DRAFT

2019 February 26

Mathematical Methods in Physics - 231B

– Group Theory –

Eric D’Hoker
Mani L. Bhaumik Institute for Theoretical Physics
Department of Physics and Astronomy
University of California, Los Angeles, CA 90095, USA
[email protected]

The purpose of this course is to present an introduction to standard and widely used
methods of group theory in Physics, including Lie groups and Lie algebras, representation
theory, tensors, spinors, structure theory of solvable and simple Lie algebras, homogeneous
and symmetric spaces.
Contents
1 Definition and examples of groups 5
1.1 A little history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Topological groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Lie Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.5 Vector spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.6 Lie Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.7 Relating Lie groups and Lie algebras . . . . . . . . . . . . . . . . . . . . . 15
1.8 Tangent vectors, tangent space, and cotangent space . . . . . . . . . . . . 16

2 Matrix Lie groups and Lie algebras 18

2.1 The general linear group GL(n) . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2 Closed subgroups of the general linear group . . . . . . . . . . . . . . . . . 21
2.3 The orthogonal groups SO(n) and Lie algebras so(n) . . . . . . . . . . . . 21
2.4 The symplectic group Sp(2n) and Lie algebra sp(2n) . . . . . . . . . . . . 23
2.5 The unitary group SU (n) and Lie algebra su(n) . . . . . . . . . . . . . . . 25
2.6 Cartan subalgebras and subgroups and the center . . . . . . . . . . . . . . 26
2.7 Summary of matrix Lie groups and Lie algebras . . . . . . . . . . . . . . . 26
2.8 Non-semi-simple matrix groups . . . . . . . . . . . . . . . . . . . . . . . . 30
2.9 Invariant differential forms, metric, and measure . . . . . . . . . . . . . . . 31
2.10 Spontaneous symmetry breaking, order parameters . . . . . . . . . . . . . 37
2.11 Lie supergroups and Lie superalgebras . . . . . . . . . . . . . . . . . . . . 38

3 Representations 41
3.1 Representations of groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2 Representations of Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3 Transformation of a vector space under a representation . . . . . . . . . . . 44
3.4 Direct sum of representations . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.5 Schur’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.6 Unitary representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.7 Tensor product of representations . . . . . . . . . . . . . . . . . . . . . . . 49
3.8 Characters of Representations . . . . . . . . . . . . . . . . . . . . . . . . . 50

4 Representations of SL(2; C), SU (2), and SO(2, 1) 53

4.1 Irreducible representations of the Lie group SU (2) . . . . . . . . . . . . . . 53
4.2 Finite-dimensional representations of sl(2, C) . . . . . . . . . . . . . . . . . 54
4.3 Infinite-dimensional representations of sl(2; C) . . . . . . . . . . . . . . . . 57
4.4 Harmonic oscillator representation of so(2, 1; R) . . . . . . . . . . . . . . . 57
4.5 Unitary representations of so(2, 1; R) . . . . . . . . . . . . . . . . . . . . . 58
4.6 An example of the continuous series . . . . . . . . . . . . . . . . . . . . . . 60

2
5 Tensor representations 62
5.1 Tensor product representations . . . . . . . . . . . . . . . . . . . . . . . . 62
5.2 Symmetrization and anti-symmetrization . . . . . . . . . . . . . . . . . . . 63
5.3 Representations of SU (3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.4 Representations of SU (n) . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

6 Spinor representations 68
6.1 Spinor representations of SO(3; R) . . . . . . . . . . . . . . . . . . . . . . 68
6.2 The Clifford Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
6.3 Representations of the Clifford algebra . . . . . . . . . . . . . . . . . . . . 70
6.4 Spinor representations of so(d) . . . . . . . . . . . . . . . . . . . . . . . . . 72
6.5 Reducibility and Weyl spinor representations . . . . . . . . . . . . . . . . . 73
6.6 Charge and complex conjugation . . . . . . . . . . . . . . . . . . . . . . . 74
6.7 Spinor representations of so(d − 1, 1; R) . . . . . . . . . . . . . . . . . . . . 75

7 Roots, weights, and representations 78

7.1 The Cartan-Killing form . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
7.2 Weights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.3 Roots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
7.4 Raising and Lowering operators . . . . . . . . . . . . . . . . . . . . . . . . 82
7.5 Finite-dimensional representations . . . . . . . . . . . . . . . . . . . . . . . 82
7.6 The example of SU (3) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
7.7 The root lattice of An = sl(n + 1) . . . . . . . . . . . . . . . . . . . . . . . 86
7.8 The root lattice of Dn = so(2n) . . . . . . . . . . . . . . . . . . . . . . . . 87
7.9 The root lattice of Bn = so(2n + 1) . . . . . . . . . . . . . . . . . . . . . . 89
7.10 The root lattice of Cn = sp(2n + 1) . . . . . . . . . . . . . . . . . . . . . . 90

8 Representations of the classical Lie algebras 92

8.1 Weyl reflections and the Weyl group . . . . . . . . . . . . . . . . . . . . . 92
8.2 Finite-dimensional irreducible representations of sl(n + 1; C) . . . . . . . . 93
8.3 Spinor representation of so(2n + 1; C) . . . . . . . . . . . . . . . . . . . . . 94
8.4 Spinor representations of so(2n; C) . . . . . . . . . . . . . . . . . . . . . . 94

9 The structure of semi-simple Lie algebras 96

9.1 Some properties of simple roots . . . . . . . . . . . . . . . . . . . . . . . . 96
9.2 Classification of finite-dimensional simple Lie algebras . . . . . . . . . . . . 97

10 Weyl’s character formulas 102

10.1 Characters on the maximal torus . . . . . . . . . . . . . . . . . . . . . . . 103
10.2 Weyl’s first formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

3
 Bibliography

Standard texts for physicists

• H. Georgi, Lie Algebras and Particle Physics, Benjamin/Cummings, 1982;
• B.G. Wybourne, Classical Groups for Physicists, Wiley, 1974;
• W. K. Tung, Group Theory in Physics, World Scientific;
• P. Di Francesco, P. Mathieu, D. Sénechal, Conformal Field Theory,
Springer 1997, Chapters 13 and 14.

Classics and more mathematical

• H. Weyl, Classical Groups, Princeton, 1946.
• C. Chevalley, Theory of Lie Groups, Princeton, 1948;
• E. Cartan, The Theory of Spinors, Hermann, Paris 1966, Dover;
• S. Helgason, Differential Geometry, Lie Groups, Symmetric Spaces, Acad. Press 1978;
• M. Nakahara, Geometry, Topology and Physics, Institute of Physics Publishing (2005)
(no group theory per se, but great for topology and differential geometry);

Very useful but more mathematical

• B.C. Hall, Lie Groups, Lie Algebras, and Representations,
Springer Verlag Graduate Texts in Mathematics 222 (2015);
• W. Fulton and J. Harris, Representation Theory, Springer Verlag, 1991;
• D.P. Zelobenko, Compact Lie Groups and Their Representations,
Tansl. Math. Monographs, Vol. 40, Am. Math. Soc., 1973;
• N.J. Vilenkin, Special Functions and the Theory of Group Representations,
Translations of Mathematical Monographs, Vol 22, Am. Math. Soc., 1968;
• F.D. Murnaghan, The Theory of Group Representations, Johns Hopkins, 1938;

Others
• A.W. Knapp, Lie Groups beyond an Introduction, Birkhäuser 2004;
• J. Humphreys, Introduction to Lie Algbras & Representation Theory, Springer, 1980;
• N. Jacobson, Lie Algebras, Wiley, 1962; Dover, 1979;
• I.R. Porteous, Clifford Algebras and the Classical Groups, Cambridge, 1995;

4
1 Definition and examples of groups
In this section, we shall begin with a little history, and then present the basic definitions and
some simple examples of groups, subgroups, topological groups, discrete groups, parametric
groups, Lie groups, and Lie algebras.

1.1 A little history

The mathematical concept of a group goes back to Joseph-Louis Lagrange (1736 - 1813)
and Niels Abel (1802 - 1829), and was articulated by Evariste Galois (1811 - 1832) in the
context of subgroups of the group of permutations of the roots of a polynomial. Lie groups
and Lie algebras were introduced by Sophus Lie (1842 - 1899) around 1870.
In physics the concept of a group arises almost invariably in the context of a symmetry
or invariance of a physical quantity or of an equation which described a physical quantity.
Greek antiquity appreciated the beauty of regular polygons and polyhedra. Galileo and
Newton undoubtedly understood the notions of translation and rotation invariance, such
as in Galilean relativity, and used them without invoking the notion of a group.
Poincaré (1854 - 1912) observed the invariance of Maxwell’s equations (without sources)
under Lorentz and Poincaré transformation. It is fair to state that Einstein (1879 - 1955)
pioneered elevating the symmetry of the equations to a physical principle in his 1905 theory
of special relativity (Poincaré invariance), but even more clearly in his 1915 theory of general
relativity (invariance under general coordinate transformations). Since then, group theory
has become a dominant organizing principle of modern physics: Herman Weyl (1885 - 1955)
applied group theory to quantum mechanics, Lev Landau (1908 - 1968) based his theory
of second order phase transitions on the group-theoretic symmetry properties of an order
parameter; Murray Gell-Mann (1929 - present) used SU (3) group theory to predict the Ω
particle; and the Standard Model of Particle Physics is built on the Yang-Mills theory for
the gauge group SU (3)c × SU (2)L × U (1)Y , though some of this symmetry is spontaneously
broken and not manifest at low energies.
One of the biggest theoretical questions in modern theoretical and experimental physics
is to find out what lies beyond the Standard Model. The most popular theoretical specu-
lations are grand unification, which is based on extending SU (3)c × SU (2)L × U (1)Y to a
larger unifying group, and supersymmetry which is based on extending the Poincaré group
to a supergroup, with fermionic parameters. One of the most active areas of research over
the past decade has been the conformal bootstrap program, based on exploiting the prop-
erties of the conformal group to study quantum field theories that do not necessarily admit
a Lagrangian or Hamiltonian description.

5
1.2 Groups
A set G equipped with a binary operation ? is a group G? provided G and ? satisfy,
1. Closure: For any pair of elements g1 , g2 ∈ G the operation obeys g1 ? g2 ∈ G;
2. Associativity: For any triplet of elements g1 , g2 , g3 ∈ G the operation obeys
(g1 ? g2 ) ? g3 = g1 ? (g2 ? g3 ) = g1 ? g2 ? g3 ;
3. Identity: There exists an element e ∈ G such that e ? g = g ? e = g for all g ∈ G;
4. Inverse: For every g ∈ G, there exists a g −1 ∈ G such that g ? g −1 = g −1 ? g = e.
The group axioms imply that the identity element e is unique, and that for every element
g, the inverse g −1 is unique. The direct product G1 × G2 of two groups G1 and G2 with
respective operations ∗1 and ∗2 is a group with elements (g1 , g2 ) under the operation ∗
defined by (g1 , g2 ) ∗ (h1 , h2 ) = (g1 ∗1 h1 , g2 ∗2 h2 ) with g1 , h1 ∈ G1 and g2 , h2 ∈ G2 . A group
G? is commutative or Abelian if the product satisfies g1 ? g2 = g2 ? g1 for all g1 , g2 ∈ G.
Otherwise, the group is said to be non-commutative or non-Abelian. When no confusion is
expected to arise, one often drops reference to the operation in denoting a group, so that
G∗ is simply denoted G.

1.2.1 Subgroups, invariant subgroups, and simple groups

A subgroup H of a group G is a subset H ⊂ G such that h1 ? h2 ∈ H for all h1 , h2 ∈ H and
h−1 ∈ H for all h ∈ H. The group G itself and the group consisting of the identity element
{e} only both trivially form subgroups of the group G. Every subgroup of an Abelian group
is Abelian, but a non-Abelian group may have Abelian and/or non-Abelian subgroups.
A subgroup H of a group G is referred to as an invariant subgroup, or normal subgroup,
iff it is invariant under all the inner automorphisms of G by conjugation,

γ ∗ H ∗ γ −1 ∈ H for all γ ∈ G (1.1)

An inner automorphism of a group G is any transformation T : G → G which preserves the

multiplication structure of the group, so that T (g1 )∗T (g2 ) = T (g1 ∗g2 ) for all g1 , g2 ∈ G with
T (e) = e and T (g −1 ) = T (g)−1 . The key example of an inner automorphism is conjugation,
Tγ (g) = γ ∗ g ∗ γ −1 for γ ∈ G. Inner automorphisms, viewed as maps from G to G, form a
group under the operation of composition of maps, Tγ ◦ Tγ 0 (g) = Tγ∗γ 0 (g). Every subgroup
of an Abelian group G is an invariant subgroup.
The quotient space G/H of a group G and one of its subgroups H is the set of equivalence
classes in G under the equivalence relation of right-multiplication by elements in H. Two
elements g1 , g2 ∈ Γ are equivalent to one another iff there exists an element h ∈ H such that
g2 = g1 ? h, or equivalently g1 = g2 ∗ h−1 . The class containing an element g ∈ G is denoted
[g], and all the elements of G in the class [g] are obtained from g by right multiplication
by H, and are thus given by g ∗ H.

6
 If H is a normal subgroup of G, then the quotient space G/H is itself a group, and
thus a subgroup of G. To prove this assertion we will show that the operation ∗ of G
induces a multiplication on the classes of the quotient space G/H. Let [g1 ] and [g2 ] be the
classes of G/H which contain respectively the elements g1 and g2 . The general elements of
G in those classes are respectively g1 ∗ h1 and g2 ∗ h2 where e h1 , h2 ∈ H. Now consider
the product of two arbitrary representatives of the classes [g1 ] and [g2 ], which is given by
g1 ∗ h1 ∗ g2 ∗ h2 . When H is a normal subgroup of G, we may use the conjugation relation
g2−1 ∗ h1 ∗ g2 = h ∈ H so that

g1 ∗ h1 ∗ g2 ∗ h2 = g1 ∗ g2 ∗ h ∗ h2 (1.2)

By definition the element g1 ∗ g2 ∗ h ∗ h2 belongs to the class [g1 ∗ g2 ], so that the operation
∗ on G induces an operation ? on the classes. When H is not a normal subgroup of G,
we cannot use the relation g2−1 ∗ h1 ∗ g2 = h ∈ H and then it is generally not possible to
consistently map the product of arbitrary members of two classes to a single class.
A group G is referred to as a simple group iff its only invariant subgroups are the group
G itself and the trivial subgroup of the identity {e}. Simple groups constitute the building
blocks of general groups which, under certain conditions, may be decomposed into a direct
product of simple groups and certain Abelian groups.

1.2.2 Fields
A field F is defined to be a set F equipped with two operations, which are usually referred
to as addition +, and multiplication ×, and which are such that F+ is an Abelian group
with identity element 0, F0× is an Abelian group with identity 1, and the operations are
related by the property of distributivity for all a, b, c ∈ F,

a × (b + c) = a × b + a × c (1.3)

Fields form the basis of virtually all the algebra needed in physics.

1.2.3 Examples
Some of the simplest groups under the operations of addition and multiplication are,

integers Z+
rationals Q+ Q0×
reals R+ R0×
complex C+ C0×
quaternions H+ H0× (1.4)

7
with the subgroup inclusions Z+ ⊂ Q+ ⊂ R+ ⊂ C+ ⊂ H+ and Q0× ⊂ R0× ⊂ C0× ⊂ H0×
where the subscript 0 stands for the removal of 0, the identity element of addition. The
above-mentioned groups are all Abelian, except for H0× which is non-Abelian. Combining
addition and multiplication, Q, R and C are fields, but H is not a field with this definition
since its multiplication is non-Abelian. In physics the fields of interest are R and C.
The group of integers has truncations to finite groups, defined as follows,
Zn = Z/nZ = {0, 1, · · · n − 1}+ addition mod n
Z0p = Z/pZ0
= {1, · · · p − 1}× multiplication mod p for p a prime number
where the function mod acts as follows. For N ∈ Z and n ∈ N, we define q ≡ N (mod n)
such that N = pn + q with 0 ≤ q < n. Note that for multiplication it is crucial to have p a
prime number, as otherwise the product of two elements may vanish mod p, as for example
in (2 × 3)(mod 6) = 0. When p is prime, Zp forms a field which is denoted Fp .
Another example is the group of permutations Sn of a set An with n elements. More
generally, a group G acting on a set A by permuting its elements is a transformation group
acting on A. When the number of elements of A is n < ∞ one always has G ⊂ Sn , but G
need not be the full permutation group. Examples of transformation groups include many
of the transformations we use in physics, including translations on the real line by discrete
translations such as in a crystal or by continuous translations, and similarly for rotations.
A very important class of groups is matrix groups, which are sets of square matrices
with the operation of matrix multiplication. They will be studied in great detail in this
course. Here we shall just quote the most general of the matrix groups GL(n) which is the
general linear group of n × n matrices. The entries of the matrices may be in Z, Q, R, C or
even quaternions H and are then denoted respectively by,
GL(n, Z) ⊂ GL(n, Q) ⊂ GL(n, R) ⊂ GL(n, C) ⊂ GL(n, H) (1.5)
In each case the determinants form a normal subgroup under multiplication, respectively
given by Z2 , Q, R, C, H (one often omits the zero superscript and the multiplication sub-
script). The quotients of each group GL(n) by its determinant invariant subgroup is again
a group referred to as the special linear groups, and denoted by,
SL(n, Z) ⊂ SL(n, Q) ⊂ SL(n, R) ⊂ SL(n, C) ⊂ SL(n, H) (1.6)
The other matrix groups will defined as closed subgroups of one of the above.
Finally, rectangular m × n matrices form a group under addition of matrices. This
group is equivalent to a subgroup of GL(m + n) under multiplication. Denoting two m × n
matrices by M1 and M2 , the correspondence is exhibited by the following matrix identity,
    
Im M1 Im M2 Im M1 + M2
= (1.7)
0 In 0 In 0 In
where Im and In respectively denote the identity m × m and n × n matrices.

8
1.3 Topological groups
The notions of open and closed sets of a set G, of a topological space G, a metric topology,
a Cauchy sequence, and continuity were introduced in section 2 of the notes for 231A, and
we shall not repeat the definitions here. A topological group G is a group G equipped with
a topology so that G is also a topological space and the group operation is continuous. For
later convenience we shall recall here a few properties of a topological group G which will
be of fundamental importance in group theory.

1.3.1 Connectedness
A subset S of a topological group G is path-connected (which we shall simply refer to as
connected when no confusion is expected to arise), if any two points x, y ∈ S can be joined
by a continuous function f : [0, 1] → S given by f (s) for s ∈ [0.1], such that f (0) = x
and f (1) = y. Note the criterion of continuity and the requirement that f ([0, 1]) must
be entirely contained in S. A set S which is not connected is disconnected. The set of
all disconnected components of S forms a group, denoted by π0 (S) and referred to as the
zero-th homotopy group of S.
A topological group G where every point is disconnected from every other point is called
discrete and we have π0 (g) = g for every g ∈ G.
Let G be a topological group and Gid the connected component of G which contains
the identity. Then Gid is an invariant subgroup and every connected component of G is
congruent to G0 with π0 (G) = G/Gid .
The set of connected components of the direct product of two topological groups G1 , G2
(or more generally of two topological spaces) is the direct product of the sets of components,

π0 (G1 × G2 ) = π0 (G1 ) × π0 (G2 ) (1.8)

so that the number of disconnected components is multiplicative.

1.3.2 Simply-connectedness
Let S be a connected subset of a topological group G. A closed curve C ⊂ S is given by a
continuous function C : [0, 1] → S such that C(0) = C(1). A subset S is simply-connected
if every closed curve in S can be continuously deformed to a point, while remaining in S.
To make this more explicit, a closed curve C can be deformed to a point p provided there
exists a continuous function Cˆ from [0, 1] × [0, 1] → S such that,
ˆ 1) = C(s)
C(s, ˆ 0) = p
C(s, s ∈ [0, 1] (1.9)

Subsets which are not simply-connected are also very important, and one can provide a
measure for how much they fail to be simply connected. The key ingredient is the notion of

9
homotopy. Two curves C0 and C1 in S which have the same end points p, q are homotopic
to one another provided there exists a continuous function Cˆ : [0, 1] × [0, 1] → S such that,

ˆ 1) = C1 (s)
C(s, ˆ t) = C1 (1) = C0 (1) = p
C(1,
ˆ 0) = C0 (s)
C(s, ˆ t) = C1 (0) = C0 (0) = q
C(0, (1.10)

Homotopy induces an equivalence relation between curves, either with the same end-points
or between closed curves in S. Thus we may state the condition of simply-connectedness
as the fact that all closed curves in S are homotopic to points.
The equivalence classes of closed curves in S under the homotopy equivalence relation
are referred to as the elements of the first homotopy group or fundamental group π1 (S).
The classes form a group under composition of curves by choosing two representatives C1
and C2 which have a point p in common and then composing the curves at the point p.
Under direct product of two connected topological groups G1 , G2 , the fundamental
group behaves as follows,

π1 (G1 × G2 ) = π1 (G1 ) × π1 (G2 ) (1.11)

For example, the fundamental group of U (1) = S 1 = SO(2) is isomorphic to Z (see the
corresponding discussions in the 231A notes). Since S 1 is connected, we apply the above
result for the torus and we have π1 (S 1 × S 1 ) = Z2 . Since R is simply connected and has
π1 (R) = 0, the fundamental group of the cylinder is π1 (S 1 × R) = Z.

1.3.3 Compactness
A cover of a subset S of aStopological group G is defined to be a class of open sets On
with n ∈ N such that S ⊂ On and thus covers S. A subspace S of a topological space
is compact if for every cover {On }n∈N one can extract a finite sub-cover. The Heine-Borel
theorem states that a subset of Rn , with the topology induced by the Euclidean metric on
Rn , is compact if and only if it is closed and bounded in the sense of the Euclidean metric
topology. If G1 , G2 are two compact spaces, then their direct product G1 × G2 is compact.
On the other hand, if G1 is not compact then the direct product of G1 with any group is
not compact.
It is important to point out that a given set may be compact under one topology and
non-compact under another. For example, consider the complex plane C = R2 . Under
the topology induced by the Euclidean metric, R2 is non-compact since it is unbounded.
However, we could equip C with a different metric, such as for example,

|dz 2
ds2 = (1.12)
(1 + |z|2 )2

10
With this metric, infinity z = ∞ is a finite distance away from any other point in C, and
infinity reduces to just a single point. Strictly speaking the point z = ∞ does not belong
to C, but we can add it to C to produce the set,
C̄ = C ∪ {∞} (1.13)
The set C̄ with the metric ds2 can be covered by just two open sets C̄ = O1 ∪ O2 , where
we can take for example O1 = {z ∈ C, |z| < 2} and O2 = {z ∈ C, |z| > 1}. Thus, the
set C̄ with the metric topology induced by ds2 is a compact set, and C̄ is referred to as a
(one-point) compactification of C.

1.3.4 Manifolds
A topological space M is a topological manifold provided every point p ∈ M has an open
neighborhood which is homeomorphic to an open set in Rn . A map is a homeomorphism
provided it is bijective, continuous, and its inverse map is also continuous. By continuity,
the value of n must be constant throughout the manifold, and n is referred to as the
dimension of M . Some important aspects of manifolds are as follows.
1. The homeomorphism ϕα from an open neighborhood Uα of a point p ∈ M into Rn
provides‘local coordinates for Uα . The pair (Uα , ϕα ) is called a chart (think of a
geographical map), and an atlas (Uα , ϕα )α∈S is a set S of charts such that,
[
Uα = M (1.14)
α∈S

so that all points in M may be described by at least one local coordinate system.
2. The only way the open sets Uα can cover M is by having non-trivial intersections with
one another. The intersection Uα ∩ Uβ of two open sets Uα and Uβ is an open set. If
the intersection is non-empty we have two different homeomorphisms (i.e. coordinate
sets) for the same point p, namely ϕα (p) and ϕβ (p). Their compositions,
ψα,β = ϕα ◦ ϕ−1 n
β : R → R
n
(1.15)
are referred to as transition functions. Since ϕα and ϕβ and their inverses are con-
tinuous, the functions ψα,β are automatically continuous.
While it follows from the definition of a topological manifold that the transition func-
tions ψα,β are continuous, we may impose conditions on ψα,β which are stronger than mere
continuity. Here are some of the most frequently used extra conditions.
• Differentiable manifold or more precisely C k differentiable manifold provided the order
k derivatives of the transition functions exist and are continuous;
• Real analytic manifold provided the transition functions are real analytic;
• Complex manifold exists provided n is even, and the transition functions are holo-
morphic functions of several complex variables.

11
1.3.5 Group manifolds and parametric groups
Requiring a topological group G to be a manifold is yet a further restriction that will turn
out to be of great importance. When the group G is a manifold, its dimension dim G is
defined to be the dimension of G as a manifold.
A group G is a parametric group if it can be parametrized (at least locally in the
neighborhood of any point) by a system of real coordinates or parameters t = (t1 , · · · , tn ),

g(t) = g(t1 , . . . , tn ) (1.16)

so that the group multiplication structure in G is given by,

g(ti ) ∗ g(si ) = g(fi (t, s)) (1.17)

where the real-valued functions fi (t, s) are continuous. Since the parameters locally describe
an Euclidean Rn neighborhood, a parametric group G is automatically a manifold endowed
with a group structure.

1.4 Lie Groups

A group G is a Lie group if G is a parametric group and the function fi (s, t) expressing
the composition of two group elements are real analytic in s, t.

Theorem 1 If the functions fi (s, t) admit first derivatives in s, t and those derivatives are
continuous, then the functions fi are analytic in s, t. (Sophus Lie, 1888)

Theorem 2 Every parametric group is a Lie group. (This gives a solution to Hilbert’s Vth
problem; see D. Montgomery and L. Zippin , 1955 )
These theorems show that the combination of the group structure and the continuity
of fi or first order differentiability automatically implies real analyticity.
Continuity in parameters is pervasive in physics, whenever we have a continuous sym-
metry, such as rotations or Lorentz transformations, or group transformations. Therefore,
our emphasis in these lectures will be on Lie groups, and we shall be brief on discrete
groups, such as permutations and crystallographic groups.

1.4.1 Examples of Lie groups

The groups R and C, under addition and multiplication, are Lie groups since the operations
on two elements x, y of addition x + y and multiplication xy are real analytic. Similarly, the
groups Rn , Cn , and the additive group of m × n matrices are all Lie groups under addition.
The groups Z and Q under addition are not Lie groups, as they are not even manifolds.

12
 Under multiplication, the matrix groups GL(n, R), GL(n, C), SL(n, R), SL(n, C) are all
Lie groups. For example, to see this for the case of GL(n, R) we exhibit the matrix in terms
of its entries xij for i, j = 1 · · · , n,
 
x11 x12 · · · x1n
 x21 x22 · · · x2n 
g(x) =  
· · · · · · · · · · · ·  (1.18)
xn1 xn2 · · · xnn
The entries xij may be used as real parameters to promote GL(n, R) into a parametric
group. Note that the parameters are labelled here by a composite index, but this is equiv-
alent to the labelling by a single parameter used in the definition of a parametric group.
The composition functions f (x, y) are given by the rules of matrix multiplication,
n
X
g(x)g(y) = g(f (x, y)) fij (x, y) = xik ykj (1.19)
k=1

They are continuous functions of x, y, so that GL(n, R) satisfies to axioms of a Lie group,
which are in fact real analytic as should be the case for a Lie group. Similarly, GL(n, C),
SL(n, R), and SL(n, C) are Lie groups.

1.5 Vector spaces

Next, we will introduce Lie algebras, but to do so we shall first recall the definition of a
vector space. A vector space V over a field F is a set V, equipped with two operation:
1. under the operation of addition, denoted +, V forms an Abelian group;
2. there is also an operation of scalar multiplication of elements of V by elements of F
which satisfies associativity and distributivity axioms, for all α, β ∈ F and all u, v ∈ V,
(α + β)u = αu + βu
α(u + v) = αu + αv
(αβ)u = α(βu) (1.20)
The zero element 0 of F is related to the zero element 0 of V by 0u = 0, and the unit 1 of
F acting trivially on V by 1u = u. The fields of interest in physics will be R or C giving
respectively a real or a complex vector space. The vector space V will then be isomorphic
to Rn , Cn or be an infinite-dimensional generalization, such as a complex function space.

1.6 Lie Algebras

A Lie algebra (defined over a field F which will be either R or C), is a vector space G
endowed with a binary operation [x, y] which is referred to as the Lie bracket and which
satisfies the following axioms,

13
 1. closure: [x, y] ∈ G for all x, y ∈ G;
2. anti-symmetry: [y, x] = −[x, y] for all x, y ∈ G;
3. linearity: [αx + βy, z] = α[x, z] + β[y, z] for all α, β ∈ F and all x, y, z ∈ G;
4. Jacobi identity: [x, [y, z]] + [y, [z, x]] + [z, [x, y]] = 0 for all x, y, z ∈ G.
When the vector space G has dimension n over F, we denote a basis of G by Xa for
a = 1, · · · , n and Xa are referred to as generators of the Lie algebra. Closure implies that
the Lie bracket of two basis vectors must be a linear combination of the Xa , so that we
have the structure relations of the Lie algebra G,
n
X
[Xa , Xb ] = fabc Xc (1.21)
c=1

The coefficients fabc are referred to as the structure constants of the Lie algebra G. By
anti-symmetry of the Lie bracket, they are anti-symmetric in the first two indices,

fbac = −fabc (1.22)

for all a, b, c = 1, · · · n. By linearity, the product of any two vectors may be decomposed in
terms of the structure constants, while the Jacobi identity on the basis vectors,

[Xa , [Xb , Xc ]] + [Xb , [Xc , Xa ]] + [Xc , [Xa , Xb ]] = 0 (1.23)

imposes the Jacobi relations on the structure constants,

n 
X 
fade fbcd + fbde fcad + fcde fabd = 0 (1.24)
d=1

Conversely, a set of coefficients fabc will correspond to structure constants of a Lie algebra
provided fabc is anti-symmetric in the indices a, b and satisfies the Jacobi identity (1.24).
We note that, while the Lie bracket [x, y] in the definition of a Lie algebra has been
denoted by the symbol customarily used for the commutator xy−yx, a Lie algebra structure
does not assume that products such as xy can be defined for x, y ∈ G: these products may
or may not exist. For the case of matrix algebras, we will in fact have [x, y] = xy − yx
because the product of square matrices exists independently of their commutator.

1.6.1 Examples of Lie algebras

Any vector space V, over R or C, may be endowed with a vanishing Lie bracket for all
pairs of elements of V. The corresponding Lie algebra is then referred to as an Abelian Lie
algebra. The vector spaces Rn and Cn where the elements are interpreted as translations
fall in this category.

14
 Examples of non-Abelian Lie algebras include the vector space V of n×n matrices where
the Lie bracket of two elements x, y ∈ V is the commutator [x, y] = xy − yx and operation
which is defined in terms of the product xy of n × n matrices, is always well-defined, and
automatically satisfies the Jacobi identity.
An example of a non-Abelian Lie algebra is provided by gl(n, R) which is the vector space
of all n × n matrices with real entries and the Lie bracket is the commutator. This vector
space has dimension n2 and a basis may be given by n2 matrices Xij with 1 ≤ i, j ≤ n whose
entries are all zero, except for entry (i, j) which is 1. Note that in our general discussion
of Lie algebras we used a single index a to label the generators Xa , but here is it much
more convenient to use the composite label (i, j). To work out the structure constants for
gl(n, R) using this notation, we exhibit the entries of the generator Xij as follows,

(Xij )αβ = δiα δjβ 1 ≤ α, β ≤ n (1.25)

The components of the commutator are then obtained by the rules of matrix multiplication,
n 
X 
[Xij , Xk` ]αβ = (Xij )αγ (Xk` )γβ − (Xk` )αγ (Xij )γβ (1.26)
γ=1

Expressing the components in terms of Kronecker δ, we find,

n 
X 
[Xij , Xk` ]αβ = δiα δjγ δkγ δ`β − δkα δ`γ δiγ δjβ = δjk δiα δ`β − δi` δkα δjβ (1.27)
γ=1

Recasting the Kronecker δ combinations in terms of components of generators, we arrive

at the Lie algebra structure relations of gl(n, R),

[Xij , Xk` ] = δjk Xi` − δi` Xkj (1.28)

Note that the same basis of generators Xij may be used to span the vector space of n × n
matrices with complex entries. They form the Lie algebra gl(n, C) whose structure relations
on the generators Xij coincide with those of gl(n, R).

1.7 Relating Lie groups and Lie algebras

In the subsections above, Lie groups and Lie algebras have been defined independently of
one another. In this subsection, we shall explain how the two are intimately related to one
another. We begin by stating the results, and will then partially prove them.

Theorem 3 To every Lie group G there corresponds a unique Lie algebra G generated by
the tangent vector fields to G at the identity.

15
 The definition and construction of tangent vectors and the tangent space to a point on
a manifold will be given in the subsequent subsection. For the special case of a Lie group
G, the tangent space at every point on the Lie group manifold may be obtained from the
tangent space at the identity by using left or right translations Lγ , Rγ on G, which are
defined by left and right multiplication,

Lγ (g) = γ g
Rγ (g) = g γ −1 (1.29)

Left translations form a group γ ∈ GL which is isomorphic to G, and right translations form
an independent group γ ∈ GR which is also isomorphic to G. Left and right translations
commute with one another, Lγ Rγ 0 = Rγ 0 Lγ for all γ ∈ GL and γ 0 ∈ GR and, upon setting
γ = γ 0 reduces to the inner automorphism discussed earlier. Under right translation by an
element γ = g, the element g is mapped to the identity. Every point in G may be mapped
to the identity by such a translation. The tangent space Tg at the point g is therefore
isomorphic to the tangent space Te at the identity in G.
Theorem 4 To every Lie algebra G there corresponds a unique connected and simply
connected Lie group G obtained by the exponential map applied to the Lie algebra.
The exponential map will be constructed in the next section for matrix groups. The
knowledge of the Lie algebra determines the Lie group up to global properties which cannot
be reconstructed from a local treatment, such as connectedness and simply connectedness.

1.8 Tangent vectors, tangent space, and cotangent space

Let M be a manifold of dimension n. Let Uα be a local open neighborhood of a point
p ∈ M and ϕα : M → Rn the corresponding homeomorphism to Rn . We shall assume
that the manifold is differentiable at least once, so that all transition functions are C 1 .
A differentiable curve γ is a map γ : [−1, 1] → M such that ϕα ◦ γ : [−1, 1] → Rn is
differentiable. We shall assume that the point p lies on γ and set γ(0) = p. The tangent
vector to the curve γ at the point p, in the chart Uα is defined by,
d d
ϕα ◦ γ(t) = ϕα (γ(t)) (1.30)
dt t=0 dt t=0

Since the image of ϕα is in Rn , the tangent vector defined above is an element of the vector
space Rn , and we may use the linearity properties of the vector space Rn to add vectors
and multiply them by a scalar. The corresponding vector space is defined to be the tangent
space Tp (M ), and has dimension n.
The tangent vectors defined in this manner depend on the chart (Uα , ϕα ), on the curve
γ, and on the choice of the parameter t. It will be convenient to isolate the dependence
on γ and t by recasting the tangent vectors in terms of vector fields. To do so, consider a

16
differentiable function f : M → R. In the chart (Uα , ϕα ) f may be expressed by the local
coordinates xi = ϕiα for i = 1, · · · , n in terms of a function f˜ : Rn → R by,
f (p) = f˜ ◦ ϕα (p) = f˜(x1 , · · · , xn ) (1.31)
The derivative of f along the curve γ at the point p is obtained by chain rule,
m 
∂ f˜

d X d i
f (γ(t)) = ϕα ◦ γ(t) ∂i f˜ ∂i f˜ ≡ (1.32)
dt t=0
i=1
dt t=0 ∂xi

and defines a linear map on the space of differentiable functions f . The coefficients of the
derivatives ∂i f˜ are the components of the tangent vector to γ at p, but might be replaced
with arbitrary coefficients which no longer refer to the specific curve used,
m
X
Dv f = v i (p)∂i f˜ (1.33)
i=1

It is a standard abuse of notation, when no confusion is expected to arise, to drop the tilde
on f , and to identify the points p ∈ M with their local coordinates (x1 , · · · , , xn ). The map
Dv is linear in f , obeys Leibnitz’s rule,
Dv (f g) = f Dv (g) + gDv (f ) (1.34)
and is referred to as a derivation or vector field. The tangent space Tp (M ) at a point p ∈ M
is isomorphic to the space of all vector fields. We may view the vector of derivatives ∂i
as the basis vectors of Tp (M ). Finally, it makes sense to act on a function f by different
vector fields. The commutator of two vector fields is again a vector field,
n 
X 
i
[Dv , Dw ]f = Du f u = v j ∂j wi − wj ∂j v i (1.35)
j=1

The commutator satisfies the Jacobi identity thanks to the Schwarz identity ∂i ∂j f = ∂j ∂i f .
The vector space dual to Tp (M ), in the sense of duality in normed vector spaces, is the
cotangent space Tp∗ (M ) at p, and its basis vectors are the differential forms dxi . A general
differential 1-form ω may be decomposed in this basis by,
m
X
ω= ωi (p)dxi (1.36)
i=1

The total differential d,

m
X ∂
d= dxi (1.37)
i=1
∂xi

is independent of the coordinates chosen in the chart.

17
2 Matrix Lie groups and Lie algebras
In this section, we shall discuss Lie groups and Lie algebras defined in terms of matrices
whose entries are real or complex. The group operation is matrix multiplication of n × n
matrices which gives again an n × n matrix, and is always associative.
The most general such groups were introduced earlier as GL(n, R) and GL(n, C). Fur-
ther matrix groups that will be of interest in Physics are subgroups of GL(n) defined by
certain equalities between the group elements. Such subgroups are referred to as closed
subgroups of GL(n) since generally spaces defined by an equality are closed sets. Matrix
groups whose entries are rational numbers or integers, such as GL(n, Q) and GL(n, Z), are
occur in Physics as duality groups, but they are not Lie groups.
We shall exhibit the relations between Lie groups and Lie algebras, introduce Car-
tan subgroups and subalgebras, evaluate the dimension and the rank, establish low-rank
coincidences and general inclusions of the standard matrix groups and algebras.

2.1 The general linear group GL(n)

The set of n × n matrices with real or complex entries and non-vanishing determinant form
groups under matrix multiplication denoted by GL(n, R) or GL(n, C). The identity matrix
is the unit element, and the condition of non-vanishing determinant guarantees that every
matrix in GL(n) has an inverse given by the inverse matrix. Some basic topological and
group-theoretic properties are as follows.

• GL(1, R) ≈ R0 is Abelian, non-compact, not connected since the positive and negative
components are disconnected, and simply-connected.
• GL(1, C) ≈ C0 is Abelian, non-compact, connected, but not simply connected since
the complex plane has one point removed.
• For n ≥ 2, the groups SL(n, R) and SL(n, C) are non-Abelian, non-compact since
2 2
they are unbounded in Rn or Cn and connected. While SL(n; C) is simply-connected,
its real counterpart SL(n; R) is not simply connected.
• For n ≥ 2, the groups GL(n, R) = SL(n, R) × R0 and GL(n, C) = SL(n, C) × C0
inherit the properties of their factors, and are non-Abelian, non-compact, the former
being not connected and the latter not simply connected.

As was already explained earlier, these matrix multiplication groups are all Lie groups.

2.1.1 The general linear Lie algebra gl(n)

The general linear Lie algebra gl(n, R) is the vector space of all n × n real matrices. The
Lie bracket is given by the matrix commutator [x, y] = xy − yx, which gives a real n × n

18
matrix for any x, y ∈ gl(n, R), is clearly anti-symmetric and linear in each argument, and
automatically satisfies the Jacobi identity, since we have,

[[x, y], z] + [[y, z], x] + [[z, x], y] = [xy − yx, z] + [yz − zy, x] + [zx − xz, y] = 0 (2.1)

for all x, y, z ∈ gl(n, R). Thus, gl(n, R) forms a Lie algebra over R. Similarly, the space of all
n × n matrices with complex entries gl(n, C) forms a Lie algebra over C. The special linear
Lie algebra sl(n, R) and sl(n, C) are the subspaces of respectively gl(n, R) and gl(n, C) with
vanishing trace, and close under the commutator since tr[x, y] = 0 for all x, y ∈ gl(n).

2.1.2 From GL(n) to gl(n)

From the Lie group GL(n, R) we may construct the corresponding Lie algebra gl(n, R) by
expanding the group elements in a neighborhood of the identity element I ∈ GL(n, R)
which corresponds to the parameter values xij = δij for i, j = 1, · · · , n. To organize this
expansion, we introduce the expansion of the group elements,

gX (α) = I + αX + O(α2 ) xij = δij + αXij + O(α2 ) (2.2)

where X is an arbitrary n × n real matrix. Performing this expansion in all directions

around the identity defines the tangent space Te at the identity, and X ∈ Te . For given X,
the group elements gX (α) are labelled by one parameter α and generate a one-parameter
subgroup of G. We can now go through the conditions under which GL(n, R) is a group,
expand those conditions to linear order in α and deduce their implications on the structure
of the tangent space Te .
Closure of the Lie group requires that the product gX (α)gY (β) be in GL(n, R). To first
order in α and β this requires that αX + βY should be in Te , which means that Te is
a vector space. The structure of an algebra on Te is obtained by expanding the product
gX (α)gY (β) to terms which are first order in both α and β. To obtain these terms it is
easiest to study the commutator of group elements,

gX (α)gY (β)gX (α)−1 gY (β)−1 = (I + αX)(1 + βY )(1 − αX)(1 − βY ) + O(α2 , β 2 )

= I + αβ(XY − Y X) + O(α2 , β 2 )
= g[X,Y ] (αβ) + O(α2 , β 2 ) (2.3)

The bracket is defined to be the commutator of the matrices,

[X, Y ] = XY − Y X (2.4)

which automatically satisfies the Jacobi identity. Thus closure in the Lie group requires
that the commutator [X, Y ] ∈ Te , so that Te with the commutator operation satisfies the
axioms of a Lie algebra, and may be identified with the Lie algebra gl(n, R).

19
2.1.3 The exponential map: from gl(n) to GL(n)

Lie’s theorem states that, given a Lie algebra G, there exists a unique connected and simply-
connected Lie group G whose Lie algebra is G. Thus, we can ask in the case of matrix
groups such as gl(n) and sl(n) (with real or complex field), how one passes from the Lie
algebra to the Lie group. This is done with the help of the exponential map, which for
matrix algebras is literally given by the exponential function of matrices.
The Taylor series of an entire function f (z) has infinite radius of convergence near any
point in C. Thus we may identify the function with its Taylor series at the origin,
∞
X zk
f (z) = f (k) (0) (2.5)
k=0
k!

The function f is defined on n × n matrices X by substituting X for z in its Taylor series,

∞
X Xk
f (X) = f (k) (0) (2.6)
k=0
k!

where X k is the k-th power of the matrix X. The result f (X) is an n×n matrix. Generally,
matrix sums of this type are difficult to compute, because one generally does not have a
good formula for successive powers of X. However, if X is diagonalizable by a matrix g,
namely if one has X = gΛg −1 for some diagonal matrix Λ with components Λij = λi δij ,
then the function may be easily worked out,

f (X) = gf (Λ)g −1 f (Λ)ij = f (λi )δij (2.7)

The calculation may be extended to the case where X cannot be diagonalized, but can only
be reduced to Jordan normal form.
The exponential map of a matrix X is the exponential function of X, given by eX . The
exponential of 0 ∈ G maps to the identity matrix, while the inverse of eX is given by e−X .
A very useful relation between the trace and the determinant is given by,

det (eX ) = etr(X) (2.8)

Given an element X of the Lie algebra gl(n) one defines a one-parameter group GX by,

gX (t) = etX (2.9)

For all X and all values of t, gX (t) ∈ GL(n), as the exponential always produces an
invertible matrix. If trX = 0 then det gX (t) = 1 and thus gX (t) ∈ SL(n).

20
2.2 Closed subgroups of the general linear group
Subgroups of the general linear groups, either over R or over C, often arise in mathematics
and in physics as the elements of GL(n) which leave a bilinear form on Vn = Rn or Cn , or
a Hermitian form on Cn invariant.
A general bilinear form is specified by an n × n matrix Q, which is real or complex
according to the nature of Vn ,
Q(x, y) = xt Qy x, y ∈ Vn (2.10)
If Q is degenerate, namely if the rank of Q is strictly less than n, then the form can be
restricted to a subspace Vp with p < n on which the restricted form is non-degenerate.
Therefore we may assume that Q is non-degenerate without loss of generality. The subset
of linear transformations M ∈ GL(n) which leave Q invariant satisfy,
Q(M x, M y) = Q(x, y) M t QM = Q (2.11)
for all x, y ∈ Vn . They form a group GQ under multiplication since when Mi satisfies
Mit QMi = Q for i, 1, 2 then their product also satisfies this relation, as may be seen from
the following manipulations
(M1 M2 )t Q(M1 M2 ) = M2t M1t QM1 M2 = M2t QM2 = Q (2.12)
Multiplication is associative since Mi are matrices; the identity matrix is the unit element
of the group; and the inverse of an element M is given by M −1 = Q−1 M t Q which makes
sense because we have assumed that Q is non-degenerate. The quadratic relation implies
that det M = ±1, so that the group GQ is disconnected and has (at least) two connected
components, of which the identity component satisfies det M = 1.
The bilinear form Q need not be symmetric or anti-symmetric, but may clearly be
decomposed into its symmetric and its anti-symmetric parts. This decomposition will give
rise to bilinear forms with definite symmetry but each of which is generically degenerate.
The non-degenerate forms with definite symmetry will be studied next.

2.3 The orthogonal groups SO(n) and Lie algebras so(n)

The orthogonal groups, either over the reals or over the complex numbers, are defined to
leave a symmetric bilinear form invariant. The real and complex cases behave differently
and it will be convenient to treat them separately from one another.

2.3.1 Classification over R

We define a real-valued symmetric non-degenerate bilinear form on the vector space Vn =
Rn in terms of an n × n real symmetric matrix Q,
Q(x, y) = xt Qy x, y ∈ Rn (2.13)

21
Equivalently, one may consider the associated quadratic form Q(x, x). Since Q is real
symmetric it may be diagonalized by an orthogonal matrix M , which satisfies M t M = I
such that Q = M t DM and D is a diagonal matrix with entries di δij . In terms of the new
coordinates x0 , y 0 , the bilinear form thus reads,
n
X
Q(x, y) = d i xi y i (2.14)
i=1

Since we have assumed that the quadratic form is non-degenerate, none of the eigenvalues di
vanish, and we shall denote the number of positive and negative eigenvalues respectively by
p and q with p + q = n. The pair (p, q) is referred to as the signature of the quadratic form.
The signature is unchanged under real linear transformations on the coordinates xi , and
is thus an intrinsic characteristic of the quadratic form. By also rescaling the coordinates,
one arrives at the canonical bilinear form of signature (p, q),
 
t Ip 0
Q(x, y) = x Ip,q y Ip,q = (2.15)
0 −Iq

where Ip is the p × p identity matrix. One usually considers the bilinear forms with signa-
tures (p, q) and (q, p) to be equivalent to one another. Important special cases are,

1. Euclidean signature (n, 0), corresponding to the Euclidean metric on Rn , given by

ds2 = Q(dx, dx) = dxt dx;
2. Minkowski signature (n − 1, 1), corresponding to the Minkowski metric on Rn , given
by ds2 = Q(dx, dx) = dxt In−1,1 dx;
3. Conformal signature (n − 2, 2), do be discussed later.

The linear group which leaves a real symmetric non-degenerate bilinear form of signature
(p, q) invariant is denoted by O(p, q; R) and is defined by,

M t Ip,q M = Ip,q M ∈ O(p, q; R) (2.16)

For n = 2, the groups are Abelian, while for n = p + q ≥ 3, the groups are non-Abelian.
The defining relation implies that det M = ±1, so that O(p, q; R) is not connected as a
manifold, but rather has two connected components. The component connected to the
identity may be defined as follows,

M t Ip,q M = Ip,q det M = 1 M ∈ SO(p, q) (2.17)

For Euclidean signature (n, 0), the groups O(n; R) and SO(n; R) are compact since they
2
are closed and bounded in Rn , while for pq 6= 0 the groups SO(p, q) are non-compact, since

22
 2
they are unbounded in Rn . None of these groups are simply-connected, a property which
is both delicate and important and will be discussed in detail when we deal with spinors.
The associated Lie algebra is so(n, R), the algebraic pairing is the commutator of ma-
trices, and the elements m ∈ so(n, R) satisfy,
 
t S U
m = −Ip,q mIp,q m= (2.18)
Ut T

where S and T are anti-symmetric p × p and q × q matrices respectively, and U is an

arbitrary p × q matrix.

2.3.2 Classification over C

We define a complex-valued symmetric non-degenerate bilinear form Q on the vector space
Vn = Cn in terms of an n × n complex symmetric matrix Q,

Q(x, y) = xt Qy x, y ∈ Cn (2.19)

Unlike in the real case, there is no need in the complex case to introduce signatures, since
a factor of −1 may always be absorbed in the quadratic form by multiplication by a factor
of i. Making a general linear transformation on the coordinates, one arrives at the canonical
symmetric bilinear form over the complex numbers,

Q(x, y) = xt y (2.20)

Its invariance group is denoted by O(n, C) and consists of the elements g ∈ O(n, C) such
that g t g = I. It follows that det g = ±1 so that the group has two connected components;
the component connected to the identity is denoted SO(n, C). The corresponding Lie alge-
bra so(n, C) is generated by the same elements as the real algebra so(n, R) but considered
now with complex coefficients.

2.4 The symplectic group Sp(2n) and Lie algebra sp(2n)

The symplectic groups, either over the reals or over the complex numbers, are defined to
leave a anti-symmetric bilinear form invariant. The real and complex cases behave very
similarly and will be treated together. A non-degenerate anti-symmetric bilinear form
can only exist in even dimensions, and may be put in canonical form by general linear
transformations on the entries,
 
t 0 In
Q(x, y) = x J y J= (2.21)
−In 0

23
where x, y ∈ R2n for the real case and x, y ∈ C2n in the complex case. The closed subgroup
Sp(2n) of GL(2n) which leaves Q invariant is defined by,
Q(M x, M y) = Q(x, y) M t JM = J M ∈ Sp(2n) (2.22)
for all x, y ∈ R2n for Sp(2n; R) or x, y ∈ C2n for Sp(2n; C). The associated Lie algebras are
sp(2n; R) and sp(2n; C) and we have for either the real or the complex case that m ∈ sp(2n)
satisfies mt J + Jm = 0.

2.4.1 Symplectic structure of Hamiltonian mechanics

Hamiltonian mechanics is based on the following three ingredients:
1. Phase space S2n which is a manifold of even dimension 2n, locally parametrized by
generalized position qi and momentum pi variables for i = 1, · · · , n;
2. The Hamiltonian H(p, q) is a function on phase space;
3. The Poisson bracket {x, y} is an anti-symmetric bilinear form on phase space. When
this form is non-degenerate, S2n is a symplectic manifold. Evaluating the Poisson
bracket on the generalized coordinates,
x = (p1 , · · · , pn ; q1 , · · · , qn ) {xα , xβ } = Jαβ (2.23)
for α, β = 1, · · · , 2n. Equivalently, one defines the symplectic 2-form by,
2n n
1 X X
ω= Jαβ dxα ∧ dxβ = dpi ∧ dqi (2.24)
2 α,β=1 i=1

and this form is closed dω = 0. The linear transformations on xα which leave the
Poisson bracket or, equivalently, the symplectic form ω invariant form the real sym-
plectic group Sp(2n; R). In this way, the symplectic group is the subgroup of all
canonical transformations which act linearly on the canonical variables.
The existence of a continuous symmetry transformation which acts on the canonical vari-
ables by infinitesimal transformations δa qi and δa pi implies, by Noether’s theorem, the
existence of a conserved charge Qa . The transformations are reproduced by the action of
the Poisson bracket,
δa qi = {Qa , qi } δa pi = {Qa , pi } (2.25)
The Poisson bracket of two conserved charges Qa , Qb produces a new conserved charge, so
that the vector space of conserved charges produces a Lie algebra under composition by
the Poisson bracket,
d
X
{Qa , Qb } = fabc Qc (2.26)
c=1

24
and fabc are the structure constants of the symmetry algebra. The Poisson bracket auto-
matically satisfies the Jacobi identity when acting on smooth functions, and therefore the
symmetry algebra is a Lie algbera. The qualification is analogous to the curl of a gradient
vanishing, provided both act on smooth functions.

2.5 The unitary group SU (n) and Lie algebra su(n)

A Hermitian form (sometimes referred to as a sesquilinear form) over Cn is given in terms
of a Hermitian matrix Q by,

Q(x, y) = x† Qy x, y ∈ Cn (2.27)

with Q† = Q. As with bilinear forms, we assume without loss of generality that the form is
non-degenerate. Since Q is Hermitian, we may diagonalize it by a unitary transformation,
and its eigenvalues di will be real and non-zero, so that the form may be expressed as,
n
X
Q(x, y) = di x̄i yi (2.28)
i=1

One may further rescale the variables xi , yi by a complex number, which rescales di by an
arbitrary but positive number. Thus, we encounter again the possibility of having forms
of different signatures (p, q) with p + q = n where p and q count the number of positive
and negative eigenvalues respectively. Thus a Hermitian form of signature (p, q) may be
reduced to the following canonical form,

Q(x, y) = x† Ip,q y x, y ∈ Cn (2.29)

The forms of signature (p, q) and (q, p) are usually assumed to be equivalent to one another.
The group which leaves this form invariant is defined by,

Q(M x, M y) = Q(x, y) M † Ip,q M = Ip,q (2.30)

and denoted by U (p, q). There is no need to indicate that this is over the complex, since
Hermitian assumes complex. The group U (p, q) has a U (1) invariant subgroup, and for this
reason one defines the more frequently used group,

SU (p, q) = {M ∈ U (p, q), det M = 1} (2.31)

The most familiar case is when q = 0, in which case the group is denoted by SU (n).
The Lie algebra associated with the Lie group SU (n) is denoted by su(n) and is the
space of traceless anti-Hermitian n × n matrices m ∈ su(n). More generally, the Lie algebra

25
associated with SU (p, q) is denoted by su(p, q) and consists of n × n complex matrices m
which satisfy the relation,
 
† m1 m3
m Ip,q + Ip,q m = 0 m= (2.32)
m†3 m2

where m1 and m2 are anti-hermitian matrices of dimension p × p and q × q respectively,

and m3 is an arbitrary p × q matrix.

2.6 Cartan subalgebras and subgroups and the center

A crucial role is played by the Cartan subgroup of a Lie group, and the Cartan subalgebra
of a Lie algebra. We shall limit the definition here to semi-simple Lie groups (the product
of simple Lie groups), and semi-simple LIe algebras (the direct sum of simple LIe algebras),
which all the matrix groups are aside from some Abelian cases. The Cartan subalgebra
of a semi-simple Lie algebra G with Lie group G is defined to be its maximal Abelian
subalgebra, and is unique up to conjugation by an arbitrary element of the Lie group G.
The dimension of the Cartan subalgebra is defined to be the rank of G and G. The Cartan
subgroup of G may similarly be defined as the maximal Abelian subgroup of G, or maximal
torus, which is unique up to conjugation.
The center Z(G) of a group G is the subgroup of elements in G which commute with all
elements g ∈ G. The center Z(G) is a normal subgroup of G so that G/Z(G) is a group.

2.7 Summary of matrix Lie groups and Lie algebras

In the table below we give a summary of simple Lie groups, their group relation, Cartan type
of their Lie algebra, rank, and dimension. The unit determinant condition is understood
for all groups but has been exhibited only for SL groups for brevity.
The above Lie groups are all connected. The groups SU (n), SO(n; R), and U Sp(2n)
are compact; all others are non-compact. The groups SL(n; C) and SU (n) are simply
connected, while π1 (SO(n)) = Z2 for n ≥ 3 and π1 (Sp(2n)) = Z2 for all n ≥ 1, so these
groups are not simply connected. The group U Sp(2n) = Sp(2n; C) ∩ U (2n) is the maximal
compact subgroup of Sp(2n; C), and is sometimes denoted Sp(n).

2.7.1 Isomorphisms between matrix groups of low rank

For complex groups we have the following identifications,

• SO(3; C) = SL(2; C)/Z2

• SO(4; C) = SL(2; C) × SL(2; C)/Z2
• SO(5, C) = Sp(4; C)

26
 type group group relation rank dimension center
An−1 SL(n; C) det M = 1 n−1 2n2 − 2 Zn
2
SL(n; R) det M = 1 n−1 n −1 1 or Z2
SU (n) M †M = I n−1 2
n −1 Zn
SU (p, q) M † Ip,q M = Ip,q , p + q = n n−1 n2 − 1 Zn
Bn SO(2n + 1; C) M tM = I n 2n(2n + 1) 1
SO(2n + 1; R) M tM = I n n(2n + 1) 1
SO(p, q; R) M t Ip,q M = Ip,q , p + q = 2n + 1 n n(2n + 1) 1
Cn Sp(2n; C) M t JM = J n 2n(2n + 1) Z2
Sp(2n; R) M t JM = J n n(2n + 1) Z2
U Sp(2n) M t JM = J, M † M = I n n(2n + 1) Z2
Dn SO(2n; C) M tM = I n 2n(2n − 1) Z2
SO(2n; R) M tM = I n n(2n − 1) Z2
SO(p, q; R) M t Ip,q M = Ip,q , p + q = 2n n n(2n − 1) Z2

• SO(6; C) = SL(4; C)
For compact groups we have the following identifications at low ranks,
• SO(3; R) = SU (2)/Z2 = U Sp(2)/Z2
• SO(4; R) = SU (2) × SU (2)/Z2
• SO(5; R) = U Sp(4)/Z2
• SO(6; R) = SU (4)/Z2
For real non-compact groups we have the following identifications at low ranks,
• SO(1, 2; R) = SL(2; R) = Sp(2; R)
• SO(1, 3; R) = SL(2, C)/Z2
• SO(2, 2; R) = SO(2, 1; R) × SO(2, 1; R)
• SO(2, 3; R) = Sp(4; R)/Z2
• SO(2, 4; R) = SU (2, 2)/Z2
All these identifications at low rank involve the orthogonal groups and are somehow related
to the existence of the Clifford algebra and spinors. Many if not all play one role or another
in physics. Let us establish the most frequently used relations.

2.7.2 Pauli matrices

We begin by introducing the Pauli matrices σi for i = 1, 2, 3 which are 2 × 2 Hermitian
matrices satisfying the following relations,

tr(σi ) = 0 {σi , σj } = 2δij I2 (2.33)

27
It is customary to use a specific representation for these matrices, given by,
     
0 1 0 −i 1 0
σ1 = σ2 = σ1 = (2.34)
1 0 i 0 0 −1

2.7.3 Proving SO(3; C) = SL(2; C)/Z2

There is a simple one-to-one map between complex 3-vectors z = (z1 , z2 , z3 ) with zi ∈ C

and traceless complex 2 × 2 matrices,
  3
z3 z1 − iz2 X
M (z) = = zi σi (2.35)
z1 + iz2 −z3
i=1

The trace of M (x) vanishes while its determinant is the complex quadratic form −det M (z) =
z12 + z22 + z32 . Under conjugation gM (z)g −1 by an arbitrary invertible matrix g, the trace
continues to vanish so that the result must be a matrix M (z 0 ) of the same type,

M (z 0 ) = gM (z)g −1 (2.36)

Since multiplying g by a complex scalar, including the central element −I ∈ SL(2, C), does
not affect M (z 0 ) we restrict g ∈ SL(2, C)/Z2 . Since det M (z 0 ) = det M (z), the action of g
on x must be a complex orthogonal transformation, so that SO(3; C) = SL(2; C)/Z2 .

2.7.4 Proving SO(3; R) = SU (2)/Z2

To prove the identification SO(3; R) = SU (2)/Z2 , we define the group U (2) by the fact
that g ∈ U (2) leaves the following quadratic form on z ∈ C2 invariant under z → gz,
 
z
†
z z=1 z= 1 z1 , z2 ∈ C (2.37)
z2

Now consider the real vector xi for i = 1, 2, 3 given by the following Hermitian forms,

xi = z † σi z x1 = z̄1 z2 + z̄2 z1 x2 = iz̄2 z1 − iz̄1 z2 x3 = z̄1 z1 − z̄2 z2 (2.38)

For z † z = 1, the vector v i has unit length, 3i=1 v i v i = 1. The invariant subgroup U (1)
P
of U (2) merely multiplies z by a phase and does not act on v i . This leaves only SU (2),
but its center element −I ∈ SU (2) also does not act on v i . Hence the action is only by
SU (2)/Z2 . Now this action leaves the length of the three-vector v i invariant, and thus acts
by an orthogonal transformation on v i , so that SU (2)/Z2 = SO(3; R).

28
2.7.5 Proving SO(4; R) = (SU (2) × SU (2)) /Z2
There is a one-to-one map between real 4-vectors x = (x1 , x2 , x3 , x4 ) with xi ∈ R and
complex matrices of the form,
  3
x4 + ix3 ix1 + x2 X
M (x) = = x4 I2 + i xi σi (2.39)
ix1 − x2 x4 − ix3
i=1

which span the space defined by the conjugation relation M + M † = I2 tr(M ). One may
also think of these matrices as being isomorphic to quaternions. The following relation
gives the Euclidean quadratic form in R4 ,

M (x)† M (x) = I2 x21 + x22 + x23 + x24

(2.40)

Under independent left and right multiplications by elements (gL , gR ) ∈ SU (2)L × SU (2)R ,

M (x0 ) = gL M (x)gR−1 gL ∈ SU (2)L , gR ∈ SU (2)R (2.41)

we have M (x0 )† M (x0 ) = M (x)† M (x) and hence this transformation performs a rotation on
the vector x. However, the transformation (gL , gR ) = (−I2 , −I2 ) does not act on M (x),
and hence we have the identification SO(4; R) = (SU (2) × SU (2)) /Z2 .

2.7.6 Proving SO(1, 3; R) = SL(2; C)/Z2

There is a one-to-one map between real 4-vectors x = (x0 , x1 , x2 , x3 ) with xi ∈ R and
Hermitian 2 × 2 matrices,
  3
x0 + x3 x1 − ix2 X
M (x) = = x0 I2 + xi σi (2.42)
x1 + ix2 x0 − x3
i=1

The determinant gives the Minkowski quadratic form on R4 det M (x) = x20 − x21 − x22 − x23
which is invariant under SO(1, 3; R). Under the following action of g ∈ SL(2; C),

M (x0 ) = g † M (x)g (2.43)

the determinant, and hence the Euclidean quadratic form is preserved. The center element
−I2 ∈ SL(2; C) does not act, and hence we can identify SO(1, 3; R) = SL(2; C)/Z2 .

2.7.7 Maximal compact subgroups

Compact Lie groups play a central role in Mathematics and in Physics, and the compact
subgroups of a non-compact group also play a central role. The maximal compact subgroup
of a non-compact group is the largest such subgroup, in the sense of largest dimension and

29
 group max compact subgroup
SL(n; C) SU (n)
SL(n; R) SO(n; R)
SU (p, q) SU (p) × SU (q) × U (1)
SO(n; C) SO(n; R)
SO(p, q; R) SO(p; R) × SO(q; R)
Sp(2n; C) U Sp(2n)
Sp(2n; R) U (n)
SO(p, q; R) SO(p; R) × SO(q; R)

rank. These groups may generally be obtained as intersection groups with the largest
compact group in a given dimension, which is U (n). Thus the maximal compact subgroups
are given in the Table below. For example, the maximal compact subgroup of Sp(2n; R) is
obtained by computing its intersection with U (2n). Let M ∈ Sp(2n; R) ∩ U (2n), so that
the matrix is real and unitary, which means that it belongs also to SO(2n; R) and we have
M t M = I2n as well as M t JM = J, which upon combining both shows that M J = JM . In
n × n block decomposition, we have,
   
0 In A B
J= M= (2.44)
−In 0 C D
The commutation of M with J implies C = −B and D = −A, while M t M = I2n implies
At A + B t B = In and At B = B t A. Identifying J with the complex number i we introduce
the complex n × n matrix m = A + iB, and we see that m† m = In using the above relations
between A and B, so that m ∈ U (n).

2.8 Non-semi-simple matrix groups

The Lie groups considered so far are,
1. Abelian groups , such as for example Rn , Cn , or U (1) = SO(2, R);
2. Simple groups, such as for example SL(n), SU (n) for n ≥ 2, SO(n) for n ≥ 5, and
Sp(2n) for n ≥ 1, along with their various different real forms;
3. Semi-simple groups, such as direct products of the above simple groups;
4. Reductive groups, which direct products of semi-simple groups and Abelian groups,
such as for example the gauge group SU (3)c ×SU (2)L ×U (1)Y of the Standard Model.
But there are plenty of other matrix groups that play an important role in Mathematics
and in Physics. Here, we shall not attempt to give a full classification, but illustrate the
ideas with the examples of the group of Euclidean motions and the Poincaré group, neither
of which belongs in any of the above four categories.

30
2.8.1 The group of Euclidean motions and the Poincaré group
The Euclidean metric or distance function in arbitrary dimension n is given by,
n
X n
X
2 i 2 2
ds = (dx ) s = (xi − y i )2 (2.45)
i=1 i=1

In either formulation, the Euclidean metric is invariant under translations xi → xi +ai , y i →

y i + ai by an arbitrary translation vector a = (a1 , · · · , an )t which form an Abelian group
Rn . The metric is also invariant under rotations x → Rx and y → Ry for an arbitrary
rotation R ∈ SO(n, R) represented by an n × n orthogonal matrix acting on the column
vector x = (x1 , · · · , xn )t , which forms a non-Abelian group. We can assemble these two
transformations into a single package,
M (R, a)x = Rx + a (2.46)
The composition of two transformations is given as follows,
M (R1 , a1 )M (R2 , a2 )x = R1 R2 x + R1 a2 + a1
M (R1 , a1 )M (R2 , a2 ) = M (R1 R2 , R1 a2 + a1 ) (2.47)
From this formula, we see that the transformations close, are associative, have unit element
M (I, 0) and inverse M (R, a)−1 = M (R−1 , −R−1 a), so they form a Lie group which is
referred to the group of Euclidean motions, and denoted ISO(n; R). The composition
law reveals that translations and rotations act on x in an intertwined manner so that the
group is not the direct product of SO(n; R) by Rn , but rather the product is referred to as
semi-direct product and denoted by SO(n; R) n Rn .
A matrix representation of the group may be obtained by considering (n + 1) × (n + 1)
matrices M̃ acting on a vector x̃ = (x1 , · · · , xn , 1)t ∈ Rn+1 . Using a block decomposition
of M̃ into blocks of size n × n, n × 1, 1 × n and 1 × 1, we have the following representation,
 
R a
M̃ (R, a) = (2.48)
0 1

One verifies that M̃ (R, a)x̃ = (Rx + a, 1) so that M̃ (R, a) acts on x as M (R, a) does.
The Poincaré group is obtained by flipping the sign of one term in the metric, and the
Poincaré group ISO(1, n − 1; R) = SO(1, n; R) n Rn . Further generalizations are manifest.

2.9 Invariant differential forms, metric, and measure

A very important notion is that of invariant differential forms on a Lie group G, and
their relation with the Lie algebra G of G. To develop these objects for arbitrary Lie
groups requires a bit of differential geometry, so we shall content ourselves here by defining
invariant differential forms for matrix Lie groups.

31
2.9.1 Invariant differential forms on a Lie group
For a matrix Lie group of dimension d, the group elements g(t1 , . . . , td ) are represented
by square matrices n × n for some value of n, and are parametrized by local coordinates
(t1 , · · · , td ). The partial derivatives of g with respect to ti are again n × n matrices, and
the total differential is defined by,
d
X ∂g i
dg = dt (2.49)
i=1
∂ti

Generally, the matrix dg is not an element of the Lie algebra, but we can define differential
forms that do take value in the Lie algebra by introducing the following combinations,
ωL (g) = g −1 dg ωR (g) = dg g −1 (2.50)
It is manifest that ωL is invariant under left multiplication of g by an arbitrary constant
(i.e. t-independent) element h ∈ G, since d(hg) = hdg so that ω(hg) = ωL (g). Similarly,
ωR (gh) = ωR (g) is invariant under right multiplication by a constant element h ∈ G.
Now the claim is that both ωL and ωR take values in the Lie algebra G which was
defined to be the tangent space to G at the identity element of G. Ig we consider ωL and
ωR at the identity element, then the statement is trivial by the definition of the tangent
space to a manifold given earlier. Next, consider ωL (an analogous argument holds for ωR )
at a point g0 ∈ G which is not the identity. But now we can use the invariance of ωL under
left multiplication by g0 to pull the form back to the identity,
ωL (g) = ωL (g0−1 g) (2.51)
but from the right side it is clear that at the point g = g0 , the form is evaluated at
the identity element and thus again belongs to the Lie algebra. The dimension of the
space of left-invariant forms (and of right-invariant forms) is n. Choosing a basis Xa with
a = 1, · · · , d for G, we may decompose the forms as follows,
d
X d
X
ωL = ωLa Xa ωR = ωRa Xa (2.52)
a=1 a=1

We shall show later on that ωK and ωR satisfy the Maurer-Cartan equations which will
prove that every Lie group admits a flat connection, which has torsion.

2.9.2 Invariant metric on a Lie group

For a matrix group G, one may use the invariant differential forms ωL or ωR to construct
the following metric ds2 on G,
ds2 = −tr g −1 dg g −1 dg = −tr (ωL ωL ) = −tr (ωR ωR )


(2.53)

32
Expressing the metric ds2 in terms of local coordinates got G, we find,
d  
2
X
i j −1 ∂g(t) −1 ∂g(t)
ds = γij (t)dt dt γij (t) = −tr g (t) g (t) j (2.54)
i,j=1
∂ti ∂t

The metric ds2 is invariant under both left and right multiplication of g, and is thus
invariant under the group G × G. The signature of the metric will depend on the form of
G, and is dictated by the signature of the quadratic form tr(X a X b ). One shows that for a
compact Lie group G, the signature of −tr(Xa Xb ) is Euclidean (all positive signs), so that
the metric is then a positive Riemannian metric.
For example, in the case of the orthogonal group SO(n, R) the Lie algebra so(n, R)
consists of real n × n anti-symmetric matrices. We take a basis Xαβ with 1 ≤ α < β ≤ n,
whose components are given by (Xαβ )ij = δαi δβj − δαj δβi and evaluate the traces,

−tr (Xαβ Xγδ ) = 2δαγ δβδ − 2δβγ δαδ (2.55)

The trace is non-zero only if (α, β) = (γ, δ) or (α, β) = (δ, γ), but the latter is excluded
since it would require β = γ < α = δ, which is excluded by the ordering α < β. Thus, the
basis vectors Xαβ with 1 ≤ α < β ≤ d are all mutually orthogonal under the trace pairing,
and −tr(Xαβ )2 = 2. Therefore the invariant metric is positive definite and Riemannian.
The Lie algebra of the unitary group U (n) consists of all n × n anti-Hermitian matrices.
We take a basis consisting of the real anti-symmetric matrices Xαβ introduced for so(n)
together with the imaginary symmetric matrices iYαβ with 1 ≤ α ≤ β ≤ n for which
the components are given by (Yαβ )ij = δαi δβj + δβi δαj . It is immediate to show that
tr(Xαβ Yγδ ) = 0 and that,

tr (Yαβ Yγδ ) = 2δαγ δβδ + 2δβγ δαδ (2.56)

The trace is zero unless (α, β) = (γ, δ) or/and (α, β) = (δ, γ) in which cases it is positive.
Thus, the full basis consisting of the union of {Xαβ } and {Yαβ } is orthogonal and non-
degenerate, so that the the invariant metric is positive definite and Riemannian.
Now let’s consider the non-compact group SO(p, q; R) with pq 6= 0. For the compact
sub-algebra SO(p) we use the basis Xαβ with 1 ≤ α < β ≤ p while for SO(q) we use Xα0 β 0
with p + 1 ≤ α0 < β 0 ≤ p + q, while for the off-diagonal block we use the basis Yαβ 0 but this
time without the factor of i since those matrices must be real symmetric. The signature of
this metric is now 12 p(p − 1) + 21 q(q − 1) plus signs and pq minus signs, adding up to the
total dimension 12 (p + q)(p + q − 1) of SO(p, q; R).

2.9.3 Geodesics on a Lie group

Having the G × G invariant metric ds2 on a Lie group G, we can investigate its geodesics,
namely the curves of extremal distance between two given points. Differential geometry

33
tell us to evaluate the Christoffel connection for the metric and then solve the geodesic
equation. Here we shall be able to take a short cut, and show that for every element X of
the Lie algebra G of G, the one-parameter subgroup,

GX = gX (t) = etX , t ∈ R

(2.57)

is the unique geodesic through the identity whose tangent vector at the identity is propor-
tional to X. Clearly the curve GX contains the identity for t = 0 and has tangent vector
X. To prove that GX is a geodesic, we fix two points, the identity for t = 0, and the point
eX for t = 1, and show that the length of GX is extremal under all infinitesimal variations
δg(t) of the curve gX which leave the end points unchanged, given by,

g −1 (t)δg(t) = ε(t) ∈ G ε(0) = ε(1) = 0 (2.58)

The length of an arbitrary curve g(t) is given by,

Z 1 
−1 −1

 21
`[g] = dt − tr g ġ(t)g ġ(t) (2.59)
0

where ġ is the derivative of g(t) with respect to t. For example the length of the curve GX
1
between the identity and eX is given by `[gX ] = (−tr(X 2 )) 2 . To compute the variation in
length due to the deformation ε, we first compute the following variation,

δ g −1 ġ = −g −1 δgg −1 ġ + g −1 δ ġ


(2.60)

Using the fact that we parametrize the variations by the Lie algebra valued function ε(t),
we have δ ġ = ġε + g ε̇, which allows us to simplify the above expression, and obtain,

δ g −1 ġ = ε̇ + [g −1 ġ, ε] = ε̇ + [X, ε]


(2.61)

The length of the ε-deformed curve is given by,

Z 1 

2  21
`[g] = dt − tr X + ε̇ + [X, ε] + O(ε3 ) (2.62)
0

Expanding the trace, we have,

2
tr X + ε̇ + [X, ε] = trX 2 + 2trX ε̇ + tr(ε̇ + [X, ε])2 (2.63)

where we have used the relation trX[X, ε] = tr(X 2 ε) − tr(XεX) = 0 due to the cyclic
symmetry of the trace. Expanding the length in powers of ε to second order we find,
Z 1 
1 (trX ε̇)2

1 2
`[g] = `[gX ] + dt −trX ε̇ − tr(ε̇ + [X, ε]) + 2
+ O(ε3 ) (2.64)
0 2 2 trX

34
The first order term vanishes since X is independent of t and ε(0) = ε(1) = 0. This
immediately shows that the curve GX has extremal length, and is a geodesic.
We may learn more about these geodesics by investigating the contributions to second
order in ε. The last term in (2.64) receives contributions only from deformations ε which
are parallel to X, but these amount to a reparametrization in t of the curve and are geo-
metrically immaterial. Thus, we may assume, without loss of generality, that tr(Xε(t)) = 0
for all t, and we are left with,
Z 1
1
`[g] = `[gX ] − dt tr(ε̇ + [X, ε])2 + O(ε3 ) (2.65)
2 0

For a compact Lie group G, the trace of gives a negative definite quadratic form on G.
Since the combination ε̇ + [X, ε] ∈ G, we conclude that for compact G, the geodesics are
actually (local) minima of the length functional. For non-compact G, we can draw no such
conclusion. We shall return to this issue when we deal with homogeneous spaces.

2.9.4 Invariant measure on a Lie group

From the metric, as always, we can construct a volume form since from distances we can
compute angles, and from distances and angles we can construct the volume. Since the
metric is invariant under G × G, the resulting measure or volume form,
d
Y
p
dµG = det (γ(t)) dti (2.66)
i=1

is also invariant under G × G and is referred to as the Hurwitz-Haar measure or often just
the Haar measure. The Haar measure allows one to integrate functions over the Lie group
with strong symmetry properties, which will be of fundamental importance when we deal
with representation theory and characters. For compact Lie groups, the volume of G with
respect to the Haar measure is finite,
Z
Vol(G) = dµG (2.67)
G

These volumes may actually computed and carry interesting topological information.

2.9.5 Invariant metric on a coset space G/H

Many interesting manifolds are obtained by taking the coset space G/H of a matrix Lie
group G by one of its Lie subgroups H. For example, the sphere S n , the hyperbolic upper
half space H n , and complex projective space CP n of arbitrary dimension n ≥ 2 may be

35
constructed as coset,

S n = SO(n + 1; R)/SO(n; R)
H n = SO(n, 1; R)/SO(n; R)
CP n = SU (n + 1)/S U (n) × U (1)


(2.68)

The examples of the sphere S 2 = CP 1 and the upper half plane H 2 are familiar already.
We shall now construct an invariant metric for such spaces. Recall that at the level
of the group, the coset space is the quotient of G by the equivalence relation whereby to
elements g, g̃ ∈ G are equivalent to one another, g1 ≈ g2 , iff there exists an element h ∈ H
such that g̃ = g h. Now consider the left-invariant differential forms g −1 dg, which we have
constructed on G, and observe its transformation under right multiplication by an element
h of G, so that g̃ = g h,
 
g̃ −1 dg̃ = h−1 g −1 dg h + g dh = h−1 (g −1 dg)h + h−1 dh (2.69)

Clearly, the left-invariant differential form g −1 dg depends on the element chosen in G

to represent the coset, and is thus not well-defined on G/H. However, we see that the
inhomogeneous term h−1 dh is always an element of the Lie subalgebra H. To construct an
invariant metric on the coset G/H, we will assume that the Lie algebra G of G admits a
decomposition into a direct sum of the Lie algebra H of H plus a complement M,

G =H⊕M (2.70)

such that the mixed trace vanishes, tr(XY ) = 0 for all X ∈ H and Y ∈ M. If the Lie
group is compact, then this decomposition is automatically orthogonal under the trace
inner product. But the decomposition extends to non-compact cases as well. In particular,
this is the case when G/H is a symmetric space iff H and M satisfy,

[H, H] ⊂ H [H, M] ⊂ M [M, M] ⊂ H (2.71)

Assuming the existence of the decomposition, we now take the projection of the differential
form g −1 dg onto the space M,

ω(g) = g −1 dg (2.72)
M

the relation between the one-form ω(g) at two points g, g̃ = gh ∈ G which belong to the
same equivalence class, and thus correspond to the same point in G/H, is given by,

ω(gh) = h−1 ω(g)h (2.73)

36
Therefore, we may now construct a metric ds2 on G/H out of these new differential forms
by taking the trace,

ds2 (g) = tr ω(g)ω(g)

(2.74)

The trace is invariant under the conjugation by h and thus we have ds2 (gh) = ds2 (g),
and this metric is well-defined on the coset G/H. Furthermore, since it is constructed out
of the left-invariant forms g −1 dg, the metric ds2 (g) is automatically invariant under left
translations in G.
We leave it up to the reader to show that the round metric on S n indeed corresponds
to this coset metric, and similarly for the other spaces. It is straightforward to check that
geodesics on G/H are generated by one-parameter subgroups on G with a tangent vector
at the identity given by Y ∈ M.

2.10 Spontaneous symmetry breaking, order parameters

Consider a statistical mechanics system with a very large number (strictly speaking an
infinite number) of degrees of freedom or a quantum field theory. Suppose its microscopic
dynamics is governed by a Hamiltonian or by a Lagrangian which is invariant under a
symmetry represented by the action of a group G on the dynamical degrees of freedom. A
familiar example is given by translations or rotations in space, which are certainly symme-
tries of the microscopic Hamiltonian or Lagrangian of Physics. In quantum mechanics, the
ground state of the system is the state of lowest energy, and we shall often adopt the same
terminology for a classical statistical mechanical system at very low temperatures (think
of it as the ~ → 0 limit of a quantum system).
Now the ground state may or may not be invariant under G. Taking again the example
of rotation invariance, the ground state of a gas is statistically rotation invariant, and
the same is true for a liquid. But the magnetic field of a magnet is not invariant under
all rotation, but only under the rotations that leave the magnetic field invariant. This
is an example of spontaneous symmetry breaking, a phenomenon which occurs whenever
the ground state fails to be invariant under all the symmetries of the microscopic system,
which are the symmetries of the Hamiltonian or Lagrangian. Thermal fluctuations may
alter the symmetry and in the case of a magnet produce a second order phase transition to
an unmagnetized (approximately) rotation symmetric state above the Curie temperature.
Quantum fluctuations also contribute and may induce a quantum phase transition at very
low temperatures.
Lev Landau (1908 - 1968) proposed a theory of second order phase transitions based
on an order parameter which reflects the change in symmetry of the system. The order
parameter is a space-time dependent field (the scalar analogue of electric or magnetic fields),
governed by an effective Lagrangian which summarizes the dynamics near the ground state.

37
For every continuous symmetry that is spontaneously broken, Goldstone’s theorem assures
us that the spectrum will contain a massless scalar particle, the Nambu-Goldstone boson.
Near the ground state energy, the only excitations in the spectrum will be the Goldstone
bosons, for which we can write down an effective Lagrangian which must be invariant under
Poincaré symmetry, the symmetry G which and whose ground states must be invariant
under a subgroup H of G. The Goldstone field g(x) takes values in G for every point x of
space-time which we will just take to be flat Rn . The Lagrangian must be invariant under
G, but take values only in G/H, since a rotation of the ground state by H is immaterial. It
is immediate to deduce such a candidate Lagrangian in terms of the invariant differential
form constructed on G/H,
1  
L = tr ωM ωM ωM = g −1 ∂µ g (2.75)
2 M

These Lagrangians are special cases of what are referred to as non-linear sigma models,
and they play a crucial role in many areas of theoretical physics, including particle physics
and condensed matter and play a role in differential geometry as well.

2.11 Lie supergroups and Lie superalgebras

Fermions obey Fermi-Dirac statistics which requires Fermi fields to obey anti-commutation
relations. In the functional integral, Fermi fields are represented by anti-commuting vari-
ables referred to as Grassmann variables. (Perturbative Yang-Mills theory in a covariant
gauge requires the ghost fields which are also represented by Grassmann variables.) By the
Coleman-Mandula theorem (1967), there are no Lie group transformations that exchange
bosons and fermions consistently with the physical principle of quantum mechanics and rel-
ativity. It was discovered by Golfand and Lichtmann (1969), Ramond (1970), Gervais and
Sakita (1971), that algebras using anti-commuting variables exist that can exchange bosons
and fermions consistently. A full implementation in field theory was pioneered by Wess and
Zumino in 1974, and goes under the name of supersymmetry. The mathematical structures
realizing supersymmetry are a generalization of Lie groups in which the parameters live in
a Grassmann algebra, and a generalization of Lie algebras with both commutators as well
as anti-commutators.

2.11.1 Grassmann algebra

We postulate an arbitrary number of anti-commuting variables ξ i , i = 1, · · · N with the
following properties,
{ξ i , ξ j } = ξ i ξ j + ξ j ξ i = 0 for all i, j = 1, · · · , N (2.76)
In particular this means that the square of each variable vanishes. These variables generate
a vector space either over the real or over the complex numbers in which linear combinations

38
of ξ i may be formed. We may supplement this vector space by a copy of RM or CM with
commuting variables xI with I = 1, · · · M , and the combined vector space is denoted either
RM |N or CM |N . One introduces a binary grading by which xI has even grading and each ξ i
has odd grading, and one only allows linear combinations that preserve the grading. Thus,
it is illegal to form xI + ξ i . The grading is preserved under multiplication. This restriction
is required by the physical super-selection rule that one cannot form a linear combination
of a fermion and a boson state.
The vector spaces RM |N and CM |N generate a Grassmann algebra which is obtained by
taking linear combinations of products of an even number of ξ i and linear combinations of
products of odd numbers of ξ i subject to the grading. Thus, we can take linear combinations
such as 2 + ξ 1 ξ 3 + 6ξ 1 ξ 2 ξ 5 ξ 7 with even grading, and πξ 1 + 19ξ 7 ξ 8 ξ 9 with odd grading.
The Grassmann algebra may appear a bit abstract at first and it may not be immediately
clear that such objects actually exist. A convenient construction in terms of matrices is
obtained by starting from γ µ which satisfy the Clifford-Dirac algebra,
{γ i γ j } = 2δ ij I i, j = 1, · · · , d (2.77)
For d = 3, one realizes the γ i by the Pauli matrices γ i = σ i , and we shall see later how to
construct the matrices γ i explicitly for arbitrary d. Now take d = 2N even and form the
following combinations,
ξ j = γ 2j−1 + iγ 2j j = 1, · · · , N (2.78)
Using (2.77), one readily verifies that {ξ i , ξ j } = 0 for all i, j = 1, · · · , N .

2.11.2 Super groups

We shall denote the elements of the vector spaces Rm|n and Cm|n by graded columns,
 
x
X= (2.79)
ξ
where xI are even with I = 1, · · · , m and ξ i are odd with i = 1, · · · , n. A general linear
transformation M takes X to a vector Y = M X in the same vector space, and thus with
the same grading. Using block matrix notation,
    
y A B y = Ax + Bξ
Y = M= (2.80)
υ C D υ = Cx + Dξ
where A, B, C, D are matrices of dimension m × m, m × n, n × m and n × n respectively.
To preserve the grading, A, D must be even, and B, C must be odd. The product of two
matrices M1 , M2 is defined as the block-wise product,
   
Ak Bk A1 A2 + B1 C2 A1 B2 + B1 D2
Mk = M1 M2 = (2.81)
Ck Dk C1 A2 + D1 C2 C1 B2 + D1 D2

39
The product closes and is associative. The identity matrix has A = D = I and B = C = 0.
The inverse of M exists when the matrices A and D are invertible, and is then given by,
 0
A B0
 0
A = (A − BD−1 C)−1
 0
B = −A−1 BD0

−1
M = (2.82)
C 0 D0 D0 = (D − CA−1 B)−1 C 0 = −D−1 CA0

The space of invertible matrices M forms a group GL(m|n; R) or GL(m|n; C) either over
R or over C. However, GL(m|n) is not a Lie group in the sense defined earlier, since some
of its parameters are not real or complex numbers, but rather Grassmann numbers. Note
that the even matrices are not just real or complex numbers, but rather even elements of
the Grassmann algebra. For this reason it is referred to as a Lie super algebra.

2.11.3 The Lie super groups SU (m|n) and OSp(m|2n)

Closed subalgebras of GL(m|n) are defined by requiring invariance of a non-degenerate
even quadratic form on Cm|n or Rm|n . The quadratic form is the sum of quadratic forms
in the even variables xi and the odd variables ξ i , the latter being very restrictive. In the
complex case, it may be diagonalized to the following canonical form,

QC = x† x + ξ † ξ (2.83)

The corresponding Lie supergroup is U (m|n) with maximal even subgroup SU (m) ×
SU (n) × U (1). In the real case, non-degeneracy requires their number to be even 2n
and the canonical form is given by,

QR = xt x + ξ t Jξ (2.84)

and the Lie supergroup is OSp(m|2n) with maximal even subgroup SO(m) × Sp(2n).

2.11.4 Lie super algebras

A Lie super algebra G is a graded algebra with even subspace G0 and odd subspace G1 , on
which one defines a graded Lie bracket, [·, ·] or {·, ·} depending on the arguments,

1. closure: [G0 , G0 ] ∈ G0 , [G0 , G1 ], [G1 , G0 ] ∈ G1 , and {G1 , G1 } ∈ G0 ;

2. (anti-)symmetry: [x, y] = −[y, x] and {x, y} = {y, x};
3. linearity: [αx + βy, z] = α[x, z] + β[y, z], {αx + βy, z} = α{x, z} + β{y, z};
4. Jacobi identity: if at most one of x, y, z ∈ G1 we have the usual Jacobi identity,
with x ∈ G0 and y, z ∈ G1 we have [x, {y, z}] + {y, [z, x]} − {z, [x, y]} = 0, and with
x, y, z ∈ G1 we have [x, {y, z}] + [y, {z, x}] + [z, {x, y}] = 0.

40
3 Representations
A group is an abstract mathematical concept. The transformation of physical quantities is
realized through a representation of the group on physical variables. When the transforma-
tions are linear, we are dealing with a linear representation (or representation for short),
in terms of square matrices or in terms of linear operators. The concept of a representa-
tion is of fundamental importance in group theory, and provides a tool to classify groups.
Representations of Lie groups will be closely related with representations of Lie algebras,
which we shall study in detail. Non-linear representations or non-linear realizations of a
group also enter physics and will be discussed in a later section.

3.1 Representations of groups

A (linear) representation R of a group G with operation ∗ is a map R : G → GL(N ) in
terms of which the group operation ∗ of G maps to matrix multiplication in GL(N ),

1. R(g ∗ g 0 ) = R(g)R(g 0 ) for all g, g 0 ∈ G;

2. R(e) = IN ;
3. R(g −1 ) = R(g)−1 for every g ∈ G.

If the map is R : G → GL(N, R) one refers to R as a real representation, while if the map
is R : G → GL(N, C) one refers to R as a complex representation. A special case which
is important in physics and in mathematics is when the map R is to the unitary group
R : G → U (N ) ⊂ GL(N ; C), in which case R is referred to as unitary representation.
The dimension of the representation R is defined to be N in either case, though it must
be stressed that this definition refers to the dimension over R when R is real, and over C
when R is complex. For N = ∞, the representation is said to be infinite-dimensional.
Two representations R, R0 of the same group G are referred to as being equivalent
representations if their dimensions are equal N = N 0 and there exists a matrix A ∈ GL(N )
(which is thus invertible) such that

R0 (g) = A R(g)A−1 (3.1)

for all g ∈ G. Clearly, the matrix A may be viewed as effecting a change of basis in the
vector space on which G acts so that we shall not be interested in distinguishing between
equivalent representations, and consider only the equivalence classes of representations.

3.1.1 Examples of representations

We shall spend a lot of time constructing and examining representations of various groups
later on. Here we wish to present just a few examples to clarify the definition.

41
 1. The trivial representation assigns to every group element g ∈ G the value R(g) = 1.
2. The multiplicative group C0× has a two-dimensional representation R in terms of real
matrices, mapping a complex number z = a + ib with a, b ∈ R to the matrix,
   
a −b 0 −1
R(z) = R(i) = (3.2)
b a 1 0

3. The additive group G+ of M × N matrices admits a representation R in GL(M + N )

by (M + N ) × (M + N ) matrices of the following form for x, y ∈ G+ ,
 
1 x
R(x) = R(x)R(y) = R(x + y) (3.3)
0 1

4. The group of permutations SN has a representation by N × N matrices. To a

permutation σ ∈ SN , we assign a matrix R(σ) whose entries as given by,

R(σ)ij = δi,σ(j) i, j = 1, · · · , N (3.4)

This matrix has exactly one entry on each row which equals 1, all others being zero.
5. Another representation of the permutation group is given by the signature of the
permutation, denoted by R(σ) = (−)σ , and takes only the values ±1. In quantum
mechanics, a system of N identical bosons is invariant under permutations, while a
system of N identical fermions transforms with the signature representation.
6. Matrix groups, introduced in the preceding section, are defined in terms of subgroups
of the general linear groups GL(N ; R) or GL(N, C) and therefore naturally exhibited
in terms of a linear representation, which is referred to as the defining representation.
For GL(N ), SL(N ), SO(N ), and SU (N ) the defining representation has dimension
N while the dimension of the defining representation of Sp(2N ) is 2N .

A non-linear representation of a group G arises when G acts not on a vector space, but
rather on a manifold such as a coset space G/H of G by one of its subgroups H. A simple
example of a non-linear representation of SL(2, R) acting on the real line x ∈ R is given
by the Möbius transformation,
 
ax + b a b
g:x→ g= ∈ SL(2, R) (3.5)
cx + d c d

Non-linear representations are ubiquitous in problems where the group action is on variables
and fields which satisfy non-linear differential equations, such as Einstein’s equations of
general relativity or Yang-Mills theory.

42
3.2 Representations of Lie algebras
Consider a Lie algebra G of dimension n with generators Xa , structure constants fab c , for
a, b, c = 1, · · · , n, and structure relations,
n
X
[Xa , Xb ] = fab c Xc (3.6)
c=1

A representation ρ of a Lie algebra G is a linear map ρ : G → gl(N ) where gl(N ) is the

space of N × N real or complex matrices (see the previous section), such that ρ preserves
the Lie algebra bilinear product,

ρ([x, y]) = [ρ(x), ρ(y)] for all x, y ∈ G (3.7)

Note that, by an abuse of notation, [x, y] on the left side stands for the Lie bracket of the Lie
algebra G, while [ρ(x), ρ(y)] stands for the commutator of matrices in gl(N ). The dimension
of the representation ρ is defined to be N (over R or C as in the case of representations of
groups, see the preceding subsection). Taking Xa to be a basis of the Lie algebra G with
a = 1, · · · , n = dim G, we see that the linearity of ρ implies that ρ(Xa ) satisfy the same
structure relations as Xa itself,
n
X
[ρ(Xa ), ρ(Xb )] = fab c ρ(Xc ) (3.8)
c=1

As introduced above, the dimension n of the Lie algebra G is independent of the dimension
N of the representation ρ.

3.2.1 Examples of representations of Lie algebras

Here we just give some immediate examples of Lie algebras to illustrate their definition and
reserve for later a more systematic discussion.
1. The trivial representation assigns to every x ∈ G the zero element, ρ(x) = 0.
2. The defining representation for each matrix Lie group SL(N ), SU (N ), SO(N ),
Sp(2N ) induces the defining representation with the same dimension of the corre-
sponding Lie algebras sl(N ), su(N ), so(N ) and sp(2N ).
3. Every Lie algebra G possesses the adjoint representation whose representation matri-
ces are given by the structure constants fabc with a, b, c = 1, · · · , n = dim G,

(Fa )b c = −fab c (3.9)

The Jacobi identity of the Lie algebra G given in (1.24) may be recast in terms of
matrix multiplication for the matrices Fa , effected by the summation over the index

43
 d in (1.24). To see this, we recast the Jacobi identity in the following form by using
anti-symmetry of the structure constants in their first two indices,
n 
X  n
X
fac d fbd e − fbc d fad e = − fab d fdc e (3.10)
d=1 d=1

Expressing the four f on the left side and the second f on the right side in terms of
F , but leaving the first f on the right side, we find,
n
X
e
[Fa , Fb ]c = fab d (Fd )c e (3.11)
d=1

Thus the matrices Fa form a representation of the Lie algebra G. This representation
has exactly the same dimension as the Lie algebra, namely n, and is referred to as the
adjoint representation of G. For an Abelian Lie algebra, the adjoint representation is
always equivalent to the trivial representation.
In a basis-independent formulation, the adjoint representation of an element x ∈ G
is denoted Adx and acts on an arbitrary element z of G (which makes sense since its
dimension is exactly the same as that of the Lie algebra) by,

Adx (z) = [x, z] (3.12)

The fact that this map is a representation of G again follows from the Jacobi identity,

[Adx , Ady ] (z) = [x, [y, z]] − [y, [x, z]] = [x, [y, z]] + [y, [z, x]]
= [[x, y], z] = Ad[x,y] (z) (3.13)

which was used in going from the first to the last line.

3.3 Transformation of a vector space under a representation

The general linear group GL(N ; C) naturally transforms the vector space V = CN into V by
linear transformations, and analogously for GL(N ; R) on V = RN . Therefore, an arbitrary
group G may also transform the vector space V through an N -dimensional representation
R : G → GL(N ) of G. By the same token, the vector space V transforms under an
N -dimensional representation ρ of a Lie algebra G, and we have,

R(g) : V → V v ∈ V → R(g)v ∈ V g∈G

ρ(x) : V → V v ∈ V → ρ(x)v ∈ V x∈G (3.14)

where we have represented the elements v of the vector space V as column vectors in a
given basis of V , and R(g)v and ρ(x)v are the products of N × N square matrices by a

44
column vector on the right which gives again a column vector. The vector space V and its
elements are then said to transform under the representation R of the group G and under
the representation ρ(x) of the Lie algebra G.1 In physics, the vector spaces V describe
N physical quantities, such as position in space and time transformed by translations,
rotations, and Lorentz transformation, or a set of electromagnetic fields, or states in the
Hilbert space of a quantum mechanical system with N degrees of freedom.

3.4 Direct sum of representations

In this subsection, we shall define the direct sum of representations, and the reducibility of
a general representation into a direct sum of representations. We shall write things out for
representations of a group G, the transcription to representations of a Lie algebra being
immediate. First, we need to review the direct sum of vector spaces.
The direct sum of two vector spaces V1 and V2 of respective dimensions N1 , N2 (either
over R or C) is a vector space V1 ⊕ V2 of dimension N1 + N2 consisting of ordered pairs
(v1 , v2 ) with v1 ∈ V1 and v2 ∈ V2 such that addition on V1 ⊕ V2 is defined by,

(v1 , v2 ) + (v10 , v20 ) = (v1 + v10 , v2 + v20 ) (3.15)

In terms of bases e11 , · · · , eN 1 N2

1 for V1 and e2 , · · · , e2 for V2 a basis of V1 ⊕ V2 may be chosen
1

to be e11 , · · · , eN 1 N2
1 , e2 , · · · , e2 thereby clearly giving a vector space of dimension N1 + N2 .
1

In this basis, we may represent the vectors in V1 and V2 by column matrices,

 1
 1  1 v1
v1 v2 · · · 
N1 2  N2 2 
 N1 
X  v X  v v1 
v1 = v1i ei1 =  1 
· · ·  v 2 = v i i
e
2 2 =  2 
· · ·  v1 ⊕ v 2 =  1  (3.16)
 v2 
i=1 N1 i=1 N2
 
v1 v2 · · · 
v2N2
This matrix notation, in a given basis, will be used throughout.
The direct sum of two representations R1 and R2 of the group G, of respective dimen-
sions N1 , N2 , is a representation R1 ⊕R2 of dimension N1 +N2 obtained by taking the direct
sum of the separate transformations of G on each vector space, and may be represented in
the above basis in terms of the following block-matrix decomposition,
 
R1 (g) 0
(R1 ⊕ R2 )(g) = (3.17)
0 R2 (g)

1
In the mathematics literature, V is referred to as the representation space of R (or module) while in
the physics literature, by a confusing abuse of notation, often refers to V as the representation as well.
Here, we shall try to stick to the terminology that the vector space and its elements transform under the
representation R.

45
The operation of direct sum on the space of representations is commutative.
Given a representation R : G → GL(N ) acting on a vector space V of dimension N ,
a vector v ∈ V is said to be invariant under a subgroup H ⊂ G if R(g)v = v for all
g ∈ H. Similarly, a vector space W ⊂ V is said to be invariant under a subgroup H ⊂ G
if R(g)w ∈ W for all w ∈ W and all g ∈ H. Clearly, V and {0} are trivial invariant
subspaces. The direct sum R1 ⊕ R2 of two representations of G, as constructed above, has
two non-trivial invariant subspaces, namely V1 and V2 .
A representation R : G → GL(N ) is said to be reducible if it admits at least one
non-trivial invariant subspace under G. A representation R : G → GL(N ) is said to be
completely reducible if every invariant subspace V1 has an invariant complement V2 , so
that V = V1 ⊕ V2 . The difference between a reducible representation R0 and a completely
reducible representation R, both with invariant subspace V2 , may be illustrated by using
the matrix notation, in which case they respectively take the following form,
   
R1 (g) 0 0 R1 (g) S3 (g)
R(g) = R (g) = (3.18)
0 R2 (g) 0 S2 (g)

The vector space V1 in the upper block is invariant under both R(g) and R0 (g) for all g ∈ G.
Its complement V2 is also invariant under R(g) but not under R0 (g), since the action of
R0 (g) on V2 will have a component along V1 generated by the block matrix S3 (g).
A representation R : G → GL(N ) acting on the vector space V of dimension N is
irreducible if its only invariant subspaces are V and {0}. Irreducible representations form
the building blocks for general representations, so that the knowledge of all irreducible
representations will suffice to classify all representations.

3.5 Schur’s Lemma

Let R : G → GL(M ; C) and S : G → GL(N ; C) be two irreducible complex representations
of the group G, such that,

AR(g) = S(g)A for all g ∈ G (3.19)

where A is an N × M matrix which is independent of g. If R and S are inequivalent then

A = 0. If R and S are equivalent, then A is unique up to a multiplicative constant. In
particular, if R = S, then A = λI with λ ∈ C.
The last assertion means that any N × N matrix which commutes for all g ∈ G with
an irreducible representation R(g) of dimension N is proportional to the identity matrix.
To prove Shur’s Lemma, denote by X and Y the vector spaces, of respective dimensions
M and N , which are transformed under R and S, so that R(g) : X → X and S(g) : Y → Y ,

46
and the matrix A is a map from X → Y . Now define the following subspaces,
X0 = Ker (A) = {x ∈ X such that Ax = 0}
Y0 = Im (A) = {y ∈ Y such that y = Ax for some x ∈ X}
It follows from the conjugation relation that X0 is a subspace of X invariant under R since
we have AR(g)x = S(g)Ax = 0 for every x ∈ X0 and thus R(g)X0 ⊂ X0 . Similarly, Y0
is a subspace of Y invariant under S. To see this, take an element y ∈ Y0 which may be
expressed as y = Ax, apply the conjugacy relation to x so that AR(g)x = S(g)Ax = S(g)y,
and observe that this implies S(g)y ∈ Y0 . Since the representations R and S have both
been assumed to be irreducible, the subspaces X0 and Y0 must both be trivial subspaces,
namely either {0} or respectively X or Y . We now examine these different cases.
The cases X0 = Y0 = {0} and X0 = X, Y0 = Y are ruled out linear algebra. The case
X0 = X and Y0 = {0} corresponds to A = 0. Finally, the case X0 = {0} and Y0 = Y
corresponds to the matrix A being invertible and the representations R and S are then
equivalent. The uniqueness up to multiplication by a constant may be seen as follows.
Suppose there were a second A0 satisfying A0 R(g) = S(g)A0 , then take the combination
(A0 − λA)R(g) = S(g)(A0 − λA) for λ ∈ C. Now let λ be a root of det (A0 − λA) = 0 (note
that it is here that we need to be in C). For that value of λ, the matrix A0 − λA is not
invertible, but then by the first part of Shur’s lemma, it must vanish. In the case where
S = R, we can clearly take A = I but since A must be unique up to multiplication by a
constant, this is in fact the only possibility, which gives rise to the last assertion.
An important application is that all irreducible representations R : G → GL(N ; C) of
an Abelian group G are one-dimensional. Note that there is no analogous result for real
representations R : G → GL(N ; R) since for example the Abelian group of multiplication
given by elements of the form,
 
a b
R(a, b) = (3.20)
−b a
for a, b ∈ R is reducible over C to a ± ib but not reducible over R.

3.6 Unitary representations

Unitary representations will be important because they preserve the Hermitian inner prod-
uct in complex vector spaces, and Hilbert spaces of quantum states or of functions. There
is also a very important immediate result, valid for representations of finite or infinite
dimension, which simplifies the study of unitary representations considerably.
Consider a complex vector space V . A Hermitian inner product is a map V × V → C,
denoted by (u, v) such that for all u, v, w ∈ V and all α, β ∈ C we have,
(v, u) = (u, v)∗
(u, αv + βw) = α (u, v) + β (u, w) (3.21)

47
One defines kuk2 = (u, u), which is a norm on V provided it is positive kuk2 ≥ 0 for
all u ∈ V ; definite so that kuk = 0 implies u = 0; and satisfies the triangle inequality
ku + vk ≤ kuk + kvk for all u, v ∈ V . The resulting normed vector space is automatically
a metric topological space. If V is complete then V is a Hilbert space.

Theorem 5 Every unitary representation is completely reducible.

To prove this result, let (x, y) be the inner product on V with respect to which the
representation R(g) is unitary, so that we have,

(R(g)x, R(g)y) = (x, y) for all x, y ∈ V (3.22)

Let V1 be an invariant subspace of the representation R(g) so that R(g)V1 = V1 . The

orthogonal complement V2 = (V1 )⊥ with respect to the inner product (x, y) is defined by,

V2 = {v ∈ V such that (v, x) = 0 for all x ∈ V1 } (3.23)

Now let y ∈ V2 then its transform R(g)y satisfies,

(R(g)y, x) = (y, R(g)−1 x) (3.24)

for all x ∈ V and thus all x ∈ V1 , and all g ∈ G. Since V1 is an invariant subspace of R,
the image R(g)−1 is in V1 , so that the second inner product vanishes by virtue of y ∈ V2 .
Therefore we have (R(g)y, x) = 0 for all x ∈ V1 so that R(g)y ∈ V2 . Therefore, V2 is an
invariant subspace under R(g) and the representation R(g) is completely reducible into the
representations on the subspaces.

Theorem 6 Every finite-dimensional representation of a finite or compact group G is

equivalent to a unitary representation.
To prove this theorem for the case of a finite group G, we consider a representation
R : G → GL(N ; C) on a vector space V of dimension N < ∞ with a non-degenerate
Hermitian inner product (x, y). An arbitrary representation R(g) may not be unitary with
respect to this inner product. We form a new inner product by summing over the group,
X
hx, yi = (R(h)x, R(h)y) (3.25)
h∈G

The sum is over a finite number of elements and thus convergent. Now transform both x, y
by the representation R(g),
X
hR(g)x, R(g)yi = (R(h)R(g)x, R(h)R(g)y) (3.26)
h∈G

48
and use the fact that R is a representation, so that R(h)R(g) = R(h ∗ g). But the sum over
h is the same as the sum over h ∗ g for any h ∈ G, so that we find,

hR(g)x, R(g)yi = hx, yi (3.27)

for all x, y ∈ V and all g ∈ G. This means that R is unitary with respect to the new
Hermitian inner product hx, yi.
The only change required for a compact group G is to replace the summation over G
by an integral over the compact Lie group G, using the Hurwitz-Haar measure dµ = dµG ,
Z
hx, yi = dµ(h)(R(h)x, R(h)y) (3.28)
G

Since G is compact and the integrand continuous in h, the integral is convergent for all
x, y. Transforming x, y under R(g), we find,
Z
hR(g)x, R(g)yi = dµ(h)(R(h ∗ g)x, R(h ∗ g)y) (3.29)
G

where we have again used the fact that R is a representation so that R(h)R(g) = R(h ∗ g).
Changing from integration variable h to h0 = h ∗ g, we use the translation invariance of the
measure dµ(h0 ∗g −1 ) = dµ(h0 ) and recover hR(g)x, R(g)yi = hx, yi so that the representation
R(g) is unitary with respect to the inner product hx, yi.

Corollary 1 It follows from combining Theorems 3.6 and 3.6 that every finite-dimensional
representation of a finite or of a compact group is completely reducible.
This result is particularly helpful as it means that for these cases it will suffice to
study just irreducible representations, all others being then given by taking direct sums of
irreducible representations.

3.7 Tensor product of representations

To define the tensor product of representations, we need to first define the tensor product
of vector spaces V1 and V2 of respective dimensions N1 and N2 . The tensor product of two
vectors v1 ∈ V1 , v2 ∈ V2 is given by a bilinear map, denoted by,

(v1 , v2 ) → v1 ⊗ v2 (3.30)

and all linear combinations of such tensor products define a vector space V1 ⊗V2 of dimension
N1 N2 . In terms of bases e11 , · · · , eN 1 N2
1 for V1 and e2 , · · · , e2 for V2 a basis of V1 ⊗ V2 may be
1

j
chosen to be ei1 ⊗ e2 where i = 1, · · · N1 and j = 1, · · · , N2 , thereby clearly giving a vector

49
space of dimension N1 + N2 . Using the decomposition of the two vectors onto these basis
vectors as given in (3.16), the tensor product v1 ⊗ v2 takes the form,
N1 X
X N2
v1 ⊗ v2 = v1i v2j ei1 ⊗ ej2 (3.31)
i=1 j=1

A general element v of the tensor product space V1 ⊗ V2 is given by a general linear

combination of the basis vectors ei1 ⊗ ej2 given as follows,
N1 X
X N2
v= v ij ei1 ⊗ ej2 (3.32)
i=1 j=1

where v ij are arbitrary coefficients in the field underlying the vector space V .
The tensor product R1 ⊗ R2 of two representations Ri : G → GL(Ni ; C) is a represen-
tation of dimension N1 N2 , namely,

R1 ⊗ R2 : G → GL(N1 N2 ; C) (3.33)

whose transformation on V = V1 ⊗ V2 is defined on the tensor product of two vectors by,

(R1 ⊗ R2 )(g)(v1 ⊗ v2 ) = (R1 (g)v1 ) ⊗ (R2 (g)v2 ) (3.34)

and in particular on basis vectors of V1 ⊗ V2 by,

N1 X
X N2
(R1 ⊗ R2 )(ei1 ⊗ ej2 ) = (R1 )ik (R2 )j` ek1 ⊗ e`2 (3.35)
k=1 `=1

where we have suppressed the dependence on g. Note that the tensor product is associative,

(R1 ⊗ R2 ) ⊗ R3 = R1 ⊗ (R2 ⊗ R3 ) for all R1 , R2 , R3 (3.36)

3.8 Characters of Representations

A group character of a representation R of a group G is a complex-valued function on G
which depends on the representation R and on the conjugacy class of each group element.
For a finite-dimensional representation R(g), we define,

χR (g) = tr R(g) (3.37)

The trace is taken of an N × N matrix where N = dim R. Invariance under conjugation

follows from the fact that R is a representation and the cyclic property of the trace,

χR (h ∗ g ∗ h−1 ) = tr R(h)R(g)R(h)−1 = trR(g) = χR (g)

(3.38)

50
One of the most important properties of group characters is that their behavior under
the operations of direct sum and tensor product of representations. For representations
R1 : G → GL(M ; C) and R2 : G → GL(N ; C) we have,

χR1 ⊕R2 (g) = χR1 (g) + χR2 (g)

χR1 ⊗R2 (g) = χR1 (g) χR2 (g) (3.39)

To prove the first line is immediate from inspection of (3.17). To prove the second line, we
recall from (3.35) that the matrix elements of R1 ⊗ R2 are given by (R1 )ik (R2 )j` . To take
the trace, we must sum the diagonal elements (R1 )ii (R2 )jj obtained by setting (k, `) = (i, j)
over the composite index ij, which clearly produces the product of the traces.
For the special case of g = e, the identity element of G, we have R(e) = IN so that,

χR (e) = dim(R) (3.40)

The direct sum and tensor product formulas for characters then reduce to the dimension
formulas derived earlier,

dim(R1 ⊕ R2 ) = dim(R1 ) + dim(R2 )

dim(R1 ⊗ R2 ) = dim(R1 ) dim(R2 ) (3.41)

For infinite dimensional representations, the definition does not make sense as it stands,
since in particular the value of the character would diverge at the identity element. Some-
times, the character may still be defined in a useful way.

Theorem 7 The following orthogonality relations hold for finite-dimensional irreducible

representations R1 and R2 of a finite group G,
X
χR1 (g) χR2 (g) = NR1 δR1 ,R2 (3.42)
g∈G

while for a compact group G with Hurwitz-Haar measure dµ we have,

Z
dµ(g)χR1 (g) χR2 (g) = NR1 δR1 ,R2 (3.43)
G

where the overline indicates complex conjugation, δR1 ,R2 equals 1 when R1 and R2 are
equivalent to one another and vanishes otherwise, and NR1 are normalizations.
The proof proceeds from Shur’s lemma. In fact we shall prove a Lemma first.

Lemma 1 Two irreducible representations R1 and R2 of a compact Lie group G (or a

finite group upon replacing the integral by a sum over G), satisfy the orthogonality relation,
Z
dµ(g) R1 (g)i j R2 (g −1 )k ` = C(R1 ) δR1 ,R2 δi ` δk j (3.44)
G

51
where δR1 ,R2 vanishes when R1 and R2 are inequivalent representations, and equals 1 when
R1 and R2 are equivalent in which case the result has been recorded for when the represen-
tations are actually equal.
To prove the lemma, we introduce the the matrix M defined by,
Z
dµ(g) R1 (g)i j R2 (g −1 )k ` = Mik j` (3.45)
G

Multiplying to the left by R1 (h)a i , summing over i to produce the product R1 (h)R1 (g) =
R1 (h ∗ g), changing variables from g to h ∗ g, and using the invariance of the measure gives
dµ(g) = dµ(h ∗ g), we find the following relation,
X Z X
a j`
R1 (h)i Mak = dµ(g) R1 (g)i j R2 (g −1 ∗ h)k ` = Mik jb R2 (h)b ` (3.46)
a G b

For fixed values of the indices j, k, denote the matrix Mik j` = Mi ` . The above relation
then implies that M satisfies R1 (h)M = MR2 (h) for all h ∈ G. By Shur’s lemma, either
R1 and R2 are inequivalent, in which case M = 0 for all values of j, k. Or R1 and R2
are equivalent in which case M is unique up to a multiplicative factor by Shur’s lemma.
Repeating the same procedure on the indices j, k we conclude the proof of the lemma.
The Theorem follows by taking the trace in the indices i, j and separately in k, `.

Theorem 8 A unitary irreducible representation of a compact group is finite dimensional.

The proof proceeds by using the orthogonality theorem, and we leave it up to the reader.

Representations of Lie algebras

So far we have discussed representations of groups, including Lie groups. The relation
between a representation ρ of a Lie algebra G and a representation R of its associated
connected and simply connected Lie group G is given explicitly by the exponential map,

R(g) = eρ(x) g = ex (3.47)

for all x ∈ G. Thus, having a representation of a Lie algebra will easily produce a repre-
sentation of the corresponding Lie group.

52
4 Representations of SL(2; C), SU (2), and SO(2, 1)
The group SU (2) is familiar to any physicist. Its elements are 2×2 complex matrices which
may be parametrized in a variety of ways. As a matrix we have,
 
a b
g= g † g = I2 det g = 1 (4.1)
c d

Equivalently, it may also be parametrized in terms of Euler angles α, β, γ,

σa
g = eiαJ3 eiβJ2 eiγJ3 Ja = (4.2)
2
where σa are the Pauli matrices.

4.1 Irreducible representations of the Lie group SU (2)

We begin by constructing the finite-dimensional irreducible representations of the Lie group
SU (2). The trivial representation is denoted D0 . The formulation of SU (2) in terms of
2 × 2 matrices provides the defining representation which is often denoted by D 1 (g) = g,
2
and which is manifestly irreducible. The direct sum of n defining representations D 12 is a
representation of dimension 2n which is obviously reducible.
To construct new irreducible representations of SU (2) we take tensor products of the
defining representation with itself and then completely reduce those into irreducible repre-
sentations. To do this, we define the vector space V = C2 on which D 1 acts, and denote
2
its vectors by vα with components vαi for i = 1, 2. The representation D 1 acts as follows,
2

2 2
i0 0
X X
vα → D 1 (g)vα vαi → D 1 (g) i
i0 vα = g i i0 vαi (4.3)
2 2
i0 =1 i0 =1

The tensor product of two vectors transforms under the tensor product representation,

vα ⊗ vβ → (D 1 ⊗ D 1 )(g)(vα ⊗ vβ ) (4.4)
2 2

Writing out the tensor products in components we have,

2 2
X 0 0 X 0 0
vαi vβj → i
D 1 (g) D 1 (g)
i0
j
j0 vαi vβj = g i i0 g j j 0 vαi vβj (4.5)
2 2
i0 ,j 0 =1 i0 ,j 0 =1

Clearly, the tensor product representation D 1 ⊗ D 1 has dimension 4.

2 2

53
 Since we are dealing with a finite-dimensional representation of a compact group, we
know by theorem 3.6 that the representation D 1 ⊗D 1 is completely reducible. Symmetriza-
2 2
tion and anti-symmetrization of the tensor product provides a general procedure to reduce
tensor product representations. In the case of D 1 ⊗ D 1 , we have,
2 2

2
1 i j X 1 0 0 0 0
i j
(vα vβ ± vβ vα ) → g i i0 g j j 0 (vαi vβj ± vβi vαj ) (4.6)
2 i0 ,j 0 =1
2

Now, the anti-symmetric combination involves,

vαi vβj ± vβi vαj = εij (vα1 vβ2 ± vβ1 vα2 ) (4.7)

where εij is the anti-symmetric symbol with ε11 = ε22 = 0 and ε12 = −ε21 = 1, and
0 0 0 0
similarly for vαi vβj ± vβi vαj . But now consider the combination,

2
0 0
X
g i i 0 g j j 0 εi j (4.8)
i0 ,j 0 =1

which is anti-symmetric in i, j and thus has a single independent component, obtained by

setting i = 1 and j = 2, and given by g 1 1 g 2 2 − g 2 1 g 1 2 = det g = 1. Thus the anti-symmetric
combination is invariant under SU (2) and transforms under the trivial representation D0 .
The symmetric combination has dimension 3, is irreducible, and is denoted D1 .

4.2 Finite-dimensional representations of sl(2, C)

The 3-dimensional Lie algebra with generators Ja for a = 1, 2, 3 and structure relations,

[J1 , J2 ] = iJ3 [J2 , J3 ] = iJ1 [J3 , J1 ] = iJ2 (4.9)

is the Lie algebra sl(2; C) or equivalently so(3; C) provided the vector space generated by
the generators Ja is over C. When the generators −iJa are real we have the algebra so(3; R),
and when the generators Ja are Hermitian it is the Lie algebra su(2) familiar from angular
momentum theory. Finally, we may also set J1 = X1 , J2 = X2 , J3 = iX3 ,

[X1 , X2 ] = −X3 [X2 , X3 ] = X1 [X3 , X1 ] = X2 (4.10)

and now require X1 , X2 , X3 to be real matrices, which gives the Lie algebra so(2, 1; R).
The Lie algebras su(2), so(3; R) and so(2, 1; R) are referred to as different real forms of the
complex Lie algebra sl(2; C), and their representation theory is closely related to that of
sl(2; C), with which we will begin.

54
 The rank of sl(2; C) is one. We choose the generator of the Cartan subalgebra to be
J3 and assume that J3 is diagonalizable.2 We organize the remaining generators in linear
combinations chosen so that the adjoint map AdJ3 is diagonal,
1
AdJ3 J± = ±J± J± = √ (J1 ± iJ2 ) (4.11)
2
In this basis, the structure relations become,

[J3 , J± ] = ±J± [J+ , J− ] = J3 (4.12)

This organization will generalize to all simple Lie algebras in which case the J± generators
will correspond to roots.
To construct the most general finite-dimensional irreducible representation ρ of sl(2; C),
we shall assume that the representation matrices ρ(Ja ) are N × N , for as yet undetermined
values of N < ∞, and satisfy the structure relations of the Lie algebra,

[ρ(J3 ), ρ(J± )] = ±ρ(J± ) [ρ(J+ ), ρ(J− )] = ρ(J3 ) (4.13)

We denote the N -dimensional vector space on which they act by V and the eigenvalues of
ρ(J 3 ) by m. Since there is no assumption of Hermiticity of the generators for sl(2; C), we
generally may have m ∈ C. The eigenspace of ρ(J3 ) may have dimension one or higher,
and we label the different vectors with the same eigenvalue m by α. The spectrum of ρ(J3 )
is the set of quantum numbers S(ρ) = {(m, α)} and has cardinality N .
We shall often use the terminology and notations of quantum mechanics, refer to vectors
as states, and denote by |ρ; m, αi the basis vectors of a basis of V in which ρ(J 3 ) is diagonal,

ρ(J 3 )|ρ; m, αi = m|ρ; m, αi (m, α) ∈ S(ρ) (4.14)

The generators ρ(J± ) map V to V by the definition of a representation, and therefore

ρ(J± )|m, α; ρi must be a linear combination of vectors |ρ; m0 , α0 i with (m0 , α0 ) ∈ S(ρ).
Actually, we may restrict the form of this linear combination by evaluating,
 
ρ(J3 )ρ(J± )|ρ; m, αi = [ρ(J3 ), ρ(J± )] + ρ(J± )ρ(J3 ) |ρ; m, αi
= (m ± 1)ρ(J± )|ρ; m, αi (4.15)

obtained using the structure relations. Thus we have the following decompositions,
X
±
ρ(J± )|ρ; m, αi = Cαβ (m, ρ)|ρ; m ± 1, βi (4.16)
β

2
When the Lie algebra is su(2), the generator J3 is Hermitian and thus automatically diagonalizable.

55
On the right side the sum is over all vectors such that (m ± 1, β) ∈ S(ρ). The matrix ρ(J+ )
raises the eigenvalue of ρ(J3 ) by 1 while ρ(J− ) lowers the eigenvalue by 1.
Next we apply ρ(J± ) repeatedly and make use of the assumption that the dimension of
the representation ρ is finite. This implies that the eigenvalue of ρ(J3 ) cannot get raised or
lowered indefinitely since this would produce a representation space of infinite dimension.
Thus, there must exist j± with non-zero vectors |j± , α± ; ρi and (j± , α± ) ∈ S(ρ) such that,

ρ(J+ )|ρ; j+ , α+ i = 0
ρ(J− )|ρ; j− , α− i = 0 (4.17)

This may occur for one or more values of j± and one or more values of α± for each value
of j± . Repeatedly applying ρ(J− ) to any one of the vectors |ρ; j+ , α+ i will produce all the
vectors of an irreducible representation of sl(2; C). If several α+ occur then applying ρ(J− )
will produce the direct sum of as many irreducible representations as there are α+ .
We shall now focus on a single irreducible representation produced from a single j+ = j
and a single α. In general, this representation is no longer ρ, but rather one of the irreducible
components of ρ. We shall denote this representation by ρj and drop the label α, labelling
the states simply |ρj ; mi. In mathematics terminology, the quantum number m is referred
to as the weight and the corresponding vector |ρj ; ji as the highest weight vector.
Applying lowering and raising operators we have,

ρj (J± )|ρj ; mi = N± (ρj ; m)|ρj ; m − 1i N+ (ρj , j) = 0 (4.18)

Enforcing the structure relation [ρj (J+ ), ρj (J− )] = ρj (J3 ) gives,

N+ (ρj , m − 1)N− (ρj , m) − N− (ρj , m + 1)N+ (ρj , m) = m (4.19)

Considering the sum of this relation over m up to m = j,

j   j
X X
N+ (ρj , n − 1)N− (ρj , n) − N− (ρj , n + 1)N+ (ρj , n) = n (4.20)
n=m n=m

The second term in the parentheses is the opposite of the first shifted by n → n + 1. Thus
the sum is given by its endpoints and, using the fact that N+ (ρj , j) = 0, we obtain,

2N+ (ρj , m − 1)N− (ρj , m) = (j + m)(j − m + 1) (4.21)

The formula reproduces N+ (ρj , j) = 0 and shows that the only value of m < j for which
N− can vanish is m = −j, so that N− (ρj , −j) = 0. Given the highest weight j of the
representation ρj , the allowed values of m are such that j − m ∈ Z+ , where Z+ stands for

56
the zero or positive integers. But −j must be such a value, so that 2j ∈ Z+ . Therefore the
states in the representation ρj are given for j − m ∈ Z+ and,

|ρj ; mi 0 ≤ j − m ≤ 2j ∈ Z+ (4.22)

and the dimension of the representation is dim(ρj ) = 2j + 1. Note that we did not use the
assumption that m or j had to be real from the outset: for sl(2; C) this result arises solely
from requiring the representations to be finite-dimensional.
Imposing the restrictions of Hermiticity for su(2), or reality for so(3; R) changes nothing
to the structure and dimensions of the representations derived above. In the case of com-
pact groups we know that all finite-dimensional representations are equivalent to unitary
representations.

4.3 Infinite-dimensional representations of sl(2; C)

Now let us remove the assumption that the representation ρ is finite-dimensional, but still
assume that ρ(J3 ) is diagonalizable, so that we still have the relations,

N+ (ρ; m − 1)N− (ρ; m) − N− (ρ; m + 1)N+ (ρ; m) = m (4.23)

Consider now a representation ρ for which N− (ρ; j) = 0, so that ρ(J− )|ρ; ji = 0 for a
non-zero state |ρ; ji. This time applying m raising operators to |ρ; ji, we find,

−2N+ (ρ; j + m)N− (ρ; j + m + 1) = (m + 1)(2j + m − 2) (4.24)

If 2j were a negative integer, then we recover the finite-dimensional case already discussed.
But if we assume that 2j is not a negative integer, then we can apply an infinite number
of raising operators for which N+ (ρ; j + m) 6= 0 and thus |ρ; j + mi =
6 0. This produces an
infinite-dimensional representation of sl(2; C).

4.4 Harmonic oscillator representation of so(2, 1; R)

Consider the one-dimensional harmonic oscillator with degrees of freedom a, a† which satisfy
the Heisenberg algebra [a, a† ] = 1 upon setting ~ = 1, and have the following Hamiltonian
and additional generators,
1
H = a† a + J+ = (a† )2 J− = (a)2 (4.25)
2
They satisfy the structure relations of so(2, 1; R) given by,

[H, J± ] = ±2J± [J+ , J− ] = −4H (4.26)

57
Clearly, J+ is a raising operator for H while J− is a lowering operator. The quadratic
Casimir take the form,

1
C = H 2 − {J+ , J− } = H 2 − 2H − J+ J− (4.27)
2

and on the harmonic oscillator representation evaluates to C = − 34 . Clearly H is positive,

and there must thus be a lowest energy state, which we shall denote |0i such that,

H|0i = E0 |0i J− |0i = 0 (4.28)

for an as yet undetermined ground state energy E0 . Applying C to |0i gives,

3 1
− = E02 − 2E0 or E0± = 1 ± (4.29)
4 2
Applying raising operators gives the excited states |ni = J+n |0i with energy,

1
En± = 1 ± + 2n n≥n (4.30)
2

The interpretation is as follows: while the operator a† raises the energy by one unit, the
raising operator J+ which is in the algebra so(2, 1; R) raises the energy by two units since
it is quadratic in a† . The two different “ground states” correspond to the even and odd
wave functions of the harmonic oscillator, and are the lowest weight states of different
representations of so(2, 1; R). Thus the entire spectrum of the harmonic oscillator is the
direct sum of two irreducible representations.

4.5 Unitary representations of so(2, 1; R)

We consider a quantum system in one dimension which generalizes the harmonic oscillator,
governed by the Hamiltonian H, and the following additional generators,

p2 g x2
H = + 2+
2 2x 2
p2 g x2
K = + 2−
2 2x 2
i
D = (xp + px) (4.31)
2
where x, p satisfy the Heisenberg algebra, [x, p] = i. The operator D generates dilations,
and has been normalized so that [D, x] = x and [D, p] = −p. From a quantum point of
view, the operators H, K and iD are self-adjoint for g ∈ R, assuming some restrictions

58
on the values of g to be spelled out later. Together with the operators K and D, the
Hamiltonian forms the so(2, 1; R) algebra, with the following structure relations,

[D, H] = −2K [D, K] = −2H [H, K] = 2D (4.32)

For g ≥ 0, the spectrum of H is discrete, real, and positive, and we choose to work in a
basis where H is diagonal. Next, we organize the operators K and D in combinations that
raise and lower the eigenvalues of H, i.e. that diagonalize AdH ,

J± = K ± D [H, J± ] = ±2J± [J+ , J− ] = −4H (4.33)

Since K is self-adjoint and D is anti-self-ajoint, we see that (J± )† − J∓ , as in the case of the
harmonic oscillator. One verifies using the structure relations that the quadratic Casimir
operator is given by

C = H 2 − K 2 + D2 = H 2 − 2H − J+ J− (4.34)

The calculation of C for the particular quantum system considered here may be simplified
by using the canonical commutation relations [x, p] = i, and we find that C is proportional
to the identity operator, and given by C = g − 43 . Clearly, the spectrum of H is real and
bounded from below, so there must be a state of lowest eigenvalue of E0 ,

H|0i = E0 |0i J− |0i = 0 (4.35)

The value of E0 is determined by using the value of the Casimir, and we get the entire
spectrum from applying the raising operator,
1p 1p
E0± = 1 ± 1 + 4g En± = 1 ± 1 + 4g + 2n n ∈ Z+ (4.36)
2 2
The spectrum makes sense even when g is negative as long as g ≥ − 41 . The entire spectrum
again would appear to be the direct sum of two irreducible representations, as was the case
for the harmonic oscillator. However, it is clear from the Hamiltonian that just its potential
√
part is bounded from below by g, so that the energy E0± must be bounded from below
√
by g. This is always the case for the E0+ branch, but holds for the E0− branch only when
− 14 ≤ g ≤ 64
9
. The energy also needs to be bounded from below by the harmonic oscillator
value, and the combined bound gives that the E0− branch can exist only for − 14 ≤ g ≤ 0.

Theorem 9 The irreducible unitary representations of SO(2, 1; R) are as follows.

1. discrete series: C ≥ −1 and there is a lowest (or highest) energy state with energy
E0 ≥ 0 (or ≥ − 41 in the above example);

2. continuous series: C < −1 and 0 ≤ E0 < 2;

3. complementary series: C = −1 + λ2 and λ ∈ R.

59
4.6 An example of the continuous series
Let me now give an example of a physically relevant representation that produces the
continuous series. We note that SO(2, 1) acts transitively on the hyperboloids,

H± = {(x, y, z) ∈ R2 x2 + y 2 − z 2 = ±1} (4.37)

While H+ is connected by not simply connected, H− has two connected components, each
of which is simply connected. Note that the compact subgroup SO(2) of SO(2, 1) rotates
(x, y). Both surfaces H± have constant negative curvature. To find the representations in
the continuous series, we diagonalize the Laplace-Beltrami operator acting on functions on
either one of these hyperboloids,
1 √
∆g = − √ ∂m g g mn ∂n (4.38)
g
and the Laplace operator represents the Casimir operator. We consider the connected
component of H− which is characterized by z ≥ 1, and introduce the following coordinates,

 x = shϕ cos θ
H− : y = shϕ sin θ (4.39)
z = chϕ


where 0 ≤ θh2π and 0 ≤ ϕ < ∞. The metric is readily obtained,

ds2 = dϕ2 + sh2 ϕ dθ2 (4.40)

and so is the Laplace operator,

∂2 chϕ ∂ 1 ∂2
∆=− − − (4.41)
∂ϕ2 shϕ ∂ϕ sh2 ϕ ∂θ2
The differential operators of SO(2, 1) in this representation are given by,
∂
T3 = −2i
∂θ
 
±iθ ∂ chϕ ∂
T± = 2e i ∓ (4.42)
∂ϕ shϕ ∂θ
Extra factors of 2 have been inserted so that the algebra is the same as in the preceding
examples, and the structure relations are as follows,

[T3 , T± ] = ±2T± [T+ , T− ] = −4T3 (4.43)

The Casimir operator is,

C = T32 − 2T3 − T+ T− = −4∆ (4.44)

60
We now solve for the eigenvalue equation:

1 ∂ 2 ψ chϕ ∂ψ 1 ∂ 2ψ
∆ψ = − Cψ = − 2 − − (4.45)
4 ∂ϕ shϕ ∂ϕ sh2 ϕ ∂θ2

We shall use the following parametrization for λ2 ∈ R,

C = −1 − λ2 (4.46)

and diagonalize T3 simultaneously with ∆, by introducing the eigenfunctions of ∂/∂θ,

ψλ,m (ϕ, θ) = eimθ ψλ,m (ϕ) (4.47)

so that the remaining eigenvalue equation reduces to,

m2
 
00 chϕ 0 1 2
−ψλ,m − ψ + ψλ,m = + λ ψλ,m (4.48)
shϕ λ,m sh2 ϕ 4

This equation is seen to be the Legendre equation by changing variables x = chϕ, and
relabeling the parameter λ in the equation as follows,
1 1
ν(ν + 1) = − − λ2 ν = − + iλ (4.49)
4 2
which converts the eigenvalue equation into the standard form for the Legendre functions,

m2
 
2 00 0
(1 − x )ψ − 2xψ + ν(ν + 1) − ψ=0 (4.50)
1 − x2

The solutions are conical functions P−m1 +iλ (chϕ) and Qm

− 21 +iλ
(chϕ) for which we have usual
2
completeness relations.

61
5 Tensor representations
Vector and tensor algebra is a topic in physics that has a somewhat separate life from group
theory but is in fact equivalent to the representation theory for the classical groups as long
as no spinors are involved. Tensor methods are organized by the pictorial Young tableaux
and are often more intuitive and practical when only representations of low dimension are
involved, and are

5.1 Tensor product representations

We consider one of the classical matrix Lie groups G, such as SL(n), SU (n), SO(n), or
Sp(n) (for n even), and start out with their respective defining representations, which are
all of dimension n. The associated vector space Vn is Rn or Cn depending on the group.
The defining representation matrix will be denoted R(g) for g ∈ G and U (g) and n × n
matrix. The group G acts on Vn by linear transformations on its vectors,
G : Vn → Vn v → R(g)v ∈ Vn v ∈ Vn (5.1)
or in component notation,
n
0
X
i
v → R(g)i i0 v i i = 1, · · · , n (5.2)
i0 =1

The transformation of Vn under the defining representation R of G induces a transformation

on the tensor product space Vn ⊗ Vn under the tensor product representation R ⊗ R,
 
G : Vn ⊗ Vn → Vn ⊗ Vn v1 ⊗ v2 → R(g) ⊗ R(g) (v1 ⊗ v2 ) (5.3)

Note that the group element g is the same in both factors. Equivalently, in components,
n
X 0 0
v1i v2j → R(g)i i0 R(g)j j 0 v1i v2j (5.4)
i0 ,j 0 =1

By taking linear combinations in Vn ⊗ Vn of tensor products of vectors we obtain general

element w ∈ Vn ⊗ Vn whose components may be denoted by wij for i, j = 1, · · · , n and is
referred to as a tensor of rank 2 since it has two indices of the defining representation, and
dimension n2 . It transforms under R ⊗ R by,
G : Vn ⊗ Vn → Vn ⊗ Vn w → (R(g) ⊗ R(g))w (5.5)
or in components,
n
0 0
X
ij
w → R(g)i i0 R(g)j j 0 wi j (5.6)
i0 ,j 0 =1

62
The process may be repeated by taking the tensor product of r copies of Vn to obtain tensors
of rank r which transform under the r-fold tensor product of the defining representation R.

G : Vn ⊗ · · · ⊗ Vn → Vn ⊗ · · · ⊗ Vn w → R(g) ⊗ · · · ⊗ R(g) w (5.7)

| {z } | {z } | {z }
r r r

or in components,
n
X
i1 ,··· ,ir
w → R(g)i1 j1 · · · R(g)ir jr wj1 ,··· ,jr (5.8)
i1 ,··· ,ir =1

When no confusion is expected to arise, we shall henceforth omit exhibiting the g-dependence.
The representations thus constructed are tensor product representations of the defining
representation. For the groups SL(n), SU (n) and Sp(n) we will show later that all finite-
dimensional representations may be constructed from tensor product representations. For
the groups SO(n), however, tensor product representations will give only a subset of all
representations, as these groups have also spinor representations which cannot be obtained
from tensor product representations.

5.2 Symmetrization and anti-symmetrization

The tensor product representations constructed above are in general reducible. Sym-
metrization and anti-symmetrization always provides operations which will reduce the
representations into direct sums of representations of smaller dimension, though the rep-
resentations obtained this way may or may not be irreducible. We begin by symmetrizing
and anti-symmetrizing the tensor product of the representation spaces Vn ⊗ Vn by defining
the operations of symmetrization S and anti-symmetrization A on Vn ⊗ Vn ,
1 1
S(v1 ⊗ v2 ) = (v1 ⊗ v2 + v2 ⊗ v1 ) S(v1i v2j ) = (v1i v2j + v1j v2i )
2 2
1 1
A(v1 ⊗ v2 ) = (v1 ⊗ v2 − v2 ⊗ v1 ) A(v1 v2 ) = (v1i v2j − v1j v2i )
i j
(5.9)
2 2
The symmetrized and anti-symmetrized tensor product representations then act as follows,
n
X 1 i j
RS : S(v1i v2j ) → (R k R ` + Rj k Ri ` )S(v1k v2` )
k,`=1
2
n
X 1 i j
RA : A(v1i v2j ) → (R k R ` − Rj k Ri ` )A(v1k v2` ) (5.10)
k,`=1
2

The dimensions of the representations RS and RA are respectively given by 21 n(n + 1) and
1
2
n(n − 1) whose sum is indeed the dimension of R ⊗ R, which is n2 .

63
5.3 Representations of SU (3)
Denote the defining representation of SU (3) by R : SU (3) → GL(3; C). Actually, since R
is a representation of SU (3), then so is the complex conjugate representation R∗ . We shall
now show that the complex conjugate representation R∗ is equivalent to the antisymmetric
tensor product of R with itself. To see this concretely, define the totally anti-symmetric
tensor in three indices,

(−)σ if (ijk) = σ(123) σ ∈ S3
εijk = (5.11)
0 otherwise
Next, we consider the relation det R = 1, expressed in terms of the ε-symbol,
3
X
εαβγ Ri α Rj β Rk γ = εijk det R = εijk (5.12)
α,β,γ=1

Multiplying by (R† )δ k on both sides, summing over k and using the unitarity relation
† k γ γ
P
k (R )δ Rk = δδ , we have,

3
X 3
X
εαβδ Ri α Rj β = εijk (R† )δ k (5.13)
α,β=1 k=1

Multiply by εij` on both sides and sum over i, j using the relation,
3
X
εijk εij` = 2δk` (5.14)
i,j=1

gives,
3
X 3
X
εijk εαβδ Ri α Rj β = 2(R† )δγ = 2Rγδ
∗
(5.15)
i,j=1 α,β=1

Therefore, the representation R∗ is equivalent to the anti-symmetrized tensor product of

two defining representations.
We have learned something interesting: each time we anti-symmetrize the tensor prod-
uct of two defining representations, we may replace that representation by its equivalent R∗ ,
while the totally anti-symmetrized tensor product of three R gives a singlet representation.
We shall use the convention that the vectors in the vector space under which the defining
representation transforms are denoted with a lower index such as vi , while a vector in the
dual vector space which transforms under R∗ will be denoted with a barred upper index,
v ī . Thus, a general tensor of SU (3) is of the form,

Ti1 ···ip ī1 ···īp̄ (5.16)

64
To reduce the tensor, we separately symmetrize it in the upper and lower indices. To fully
reduce the tensor, we remove all the traces between upper and lower indices.
To count the dimension of the irreducible representation with p defining and p̄ anti-
defining representations, we first ignore the trace condition and count the dimension of
just the tensor ti1 ···ip . Since we have total symmetrization, it suffices to count the number
of ways we can put the three possible values i = 1, 2, 3 in three boxes with the sum of
the lengths of the boxes being n. Denote by p1 , p2 , p3 the number of values 1, 2, and 3
respectively, so that p1 + p2 + p3 = n. Each pi satisfies 0 ≤ pi ≤ n. thus the dimension is,
n n−p
X1 n
X X (p + 2)!
dim(t) = 1= (n + 1 − p1 ) = (5.17)
p1 =0 p2 =0 p1 =0
2! p!

Thus, the dimension of the tensor T fully symmetrized in upper and lower indices, but
ignoring the trace is given by,

(p + 2)! (p̄ + 2)!

(5.18)
2! p! 2! p̄!

Removing the traces is equivalent to removing the tensor with any pair of upper and lower
indices contracted in view of the symmetry of the tensor. Thus, the dimension of the
irreducible tensor T is given by,

(p + 2)! (p̄ + 2)! (p + 1)! (p̄ + 1)!

D(p, p̄) = −
2! p! 2! p̄! 2! (p − 1)! 2! (p̄ − 1)!
1
= (p + 1)(p̄ + 1)(p + p̄ + 2) (5.19)
2
We can check this formula for example on the adjoint representation, which is of the form
Ti j so that p = p̄ = 1, and we find dim(T ) = 8, which is the famous meson multiplet of
π 0 , π ± , K ± , K 0 , K̄ 0 , η, when only the u, d, s quarks are included. Another is the multiplet
in which the proton and neutron live, which is the symmetric tensor product of 3 defining
representations, so that p = 3 and p̄ = 0, and we get dim(T ) = 10. This is the multiplet
that contains the Ω− consisting of three s quarks, which was predicted on the basis of
SU (3) group theory by Murray Gell-Mann for which he got the Nobel Prize.
Young tableaux provide a pictorial image of irreducible representations. One begins by
denoting the defining representation by a single square 2, and the trivial representation by
a center dot •. Taking the tensor product with another copy of the defining representation
is indicated by adjoining another box, either to the right or underneath the first box. In the
vertical alignment of boxes, the tensor product is anti-symmetrized, while in the horizontal
alignment, they are symmetrized. For SU (3), the anti-symmetric tensor product of three
defining representations is the trivial representation, so non-trivial representations can have

65
 p̄ p
··· ···
···

either one or two vertically aligned boxes, and the most general irreducible representation
labelled by (p, p̄) constructed above has the following Young tableaux,
On may now also give a pictorial representation for tensor multiplication of an arbitrary
representation by the defining representation: the result is the sum of three irreducible
representations, each one obtained by adding a square in each row to the utmost left.
Adding the square in the third row makes the corresponding column a singlet and produces
the representation (p, p̄ − 1). Adding the square in the second row diminishes p by one and
increases p̄ by one, so we have the representation (p − 1, p̄ + 1), and finally in the first row
we get (p + 1, p̄). One verifies that the dimensions add up:
3D(p, p̄ − 1) = D(p − 1, p̄ + 1) + D(p, p̄) + D(p + 1, p̄) (5.20)
There are rules for taking the tensor product not just with the defining representation, but
with a general representation. We shall see an easier route using weights.

5.4 Representations of SU (n)

For the case of SU (n), we introduce the totally anti-symmetric symbol in n indices,

(−)σ if ik = σ(k) σ ∈ Sn
εi1 ···in = (5.21)
0 otherwise
The determinant relation is now,
X
εα1 ···αn Ri1 α1 · · · Rin αn = εi1 ···in det R = εi1 ···in (5.22)
α1 ,··· ,αn

Multiplying by (R† )k in and summing over in gives,

X X
εα1 ···αn−1 k Ri1 α1 · · · Rin−1 αn−1 = (R† )k in εi1 ···in (5.23)
α1 ,··· ,αn in

The remaining step is straightforward and now proves that R∗ is equivalent to the (n − 1)-
fold anti-symmetric tensor product of the defining representation.
Are the symmetric/antisymmetric reductions now irreducible representations? This
depends upon what the group is. In general, for representations of
G`(n, C) S,A
SU (n) S,A, δij
SO(2n + 1), SO(2n) δij
Sp(2n) Jij

66
 By isolating the parts proportional to these tensors, they are further reduced.
Some illustrative examples.
SO(n) n

Ti1 ...ip (5.24)

Now you can take the trace as well on symmetric indices

T i1 ...ip−2 = Ti1 ...ip δ ip−1 ip (5.25)

{ip−1 ip }

There is also the fully antisymmetric tensor, but it does not lead to new reductions.
Sp(2n) One has the defining representation 2n.
Now there is a trace with respect to Jij : on antisymmetric indices

Ti1 ...ip−2 =Ti1 ...ip Jip−1 ip [ip−1 ip ] (5.26)

For exceptional groups (they are all subgroups of classical groups) there is usually yet
another invariant tensor, which further reduces the representations.

67
6 Spinor representations
In this section, we discuss spinor representations of the orthogonal Lie algebras so(d; R)
and so(d − 1, 1; R), starting from the respective Clifford algebras. We begin with a brief
review of the spinor representation of the group of rotations SO(3; R).

6.1 Spinor representations of SO(3; R)

The fundamental representation of SO(3; R) has dimension 3. All the higher dimensional
representations may be obtained by tensor product decomposition, and give us the irre-
ducible representations of integer spin j. So where do the spinor representations enter ?
There are two ways of looking at it. First, look at the generators of the Lie algebra so(3; R),
     
0 0 0 0 0 1 0 −1 0
L1 = 0 0 −1 L2 =  0 0 0  L 3 = 1 0 0  (6.1)
0 1 0 −1 0 0 0 0 0

which satisfy the following structure relations,

[L1 , L2 ] = L3 [L2 , L3 ] = L1 [L3 , L1 ] = L2 (6.2)

The group elements g ∈ SO(3; R) may then be parametrized by Euler angles, α, β, γ,

g(α, β, γ) = eαL3 eβL1 eγL3 0 ≤ α, β < π, 0 ≤ γ < 2π (6.3)

It may be verified that the above range of parameters covers SO(3; R) exactly once. From
the point of view of the Lie algebra, the spinor representations enter because the structure
relations are also satisfied by setting La = −i σ2a where σa are the Pauli matrices. This
gives a 2-dimensional representation R which cannot be obtained from the tensor product
of the defining representation of SO(3; R). But actually, it is not a genuine representation
of SO(3; R), because the Euler angle parametrization now gives,

R(g) = e−iασ3 /2 e−iβσ1 /2 e−iγσ3 /2 0 ≤ α, β < π, 0 ≤ γ < 2π (6.4)

While g was periodic with period 2π in γ, this is not the case for R(g) since we have,

R(g(α, β, γ + 2π)) = −R(g(α, β, γ + 2π)) (6.5)

so that R(g) is actually double-valued function of g, and the representation R is said to

be double-valued or to be a projective representation of SO(3; R). But R(g) is a single-
valued representation of SU (2), and in fact coincides with its defining representation. The
sign ambiguity is absent in SO(3; R) because it is isomorphic to SU (2)/Z2 , the Z2 exactly
accounting for the sign indeterminacy.

68
6.2 The Clifford Algebra
The generalization to spinor representations for higher rank orthogonal groups makes use of
the Clifford algebra whose representations induce spinor representation in the orthogonal
groups. We begin by recalling the parametrization and the structure relations of so(d).
The generators Mµν are labelled by a composite index (µν) with 1 ≤ µ, ν ≤ d and they are
anti-symmetric, so that Mνµ = −Mµν . The independent generators may thus be chosen to
be Mµν with 1 ≤ µ < ν ≤ d. Their matrix elements are given by,
(Mµν )αβ = δµα δνβ − δµβ δνα (6.6)
The structure relations are given by,
[Mµν , Mρσ ] = δνρ Mµσ − δνσ Mµρ − δµρ Mνσ + δµσ Mνρ (6.7)
Note that one should think of the Kronecker symbols δµν etc as representing the identity
matrix in so(d; R). It is possible to generalize the construction of spinors to so(p, q; R)
with pq 6= 0 and δµν will then be replaced by the matrix elements of Ip,q , and indeed we
shall do so for the Minkowski metric later. In the special case of d = 3, we may recast the
generators Mµν in terms of the notation La used earlier for so(3) as follows,
L1 = M32 L2 = M13 L3 = M21 (6.8)
We shall now introduce the Clifford algebra to produce representations of this Lie algebra.
The Clifford algebra in d dimensions, for Euclidean signature, is defined by,
{γµ , γν } = 2δµν I µ, ν = 1, · · · , d (6.9)
Now let’s assume that we can construct a representation of the Clifford algebra (we shall
do so explicitly soon). The claim is then that if γµ forms a representation of the Clifford
algebra, then we automatically obtain a representation of the Lie algebra so(d) with the
following representation matrices,
1 1
Mµν = [γµ , γν ] = (γµ γν − γν γµ ) (6.10)
4 4
This result is so simple and so fundamental that it is worth proving it explicitly by com-
puting the commutator,
1
[Mµν , Mρσ ] = [γµ γν , γρ γσ ] (6.11)
4
where in each entry of the commutator, we have used the Clifford algebra relations to
convert γν γµ into −γµ γν plus the identity matrix which cancels out of the commutator.
Next, we write out the commutator,
1 1
[Mµν , Mρσ ] = γµ γν γρ γσ − γρ γσ γµ γν (6.12)
4 4

69
By systematically permuting the γ-matrices to the same position with the help of the
Clifford relations, we get,

1 1 1 1
[Mµν , Mρσ ] = γµ {γν , γρ }γσ − γρ {γσ , γµ }γν − {γµ , γρ }γν γσ + γρ γµ {γσ , γν } (6.13)
4 4 4 4
Using now the Clifford algebra relation to convert all anti-commutators, we find,

1 1 1 1
[Mµν , Mρσ ] = δνρ γµ γσ − δµσ γρ γν − δµρ γν γσ + δνσ γρ γµ (6.14)
2 2 2 2
Using the Clifford algebra relations to anti-symmetrize the products of γ-matrices we pro-
duce terms proportional to the identity matrix, which all cancel. We find that the matrices
M indeed obey the structure relations of so(d; R). To see why the representation matrices
have anything to do with spinors, we need to construct representations of the γ-matrices.

6.3 Representations of the Clifford algebra

For d = 3 the representation of the Clifford algebra is in terms of the Pauli matrices,
     
0 1 0 −i 1 0
σ1 = σ2 = σ3 = (6.15)
1 0 i 0 0 −1

The Pauli matrices may also be given in the raising/lowering basis,

   
1 0 1 1 0 0
σ+ = (σ1 + iσ2 ) = σ− = (σ1 − iσ2 ) = (6.16)
2 0 0 2 1 0

where they satisfy [σ3 , σ± ] = ±σ± and [σ+ , σ− ] = σ3 . For d = 4, it was Dirac who, in one
of his strokes of genius, wrote down the Dirac representation in terms of 4 × 4 matrices for
the case of the Minkowski metric. What he used was, effectively, taking the tensor product
of the Pauli matrices, though he did not use that terminology. He used this representation
to construct the Dirac equation which led to a revolution in physics.
For general d we proceed by transforming the Clifford algebra into an algebra of
fermions, by defining the following operators, for j = 1, · · · n,

1 1
bj = (γ2j−1 − iγ2j ) b†j = (γ2j−1 + iγ2j ) (6.17)
2 2

when d = 2n is even. When d = 2n + 1 is odd, we add b0 = b†0 = γ2n+1 . Expressing the

Clifford algebra relations in terms of the bi and b†i , we find the following algebra,

{bj , b†k } = δjk I {bj , bk } = {b†j , b†k } = 0 j, k = 1, · · · , n (6.18)

70
when d is even, and when d is odd we supplement the above relations with the anti-
commutators with b0 ,
(b0 )2 = I {bj , b0 } = {b†j , b0 } = 0 for d odd (6.19)
The algebra for d = 2n even consists of n complex fermion operators which are independent
of one another.
This means that their representation is a tensor product of the representations for each
one of the fermion oscillator pairs. We are interested in finite-dimensional representations,
because we wish to construct finite-dimensional representations of so(d). To construct the
representation for a single pair, bi , b†i we use the fact that it is finite-dimensional so that
there must be some state |−ii 6= 0 in the representation for which bi |−ii = 0. Applying b†i
to |−ii , we obtain a new state |+ii = b†i |−ii . This state is different from |−ii because when
we apply bi to it, the result is |−ii which is non-zero while applying bi to |−ii gave zero.
But now were are finished because no other states are obtained by applying the generators
bi or b†i . Thus the irreducible representation is two-dimensional, and in fact it is just the
algebra of σ± and σ3 . To obtain the representation of the full Clifford algebra for arbitrary
d, we take the tensor product ground state,
|0i = |−i1 ⊗ |−i2 ⊗ · · · ⊗ |−in (6.20)
It is annihilated by all the bi for i = 1, · · · , n, and is an eigenstate of b0 if d is odd. The
most general state is a linear combination of the states,
(b†1 )m1 (b†2 )m2 · · · (b†n )mn |0i (6.21)
where each mi can take the values 0 or 1, giving a total of 2n states. Thus the representation
n
has dimension 2n = 2[ 2 ] , and it is irreducible for all d.
We may equivalently return to the representation in terms of the original γ-matrices.
The states map as follows,
   
1 0
|+ii → u+ = |−ii → u− = (6.22)
0 1
and the ground state is given by,
|0i → u− ⊗ u− ⊗ · · · ⊗ u− (6.23)
| {z }
n

The individual generators map as follows bi → σ− and b†i → σ+ , from which we can now
construct the representations of the b and b† operators,
b1 = σ− ⊗ I ⊗ I ⊗ · · · ⊗ I ⊗ I b†1 = σ+ ⊗ I ⊗ I ⊗ · · · ⊗ I ⊗ I (6.24)
b2 = σ3 ⊗ σ− ⊗ I ⊗ · · · ⊗ I ⊗ I b†2 = σ3 ⊗ σ+ ⊗ I ⊗ · · · ⊗ I ⊗ I
··· ···
bn = σ3 ⊗ σ3 ⊗ σ3 ⊗ · · · ⊗ σ3 ⊗ σ− b†n = σ3 ⊗ σ3 ⊗ σ3 ⊗ · · · ⊗ σ3 ⊗ σ+

71
Finally, converting σ± into σ1 and σ2 and bi and βi† into γµ , we obtain, the expressions for
the γ matrices,

γ1 = σ1 ⊗ I ⊗ I ⊗ · · · ⊗ I ⊗ I γ2 = σ2 ⊗ I ⊗ I ⊗ · · · ⊗ I ⊗ I (6.25)
γ3 = σ3 ⊗ σ1 ⊗ I ⊗ · · · ⊗ I ⊗ I γ4 = σ3 ⊗ σ2 ⊗ I ⊗ · · · ⊗ I ⊗ I
γ5 = σ3 ⊗ σ3 ⊗ σ1 ⊗ · · · ⊗ I ⊗ I γ6 = σ3 ⊗ σ3 ⊗ σ2 ⊗ · · · ⊗ I ⊗ I
··· ···
γ2n−1 = σ3 ⊗ σ3 ⊗ σ3 ⊗ · · · ⊗ σ3 ⊗ σ1 γ2n = σ3 ⊗ σ3 ⊗ σ3 ⊗ · · · ⊗ σ3 ⊗ σ2

We also define the matrix

γ̄ = σ3 ⊗ σ3 ⊗ σ3 ⊗ · · · ⊗ σ3 ⊗ σ3 (6.26)

which satisfies

γ̄ 2 = I {γ̄, γj } = 0 (6.27)

If d is even, so that d = 2n, then γµ , µ = 1, · · · , 2n defined above, indeed satisfy (6.9). If d

is odd, we define γd = γ̄, and then γµ , i = 1, · · · , 2n + 1 satisfies the Clifford algebra. By
construction, all 㵠are Hermitian, 㵆 = 㵠.
The γ-matrices form only a small subset of all 2n ×2n matrices. All the other generators
of 2n × 2n matrices are obtained by anti-symmetrized products of the basic γµ ,

γµ1 µ2 ···µp ≡ γ[µ1 γµ2 · · · γµp ] (6.28)

where [· · · ] denotes anti-symmetrization of the p indices. The identity I together with all
matrices γµ1 µ2 ···µp for 1 ≤ p ≤ n forms a complete basis, with the following properties,
1
(γµ1 µ2 ···µp )† = (−1) 2 p(p−1) γµ1 µ2 ···µp
γ̄ γµ1 µ2 ···µm = (−1)p γµ1 µ2 ···µp γ̄ (6.29)

6.4 Spinor representations of so(d)

We have shown earlier that the generators Σµν = 41 [γµ , γν ] satisfy the structure relations of
so(d). Here we shall provide more details and show why this is the spinor representation.
We shall denote the representation matrices of the spinor representation of SO(d) by S(g)
where g is an element of SO(d). Near the identity g ≈ I, we have,

gµ ν = δν ν + ωµ ν + O(ω 2 )
d
1 X
S(g) = I + ωµν Σµν + O(ω 2 ) (6.30)
2 µ,ν=1

72
The spinor representation S(g) leaves the γ-matrices invariant,
d
X
gµ ν S(g)γν S(g)−1 = γµ (6.31)
ν=1

a relation which follows from integrating the infinitesimal version,

[Σµν , γρ ] = δµρ γν − δνρ γµ (6.32)
Notice that the generators have been normalized here to be anti-hermitian.
The reason this representation is a spinor representation may be seen by performing a
rotation in just a single direction, say by,
i
Σ12 = σ3 ⊗ I| ⊗ I ⊗
{z· · · ⊗ I} (6.33)
2
n−1

But this matrix is proportional to a Pauli matrix with coefficient 1/2, so that a rotation by
2π will again produce a minus sign. This double-valuedness is the characteristic property
of a spinor representation. For low rank, it may be understood as it was for the SO(3; R)
case in terms of SU (2), as we have,
SO(4; R) = SU (2) × SU (2)/Z2
SO(5; R) = U Sp(4)/Z2
SO(6; R) = SU (4)/Z2 (6.34)
In each case the Z2 identification takes care of the double-valuedness. But the special
identifications end there. For d ≥ 7, one sometimes refers to the simply-connected double
cover of SO(d) as Spin(d), so that SO(d) = Spin(d)/Z2 .

6.5 Reducibility and Weyl spinor representations

For d = 2n + 1 odd, the Dirac spinor representation is irreducible. However, for d = 2n
even, the chirality matrix γ̄ commutes with the representation Σµν ,
[γ̄, Σµν ] = 0 (6.35)
Since γ̄ 2 = I but trγ̄ = 0, we see that γ̄ is not proportional to the identity matrix.
Putting these two observations together, we see that the Dirac spinor representation Σ
must reducible into two representations of half the dimension of 2n . The components are
referred to as Weyl spinor representations Σ+ and Σ− , and are defined by projection onto
the ±1 eigenspaces of γ̄ by,
1
Σ± ≡ (I ± γ̄)Σµν (6.36)
2
each of which is irreducible.

73
6.6 Charge and complex conjugation
If the matrices γµ satisfy the Clifford algebra relations, then so must their complex con-
jugate matrices γµ∗ . The matrices γµ are not all real. Since we have found only a single
irreducible representation of the Clifford algebra, and hence γµ∗ must actually correspond
to an equivalent representation. Therefore, there exists an invertible matrix C such that,3

γµ∗ = −Cγµ C −1 (6.37)

for all 1 ≤ µ ≤ d. The matrix C is referred to as the charge conjugation matrix. We may
compute C explicitly, in the basis given above for example for d = 2n even, by observing
that half of the matrices γµ are real symmetric while the other half are imaginary anti-
symmetric. To find C, we choose a basis, such as the one above, and we have

Cγµ C −1 = −γµ µ odd

Cγµ C −1 = +γµ µ even (6.38)

It is easy to find C, up to an overall multiplicative factor, by considering the product of

the n symmetric Dirac matrices, together with a power of Γ. Choosing the multiplicative
factor as follows, we find,

C = (γ̄)n γ1 γ3 γ5 · · · γ2n−1 (6.39)

The matrix C then satisfies,

C †C = C tC = I
1
C t = (−1) 2 n(n+1) C
1
C 2 = (−1) 2 n(n+1) I (6.40)

The complex conjugation properties of the representation Σ of so(d) are then give as follows,

±
C Σij C −1 n even


∗
Σ±
ij = (6.41)
C Σ∓ C −1
n odd

ij

For n even, the Weyl spinor representations Σ± are self-conjugates, while for n odd, they
are complex conjugates of one another. An analogous result may be derived for odd d.

3
The sign prefactor is a matter of convention since, if C satisfies the above equation, then C γ̄ satisfies
the equation with the opposite sign, γµ∗ = γµt = (C γ̄) γµ ( C γ̄)−1 .

74
6.7 Spinor representations of so(d − 1, 1; R)
When the metric is changed from Euclidean signature (+ + · · · +) to Minkowski signature
(+ · · · + −) with metric ηµν , the structure equation becomes,

[Mµν , Mρσ ] = ηνρ Mµσ − ηνσ Mµρ − ηµρ Mνσ + ηµσ Mνρ (6.42)

The Clifford algebra is defined with respect to the metric η as well and we have,

{Γµ , Γν } = 2ηµν I (6.43)

and the spinor representation matrices take the form,

1
Σµν = [Γµ , Γν ] (6.44)
4
We use capital letters here to represent the γ-matrices to clearly distinguish them from the
case of Euclidean signature. It is straightforward to obtain a representation of the Clifford
algebra matrices Γµ in terms of γµ by setting,

Γd = i γd
Γµ = γµ for µ = 1, · · · , d − 1 (6.45)

For even d, we define the matrix Γ̄ such that Γ̄2 = I. The representation Σ obtained this
way is a spinor representation for the same reasons as with Euclidean signature. For even
d, the Dirac spinor representation is reducible using the matrix γ̄ in terms of Weyl spinor
representations Σ± with half the dimension of 2n .
What changes when the signature is changed is the reality conditions that one can im-
pose on the representations. Complex conjugation may still be defined in an SO(d−1, 1; R)-
invariant way, but there is now a further distinction one can make since transposition is
not related to complex conjugation by Hermiticity. Thus, if Γµ satisfy the Clifford algebra
relations, then so must Γ∗µ and Γtµ , but these representations must be equivalent to Γµ , so
that there must exist invertible matrices C and B such that,4

Γtµ = −C Γµ C −1
Γ∗µ = +B Γµ B −1 (6.46)

To find C, we must pick a basis, such as the one above, and we have

C Γµ C −1 = −Γµ µ odd
−1
C Γµ C = +Γµ µ even (6.47)

4
The sign prefactors are a matter of convention since, if B and C satisfy the above equations, then ΓB
and ΓC satisfy the equation with the opposite sign, Γtµ = +(ΓC)Γµ (ΓC)−1 , and Γ∗µ = −(ΓB)Γµ (ΓB)−1 .

75
For even d = 2n it is easy to find C by considering the product,

C = (Γ̄)n Γ1 Γ3 Γ5 · · · Γ2n−1 (6.48)

The matrix C is then the same matrix as in the Euclidean case and satisfies,

C †C = C tC = I
1
C t = (−1) 2 n(n+1) C
1
C 2 = (−1) 2 n(n+1) I (6.49)

Complex conjugation on the other hand behaves differently, and is most easily gotten as
the composition of Hermitian conjugation and charge conjugation,

Γ∗µ = (Γ†µ )t = −(Γt0 )−1 Γtµ Γ0 = Γ0 CΓµ (Γ0 C)−1 (6.50)

Using the definition of the matrix B and Shur’s lemma, we find that B must be proportional
to CΓ0 . We shall define B as follows,

B = CΓ0 (6.51)

For even d = 2n, the complex conjugation properties of the representation are,

B Σ± −1

µν B n odd
±

∗
Σµν = (6.52)
B Σ∓
µν B
−1
n even


Thus, for n odd, the spinor representations ψ+ and ψ− are self-conjugates, while for n even,
they are complex conjugates of one another.
The spinors ψ ∗ and Bψ and ΓBψ transform under the same representation. Thus, we
may impose the following reality conditions,

ψ ∗ = Bψ (6.53)

which requires the consistency condition B ∗ B = I. In view of B ∗ = B we have,

1
B ∗ B = −(−1) 2 n(n+1) I = I (6.54)

has the solutions n ≡ 1, 2 (mod 4). Alternatively, one may impose the reality condition,

ψ ∗ = ΓBψ (6.55)

which requires that the consistency condition (ΓB)∗ (ΓB) = I be obeyed, or

1
(ΓB)∗ (ΓB) = +(−1) 2 n(n−1) I = I (6.56)

76
has the solutions n ≡ 0, 1 (mod 4).
The Majorana and Weyl conditions may be enforced when we have Majorana spinors,
and [B, Γ] = 0. So, this can happen only when n ≡ 0, 1, 2 (mod 4). To fulfill the remaining
condition, we require,

B = Γ B Γ = Γ Γ0 C Γ = (−1)n+1 B (6.57)

so that only n ≡ 1 (mod 4) i.e. d ≡ 2 (mod 8) survives.

77
7 Roots, weights, and representations
The construction given in the earlier section for the representations of sl(2; C) basically con-
tains all the ingredients for the construction of the representations of all finite-dimensional
representations of an arbitrary semi-simple Lie algebra G over C. The construction for Lie
algebras of rank higher than 1 is considerably more complicated, but may be handled with
great ease using the method of roots and weights due to Cartan.
The adjoint representation of G is such a representation, whose matrices are the struc-
ture constants of G and whose dimension is that of G. The classification of all represen-
tations thus contains the classification of all semi-simple Lie algebras. The power of the
method of roots and weights is illustrated by Cartan’s classification theorem (1920) of all
simple finite-dimensional Lie algebras, and by Dynkin’s complete theory of representations.

7.1 The Cartan-Killing form

We consider a Lie algebra G over C of finite dimension n = dimC G and rank r. In a basis
{X1 , · · · , Xn } of G the structure relations are given by the structure constants fab c ,
n
X
[Xa , Xb ] = fab c Xc (7.1)
c=1

The Cartan-Killing form is the bilinear symmetric form, or metric, γab on G defined by,
n
X
γab = tr (AdXa AdXb ) = fac d fbd c (7.2)
c,d=1

Cartan’s definition of a semi-simple Lie algebra is that its Cartan-Killing form is invertible,
in which case we denote the inverse by γ ab ,
n
X
γab γ bc = δb c (7.3)
b=1

We shall denote the inner product of two vectors va , wb by,

n
X
(v, w) = γ ab va wb (7.4)
a,b=1

One could, of course, take the convention to choose γ proportional to the identity ma-
trix, as is done in Georgi’s book. However, this choice leads to unnecessarily complicated
normalizations of the Cartan generators.
An alternative and equivalent definition of a complex semi-simple Lie algebra is that it
is the complexification of the Lie algebra K of a compact matrix Lie group. This definition,
advocated for example in Brian Hall’s book, is especially convenient to use.

78
Theorem 10 Let G be a semi-simple Lie algebra with Cartan-Killing form γab .

1. The Cartan-Killing form is invariant under the action of G;

2. The combination fabc = d fab d γdc is totally anti-symmetric in its indices;
P

3. For an arbitrary representation ρ of G, we define a symmetric bilinear form by,

tr ρ(Xa )ρ(Xb ) = γ(ρ)ab (7.5)

If G is a simple Lie algebra (i.e. its adjoint representation is irreducible), then for any
representation ρ of G the form γ(ρ) is proportional to the Cartan-Killing form,

γ(ρ)ab = K(ρ)γab (7.6)

where K(ρ) depends only on ρ and equals 1 for the adjoint representation.

To prove these results, we start from the following identity, valid in any representation,
  Xn n
X
d
0 = tr [ρ(Xa ), ρ(Xb )ρ(Xc )] = fab γ(ρ)dc + fac d γ(ρ)db (7.7)
d=1 d=1

This relation proves invariance of γ(ρ)ab in any representation ρ. Taking ρ to be the

adjoint representation shows that fabc , defined above, is anti-symmetric in its last two
indices and is thus anti-symmetric in all indices. To prove 3, we write (7.7) in matrix form
Fa γ(ρ) + (Fa γ(ρ))t = 0 using (Fa )b c = fab c . Eliminating Fat between the relations for an
arbitrary representation ρ and the adjoint representation we find,

Fa γ(ρ)γ −1 = γ(ρ)γ −1 Fa (7.8)

When G is a simple Lie algebra, its adjoint representation is irreducible, and by Shur’s
lemma γ(ρ)γ −1 must be proportional to the identity matrix, thereby proving 3.

7.2 Weights
The Cartan subalgebra H of G is defined to be the maximal Abelian subalgebra of G for
which the adjoint map is diagonalizable. The dimension r of the vector space H over C is
defined to be the rank of G. We denote by H1 , · · · , Hr a choice of basis generators of H
which, by definition, mutually commute,

[Hi , Hj ] = 0 for all i, j = 1, · · · , r (7.9)

Since G is the complexification of the Lie algebra K of a compact Lie group, by the above
definition of a semi-simple Lie algebra, we may choose the Cartan generators to be the Her-
mitian generators of K. Furthermore, since every finite-dimensional representation of K is

79
equivalent to a unitary representation (by theorem 3.6), we may choose the representations
matrices ρ(Hi ) in any finite-dimensional representation of the complexified Lie algebra G
to be Hermitian as well. For sl(2; C) the Cartan generator may be chosen to be J3 . For
sl(3; C) there will be two such generators, which we denoted λ3 , λ8 in Problem set 2.
Consider an arbitrary representation ρ of G, whose dimension N is finite. The repre-
sentation matrices ρ(Hi ) mutually commute,

[ρ(Hi ), ρ(Hj )] = 0 for all i, j = 1, · · · , n (7.10)

and may be chosen to be Hermitian, as argued above. Since they mutually commute and
are Hermitian, they may be diagonalized in the same basis and have real eigenvalues. We
shall label the states of the representation ρ, in the basis where all ρ(Hi ) are diagonal, by
the real eigenvalues µi of the matrices ρ(Hi ),

ρ(Hi )|ρ; µ i = mi |ρ; µ i µ = (m1 , · · · , mn ) (7.11)

The array µ of eigenvalues is referred to as the weight vector of the state |ρ; µ i. We shall
show later that, when ρ is an irreducible representation, all the states are uniquely labelled
by the set of all weight vectors. This result is familiar from quantum mechanics, where one
uses a maximal set of commuting observables to generate all the quantum numbers.

7.3 Roots
Since the generators Hi mutually commute and are Hermitian, we may choose the remaining
generators of G to diagonalize simultaneously the adjoint maps,

[Hi , Eα ] = ai Eα α = (a1 , · · · , an ) (7.12)

where each ai is real, and α is referred to as the root vector. Diagonalizing the adjoint map
generally requires complexifying the Lie algebra. For the case of sl(2; C) there were two
such generators which we denoted J± = J1 ± iJ2 which clearly required introducing i ∈ C.
The generators satisfied [J3 , J± ] = ±J± so in this case the root vectors would be (±1).
The root vectors α are identical to the weight vectors of the adjoint representation.
Denoting the subspace of G generated by Eα by Gα , and the space of all root vectors α by
R, we have the Cartain decomposition of the Lie algebra G as follows,
M
G =H⊕ Gα (7.13)
α ∈R

The result follows from the Hermiticity of the generators Hi of H. In addition, we prove
the following results.

80
Theorem 11 Let G be a semi-simple Lie algebra with Cartan subalgebra H = {H1 , · · · Hr }.
α and we have E−αα = (Eα )† .
1. If α is a root, then so is −α
2. Opposite root generators close onto a sl(2; C) subalgebra as follows,
r
X
[Eα , E−αα ] = Hα α, H) =
Hα = (α γ ij ai Hj
i,j=1

[Hα , E±αα ] = ±(α

α, α )E±αα (7.14)
3. The commutator of two generators associated with roots α , β 6= −α
α vanishes if α + β
is not a root, and gives a generator Eα +ββ if α + β is a root,
[Eα , Eβ ] ∼ Eα +ββ (7.15)
4. The root spaces Gα are one-dimensional and the only root proportional to α is −α
α.
To prove 1. we take the adjoint of [Hi , Eα ] = ai Eα and use the fact that Hi is Hermitian,
and ai is real. It follows that (Eα )† has eigenvalue −ai . To prove 2, we use the Jacobi
identity to evaluate the following commutator,
[Hi , [Eα , E−αα ]] = [Eα , [Hi , E−αα ]] − [E−αα , [Hi , Eα ]]
= −ai [Eα , E−αα ] − ai [E−αα , Eα ] = 0 (7.16)
Hence, [Eα , E−αα ] must be in the Cartan subalgebra of G, and there exist bi such that,
n
X
[Eα , E−αα ] = bi Hi (7.17)
i=1

Taking the inner product with Hi and using the Cartan-Killing form,
  Xr
tr Hi [Eα , E−αα ] = γij bj (7.18)
j=1

But the left hand side may be rearranged as tr([Hi , Eα ]E−αα ) which evaluates to ai tr(Eα E−αα ).
It will be convenient to normalize the generators Eα so that,
tr(Eα E−αα ) = 1 (7.19)
so that bi = j γ ij aj for all i = 1, · · · , r. To prove 3. we consider the commutator [Eα , Eβ ]
P

for α + β 6= 0, and evaluate its commutator with Hi using the Jacobi identity,
[Hi , [Eα , Eβ ]] = [[Hi , Eα ], Eβ ] + [Eα , [Hi , Eβ ]] = (ai + bi )[Eα , Eβ ] (7.20)
If α + β is not a root, then [Eα , Eβ ] has no components along Eγ for any root γ, and it
can have no components along Hj since the left side vanishes, but the right side would not.
Thus if α + β is not a root, then [Eα , Eβ ] = 0. Finally, if α + β is a root, then we have
[Eα , Eβ ] ∼ Eα +ββ , as stated above. Point 4. will be proven later.

81
7.4 Raising and Lowering operators
A generator Eα of G corresponding to a root α modifies the weights of a representation ρ.
To see this, we use the structure relations satisfied by the representation matrices,

[ρ(Hi ), ρ(Eα )] = ai ρ(Eα ) (7.21)

and evaluate the weight vector µ = (m1 , · · · , mr ) of a state ρ(Eα )|ρ; µ i as follows,
 
ρ(Hi )ρ(Eα )|ρ; µ i = [ρ(Hi ), ρ(Eα )] + ρ(Eα )ρ(Hi ) |ρ; µ i
= (ai + mi )Eα |ρ; µ i (7.22)

Since all the states in an irreducible representation ρ were uniquely labelled by their weight
vector, we conclude that we must have,

ρ(Eα )|ρ; µ i = Nα (ρ; µ )|ρ; µ + α i (7.23)

where Nα (ρ; µ ) are normalization constants which may vanish. Since we have,
n
X
[ρ(Eα ), ρ(E−αα )] = ρ(Hα ) = γ ij ai ρ(Hj ) (7.24)
i,j=1

for any representation, we obtain relations between the normalizations Nα (ρ; µ ) by applying
this relation to an arbitrary state |ρ; µ i =
6 0,
n
X
ρ(Eα )ρ(E−αα )|ρ; µ i − ρ(E−αα )ρ(Eα )|ρ; µ i = γ ij ai ρ(Hj )|ρ; µ i (7.25)
i,j=1

we obtain the following fundamental relation,

Nα (ρ; µ − α )N−αα (ρ; µ ) − Nα (ρ; µ )N−αα (ρ; µ + α ) = (α

α, µ ) (7.26)

7.5 Finite-dimensional representations

For a finite-dimensional representation ρ, the vector space generated by applying ρ(Eα )
to any given state transforming under the representation ρ must have finite dimension.
Consider the state |ρ; µ i associated with weight vector µ , and the vector space generated
by applying a single root generator multiple times,
κ
Y
κ
ρ(Eα ) |ρ; µ i = Nα (ρ; µ + (k − 1)α
α) |ρ; µ + κα
αi
k=1
Yκ
ρ(E−αα )κ |ρ; µ i = N−αα (ρ; µ − (` − 1)α
α) |ρ; µ − κα
αi (7.27)
`=1

82
with κ ∈ N. We shall denote by p and q the largest values of κ such that the states satisfy
|ρ; µ + pα
αi =
6 0 and |ρ; µ − qα
αi =
6 0 and,

Eα |ρ; µ + pα
αi = 0 E−αα |ρ; µ − qα
αi = 0 (7.28)

Since the representation ρ has finite dimension, we must have 1 ≤ p, q < ∞. Now we use
α with −q ≤ k ≤ p ,
the fundamental formula (7.26) for all weights µ + kα

Nα (ρ; µ + (k − 1)α
α)N−αα (µ α) − Nα (ρ; µ + kα
µ + kα α)N−αα (µ
µ + (k + 1)α
α) = (α, µ + kα
α) (7.29)

Summing the contributions over this range,

p
X
Nα (ρ; µ + (k − 1)α
α)N−α (µ
µ + kα
α)
k=−q
p p
X X
− α)N−αα (µ
Nα (ρ; µ + kα µ + (k + 1)α
α) = α, µ + kα
(α α) (7.30)
k=−q k=−q

Changing summation variables k → k − 1 in the second sum, we see that all the terms
except at the end points of the summation on the left side cancel one another, and we are
left with,

Nα (ρ; µ − (q + 1)α µ − qα
α)N−αα (µ α)
p
X
−Nα (ρ; µ + pα
α)N−αα (µ
µ + (p + 1)α
α) = α, µ + kα
(α α) (7.31)
k=−q

In view of the definition of p and q we have Nα (ρ; µ + pαα) = 0 and N−αα (ρ; µ − qα
α) = 0, so
that the left side cancels. The vanishing of the right side then gives the equation,
 
1 1
α, µ ) +
(p + q + 1)(α p(p + 1) − q(q + 1) (α
α, α ) = 0 (7.32)
2 2

Dividing through by (p + q + 1) gives the following fundamental formula,

α, µ )
2(α
=q−p (7.33)
α, α )
(α

Since q, p ∈ N, we see that the weights µ lie in a lattice that is dual to the root lattice.
Note that in this ratio of inner products the overall normalization of γab cancels out, and
one may use the Cartan Killing form defined in an arbitrary representation.

83
7.6 The example of SU (3)
The choice Gell-Mann made for the generators of SU (3) is as follows.
     
0 1 0 0 −i 0 1 0 0
1 1 1
T1 = 1 0 0  T2 =  i 0 0  T3 =  0 −1 0 
2 2 2
0 0 0 0 0 0 0 0 0
   
0 0 1 0 0 −i
1 1
T4 = 0 0 0  T5 =  0 0 0 
2 2
1 0 0 i 0 0
     
0 0 0 0 0 0 1 0 0
1 1 1
T6 = 0 0 1  T7 =  0 0 −i  T8 = √  0 1 0 
2 2 2 3
0 1 0 0 i 0 0 0 −2
The matrices Ta form an orthogonal basis under the inner product provided by the
trace, and their normalization has been chosen so that they satisfy,
1
tr(Ta Tb ) = δab (7.34)
2
Thus the inner product (u, v) may be simply replaced by the vector inner product u · v.
Ultimately, we shall prefer the basis in which T8 is replaced by the generator,
√
 
0 0 0
1 3 1
T80 = 0 1 0  = T8 − T3 (7.35)
2 2 2
0 0 −1

at the cost of having tr(T80 T3 ) = − 14 , and thus a non-diagonal metric. We observe the
following subalgebras isomorphic to su(2) or so(3),

S1 = {T1 , T2 , T3 }
S2 = {T4 , T5 , T3 + T80 }
S3 = {T6 , T7 , T80 }
S4 = {T2 , T5 , T7 } (7.36)

The sub-algebras S1 , S2 , S3 are in the two-dimensional representation of su(2), and are

equivalent to one another, but S4 is in the three-dimensional representation of su(2) or in
the defining representation of so(3) and is inequivalent to the first three.
Now we proceed to choosing a Cartan subalgebra, and constructing the roots of su(3)
and the weights of arbitrary representations of su(3). The rank is 2, and clearly we will
choose the Cartan subalgebra generators to be H1 = T3 and H2 = T8 . The states in any
representation ρ will be labelled by the weight vector µ = (m1 , m2 ),

Hi |ρ; µi = mi |ρ; µi i = 1, 2 (7.37)

84
The states of the defining representation 3 in the standard basis of C3 correspond to the
following basis in terms of the weight vectors λ of the defining representation,
     
1 0 0
0 = |3; λ1 i 1 = |3; λ2 i 0 = |3; λ3 i (7.38)
0 0 1
where the weight vectors are given by,
     
1 1 1 1 1
λ1 = , √ λ2 = − , √ λ3 = 0, − √ (7.39)
2 2 3 2 2 3 3
For the adjoint representation, we choose the following linear combinations which diago-
nalize AdH1 and AdH2 ,

T1 ± i T2 = E±α1
T4 ± i T5 = E±α3
T6 ± i T7 = E±α2 (7.40)

where the root vectors αi are given by,

√ ! √ !
1 3 1 3
α1 = (1, 0) α2 = − , α3 = , (7.41)
2 2 2 2

Notice that each root satisfies αi · αi = 1, so that the master formula becomes,

2α · β ∈ Z (7.42)

for any pair of roots. We readily verify that 2α1 · α3 = 2α2 · α3 = −2α1 · α2 = 1. But the
master formula tells us more than that. If µ is a weight, and if µ + pα and µ − qα are the
largest addition and subtraction of µ by the root α, then p and q satisfy,

2α · µ = q − p (7.43)

Taking µ = α2 and α = α1 we have 2α · µ = q − p = −1. This means that p ≥ 1, so α1 + α2

must be a root, and indeed α1 + α2 = α3 . We can choose as a basis of all roots the vectors
α1 and α2 . If 2α1 · µ ∈ Z and 2α2 · µ ∈ Z, then it follows that 2α3 · µ ∈ Z.
Now consider a more general representation ρ of su(3). Its weight vectors must satisfy,

2α1 · µ = p1 ∈ Z 2α2 · µ = p2 ∈ Z (7.44)

Solving for µ in terms of p1 , p2 , we find,

 
p1 p1 + 2p2
µ= , √ (7.45)
2 2 3

85
Comparison with the weights of the fundamental representation, we see that µ is a linear
combination with integer coefficients of the weights of teh defining representation,

µ = (p1 + p2 )λ1 + p2 λ2 (7.46)

This is no accident: the weights of any representation lies in the dual lattice to the root
lattice, of which the weight lattice of the defining representation is a basis. Of course, this
is not the full story since these states must truncate somewhere.

7.7 The root lattice of An = sl(n + 1)

Although the rank of the algebra is n, it will be convenient to consider the root vectors in
Rn+1 , generated by an orthonormal basis of vectors ei for i = 1, · · · , n + 1. Parametrize
the Lie algebra sl(n + 1) by the matrices Mij with components,

(Mi,j )αβ = δi,α δj,β (7.47)

The commutator of two such matrices is given by,

[Mi,j , Mk,` ] = δj,k Mi,` − δi,` Mk,j (7.48)

We shall chose the Cartan generators to be diagonal and traceless,

Hi = Mi,i − Mi+1,i+1 (7.49)

The commutator of Hi with an arbitrary generator Mk,` is given by,

[Hi , Mk,` ] = δi,k − δi,` − δi+1,k + δi+1,` Mk,` (7.50)

Therefore, the root vector associated with the generator Mk,` with k 6= ` is given by,

α = ek − e` Eα = Mk,` (7.51)

The set of all roots is given by,

R = {ek − e` for 1 ≤ k 6= ` ≤ n + 1} #R = n(n + 1) (7.52)

One defines a root α = ek − e` to be a positive root iff k < ` and a root α to be a negative
root iff −α
α is a positive root, which is equivalent to k > `. All positive roots may be
expressed as a linear combinations of the simple roots,

α i = ei − ei+1 i = 1, · · · , n (7.53)

86
with positive integer, or zero, coefficients. To see this, we decompose the positive root
ek − e` for k < ` as follows,
`−1
X
ek − e` = (ei − ei+1 ) (7.54)
i=k

All the coefficients are either zero or 1, and hence all positive or zero, as announced.
The inner products of the simple roots are given as follows,

2 if i=j
(α i , j ) = −1 if |i − j| = 1
α α (7.55)
0 if |i − j| ≥ 2


Using the Dynkin diagran representation introduced earlier, in which roots are represented
by nodes, two roots with inner product −1 are linked by a single link, and two roots which
are orthogonal re not linked, we have the following Dynkin diagram for An = sl(n + 1; C).
···
α1 α2 α3 α n−1 αn

The Cartan matrix is the matrix of inner products of the simple roots, normalized in
such a manner that the entries have to be integers by construction, and are independent
of the overall scaling of the root system,
αi , α j )
2(α
Cij = (7.56)
αi , α i )
(α
Note that the Cartan matrix is symmetric when all the simple roots have the same length,
as is the case for he algebra An = sl(n + 1), but will not be symmetric when simple roots
of different lengths occur in the root space. For sl(n + 1), the Cartan matrix is simply
Cij = (ααi , α j ), and takes the form,
 
2 −1 0 0 ··· 0 0
 −1 2 −1 0 · · · 0 0
 
 0 −1 2 −1 · · · 0 0
C=   (7.57)
 · · · · · · · · · · · · · · · −1 0 

 0 0 0 0 · · · 2 −1
0 0 0 0 · · · −1 2

7.8 The root lattice of Dn = so(2n)

The algebra so(2n) is generated by 2n × 2n anti-symmetric matrices which we denote by
t
Mab = −Mba = −Mab for a, b = 1, · · · , 2n and with components,
(Mab )α,β = δaα δbβ − δaβ δbα (7.58)

87
Their structure relations are given as follows,

[Mab , Mcd ] = δad Mbc + δbc Mad − δbd Mac − δac Mbd (7.59)

These are actually the generators of so(2n; R) which is the Lie algebra of the compact Lie
group SO(2n; R). We now complexify the algebra and choose the following generators of
the Cartain subalgebra,

Hi = i M2i−1,2i i = 1, 2, . . . , n (7.60)
√
where the prefactor i = −1 and not the index. The remaining generators may be or-
ganized into four sets, depending on the parity of each index, and we have the following
commutation relations,

[Hi , M2j−1,2k−1 ] = −iδi,j M2j,2k−1 − iδi,k M2j−1,2k

[Hi , M2j,2k−1 ] = +iδi,j M2j−1,2k−1 − iδi,k M2j,2k
[Hi , M2j−1,2k ] = −iδi,j M2j,2k + iδi,k M2j−1,2k−1
[Hi , M2j,2k ] = +iδi,j M2j−1,2k + iδi,k M2j,2k−1 (7.61)

The roots are identified as follows,

α = η ej + η 0 ek j 6= k (7.62)

where η = ±1 and independently η 0 = ±1, and the associated generators are,

1 0 0

Eα = M2j−1,2k−1 + ηM2j,2k−1 + η M2j−1,2k − ηη M2j,2k (7.63)
2
One verifies that Eα is indeed a simultaneous eigenstate with the following eigenvalues,

[Hi , Eα ] = (ηδi,j + η 0 δi,k )Eα (7.64)

Counting the number of generators, we have n in the Cartan subalgebra and 2n(n − 1) root
generators, adding up to 12 2n(2n − 1) which indeed is the dimension of so(2n). The roots
are usually denoted,

R = {±ej ± ek , 1 ≤ j 6= k ≤ n} (7.65)

where the ± signs are independent of one another. All roots have equal length and satisfy
α, α ) = 2. By the same lexicographical rule that we used for (n), the positive roots are,
(α

R+ = {ej − ek , ej + ek , 1 ≤ j < k ≤ n} (7.66)

88
The simple roots system may be chosen as follows,

α i = ei − ei+1 , i = 1, · · · , n − 1
α n = en−1 + en (7.67)

It is easy to see that every positive root is a linear combination with positive integer
coefficients of simple roots. The inner products of the simple roots are given as follows,


 2 if i=j for 1 ≤ i, j ≤ n
−1 if |i − j| = 1 for 1 ≤ i, j ≤ n − 1

αi , αj ) =
(α (7.68)

 0 if |i − j| ≥ 2 for 1 ≤ i, j ≤ n
0 if i = n − 1, j = n


We find the following Dynkin diagram. Since all roots and all simple roots have the same

αn

···
α1 α2 α n−3 α n−2
α n−1

length, the Cartan matrix is symmetric and given αi , α j ) and is given by a small
by Cij = (α
modification of the one for sl(n + 1; C),
 
2 −1 0 0 ··· 0 0 0
 −1 2 −1 0 · · · 0 0 0 
 
 0 −1 2 −1 · · · 0 0 0 
 
C= · · · · · · · · · · · · · · · ··· ··· ···
 (7.69)
 0
 0 0 0 ··· 2 −1 −1  
 0 0 0 0 ··· −1 2 0 
0 0 0 0 ··· −1 0 2

7.9 The root lattice of Bn = so(2n + 1)

Since so(2n) is a subalgebra of so(2n + 1) of the same rank, we shall choose the Cartan
generators of so(2n+1) to be those chosen for so(2n). The roots ±ej ±ek with 1 ≤ j, k ≤ n
of so(2n) are all roots of so(2n + 1) as well. But there are more roots, which are given by,

[Hi , Eα ] = η δi,j Eα α = η ej η = ±1 (7.70)

with associated generators,

Eα = M2j−1,2n+1 + ηM2j,2n+1 (7.71)

89
Hence the system of all roots is given by,
R = {±ej ± ek , 1 ≤ j 6= k ≤ n and ± ej , 1 ≤ j ≤ n} (7.72)
where the ± signs are independent of one another. By the same lexicographical rule that
we used for sl(n), the positive roots are,
R+ = {ej − ek , 1 ≤ j < k ≤ n and ej , 1 ≤ j ≤ n} (7.73)
The simple roots system may be chosen as follows,
α i = ei − ei+1 , i = 1, · · · , n − 1
α n = en (7.74)
We encounter here, for the first time, roots of different lengths, and simple roots of different
lengths. The Cartan matrix is not now symmetric. For example, in the case of n = 3,
namely so(7), we have,
 
2 −1 0
C = −1 2 −1 (7.75)
0 −2 2
The Dynkin diagram is given by
···
α1 α2 α3 α n−1 αn

7.10 The root lattice of Cn = sp(2n + 1)

Without derivation we shall give here the root lattice of sp(2n),
R = {±ej ± ek , 1 ≤ j 6= k ≤ n and ± 2ej , 1 ≤ j ≤ n} (7.76)
where the ± signs are independent of one another. By the same lexicographical rule that
we used for sl(n + 1; C), the positive roots are,
R+ = {ej − ek , 1 ≤ j < k ≤ n and 2ej , 1 ≤ j ≤ n} (7.77)
The simple roots system may be chosen as follows,
α i = ei − ei+1 , i = 1, · · · , n − 1
α n = 2en (7.78)
The Dynkin diagram is given by Superficially, this is looks like the same Dynkin diagram
as we had for so(2n + 1), but the key difference is that the role of the long and short roots
has been exchanged. Sometimes this is indicated by an arrow pointing from the short to
the long roots.

90
 ···
α1 α2 α3 α n−1 αn

Coincidences of Lie algebras of low rank

From looking at the Dynkin diagrams of the classical Lie algebras, it is immediately manifest
that we have the following coincidences,

so(3; C) = sl(2; C) = sp(2; C)

so(4; C) = sl(2; C) ⊕ sl(2; C)
so(5; C) = sp(4; C)
so(6; C) = sl(4; C) (7.79)

and that there are no other coincidences.

91
8 Representations of the classical Lie algebras
The classical Lie algebras are An = sl(n + 1; C), Bn = so(2n + 1; C), Cn = sp(2n; C), and
Dn = so(2n; C). Their root systems have been constructed in the preceding section. In the
present section, we shall discuss the construction of their finite-dimensional representations,
and give details for the case of sl(n + 1; C) and so(2n; C) only.
The key tool is the following result obtained earlier. Consider a semi-simple Lie algebra
G of rank n and a finite-dimensional irreducible representation ρ of G. If µ is a weight of ρ
and α a root of G, and µ + kα α is a weight of ρ for −q ≤ k ≤ p with p, q ≥ 1, but µ + kαα is
not a weight of ρ for k = p + 1 and k = −q − 1 then p, q satisfy the following magic formula,

α, µ )
2(α
q−p= (8.1)
α, α )
(α

As established earlier, the weights of all finite-dimensional representations lie on the weight
lattice which consists of all vectors µ such that 2(α α, α ) ∈ Z for all roots α of G.
α, µ )/(α
A basis of the weight lattice is given by the fundamental weights λ i for i = 1, · · · n,
defined as duals to the simple roots α j by,

αj , λ i )
2(α
= δij (8.2)
αj , α j )
(α

The weights of every finite-dimensional representation ρ span a finite sub-lattice of the

lattice of linear combinations with integer coefficients of the fundamental weights λ i .

8.1 Weyl reflections and the Weyl group

The system of root and weight vectors is highly symmetric, a property which is reflected
in the fact that it is invariant under the Weyl group W (G) generated by Weyl reflections.
To quantify Weyl reflections, we start from a weight µ of the Lie algebra G in a finite-
dimensional irreducible representation ρ and use the magic formula in the following way,

α, µ )
2(α α, µ )
2(α
If ≥0 then q≥
α, α )
(α α, α )
(α
α, µ )
2(α α, µ )
2(α
If ≤0 then p≥− (8.3)
α, α )
(α α, α )
(α

When 2(α α, α) is negative, we can always add −2(α

α, µ)/(α α, µ)/(α
α, α) times α to the weight
0
µ to obtain a new weight µ ,

α, µ )
2(α
µ0 = µ − α (8.4)
α, α )
(α

92
When 2(α α, µ )/(α
α, α ) is positive, we can always subtract 2(α
α, µ )/(α
α, α ) times α to the weight
µ to obtain a new weight which turns out to be given by the same formula as µ 0 . Defining
the following linear operation Πα (µ µ) on the weight space,
α, µ )
2(α
µ) = µ −
Πα (µ α (8.5)
α, α )
(α
we have just shown that it will take a weight to another weight of the representation ρ.
The operation Πα reverses the sign of the root vector α ,
α) = −α
Πα (α α (8.6)
and leaves invariant any vector which is orthogonal to α . Thus, geometrically, Πα reflects
the vector µ in the plane orthogonal to α , and is referred to as a Weyl reflection. The
reflection property is confirmed by showing that the square of Πα is the identity,
  α, µ )
2(α
µ) = Πα (µ
Πα Πα (µ µ) − α) = µ
Πα (α (8.7)
α, α )
(α
The set of Weyl reflections for all roots generates a group, referred to as the Weyl group
W (G) which is a subgroup of the group of all permutations of all the roots. The Weyl group
provides an invaluable tool for constructing the full weight diagram of a representation,
given its highest weight.

8.2 Finite-dimensional irreducible representations of sl(n + 1; C)

Using the simple roots of sl(n + 1; C),
α i = ei − ei+1 i = 1, · · · , n (8.8)
αi , α )i) = 2, the fundamental weights λ j are defined to satisfy,
which all have length (α
(ei − ei+1 ) · λ j = δij 1 ≤ i, j ≤ n (8.9)
The solution is unique and given as follows,
j
X
λj = e` (8.10)
`=1

An arbitrary representation is specified by its highest weight vector,

n
X
λ= qiλ i (8.11)
i=1

The numbers qi correspond to the lengths of the horizontal arrays of boxes in the Young
tableau. For example, taking the case of sl(5; C) with n = 4, the values of qi parametrize
the Young tableau as follows.

93
 q1 q2 q3 q4

8.3 Spinor representation of so(2n + 1; C)

The simple roots of so(2n + 1; C) are given by,

α i = ei − ei+1 i = 1, · · · , n − 1
α n = en (8.12)

The fundamental weights λ j for j = 1, · · · , n − 1 obey identically the same equations as

the first n − 1 fundamental weights of sl(n + 1; C) and are thus given by,
j
X
λj = e` j = 1, · · · , n − 1 (8.13)
`=1

But the fundamental weight λ n satisfies a different equation than for sl(n + 1; C) in view
of the fact that α n is a sorter rot, and we have, 2(α αn , λ n ) = 1 while (αi , λ n ) = 0 for
i = 1, · · · , n − 1. The solution is uniquely given by,
n
1X
λn = e` (8.14)
2 `=1

All the weights of the representation with highest weight λ n may be constructed by Weyl
reflections in the roots ei , and we get,
1
(±e1 ± e2 ± · · · ± en ) (8.15)
2
where the ± assignments are independent of one another. Therefore, the representation has
dimension 2n , and is referred to as the fundamental spinor representation of so(2n + 1; C).

8.4 Spinor representations of so(2n; C)

The simple roots of so(2n; C) are given by,

α i = ei − ei+1 i = 1, · · · , n − 1
α n = en−1 + en (8.16)

94
The fundamental weights λ j for j = 1, · · · , n − 2 obey identically the same equations as
the first n − 2 fundamental weights of sl(n + 1; C) and are thus given by,
j
X
λj = e` j = 1, · · · , n − 2 (8.17)
`=1

But the fundamental weights λ n−1 , λ n satisfy,

λn−1 , α i ) = (λ
(λ λn , α i ) = 0 i = 1, · · · , n − 2
λn−1 , α n−1 ) = (λ
(λ λn , α n ) = 1
λn−1 , α n ) = (λ
(λ λn , α n−1 ) = 0 (8.18)
Solving these equations, we find,
1
λ n−1 = (e1 + · · · + en−2 + en−1 − en )
2
1
λn = (e1 + · · · + en−2 + en−1 + en ) (8.19)
2
αn , λ n ) = 1 while (αi , λ n ) = 0 for i = 1, · · · , n − 1. The solution is uniquely given by,
2(α
n
1X
λn = e` (8.20)
2 `=1

All the weights of the representation with highest weight λ n may be constructed by Weyl
reflections in the roots, and we get,
1
λ n−1 : (±e1 ± e2 ± · · · ± en ) odd number of minus signs
2
1
λn : (±e1 ± e2 ± · · · ± en ) even number of minus signs (8.21)
2
where the ± assignments are independent of one another, except for the overall number of
minus signs as indicated above. Therefore, we find two independent fundamental spinor
representations, each one having dimension dimension 2n−1 .
A general finite-dimensional irreducible representation ρ of so(2n; C) is given by its
highest weight vector λ , which is a linear combination of the n fundamental weight vectors
λ i with integer coefficients,
n
X
λ= qiλ i qi ∈ Z, qi ≥ 0 (8.22)
i=1

When qn−1 + qn is even, then the representation is single-valued and referred to as a tensor
representation, while when qn−1 + qn is odd, the representation is double-valued and is
referred to as a spinor representation.

95
9 The structure of semi-simple Lie algebras
A semi-simple Lie algebra is specified by its adjoint representation. Thus, in the magic
formula, we take the weight vectors µ to be the roots, and find that the root lattice must
obey the following conditions,
α, β )
2(α α, β )
2(α
M= ∈Z M0 = ∈Z (9.1)
α, α )
(α β, β)
(β
which implies that we also must have, for all roots α, β,
α , β )2
4(α
MM0 = ∈Z (9.2)
α, α ) (β
(α β, β)
This combination is invariant under overall rescaling of the root lattice, and in fact is given
by the angle θ between the root vectors α , β ,

M M 0 = 4 cos2 θ ∈ Z (9.3)

This condition is extremely restrictive, and allows for

MM0 θ Dynkin
0 90o
1 60o , 120o
2 45o , 135o
3 30o , 150o
4 0o , 180o same root

Clearly these conditions provide very strong restrictions on the structure of semi-simple
Lie algebras of finite dimension, and in fact lead to their complete Cartan classification,
which we shall now carry out.

9.1 Some properties of simple roots

One defines a root vector to be positive as one defines in general a weight vector to be
positive. Let the root and weight vectors be given in an ordered orthonormal basis ei for
i = 1, · · · , n. A root or weight µ is a positive root iff its first non-zero component is positive.
A root or weight µ is a negative root if −µ µ is a positive root. The only null root or weight
is µ = 0. With this lexicographical partial ordering, every root or weight vector is either
positive, negative or null.

96
 One defines a simple root as a positive root which cannot be written as the sum of
two positive roots. The entire root system is determined by giving the simple roots. Some
important properties of simple roots are as follows.

1. A semi-simple Lie algebra of rank n has n linearly independent simple roots, which
we shall denote by α i for i = 1, · · · , n.
2. If α , β are simple roots, then α − β is not a root. Obviously, this is true when α = β ,
so consider α 6= β . Assume that α − β 6= 0 is a root. Then ether α − β is positive or
β − α is positive. Consider the first and write α = (α α − β ) + β , but then α would be
the sum of two positive roots, which contradicts the fact that α is simple.
3. As a result, the magic formula tells us that if α , β are two simple roots, we have,

α, β )
2(α
=q−p (9.4)
α, α )
(α

but we must have q = 0 since otherwise β − α would be a root, which contradicts

the item above. Hence we must have (α, β ) ≤ 0. Thus the angle θ between different
simple roots must obey π2 ≤ θ < π. The case θ = π is excluded since it would
correspond to opposite roots which cannot both be positive.
4. The n simple roots determine all the roots of the algebra, either by using the magic
formula, or by using the Weyl group.
5. Every root α can be uniquely decomposed as a sum of simple roots:
n
X
α= ni α i (9.5)
i=1

where either all ni ≥ 0 for a positive root α , or all ni ≤ 0 for a negative root α .
6. All the information on the simple roots is provided by the Cartan matrix or equiva-
lently by the Dynkin diagram.

9.2 Classification of finite-dimensional simple Lie algebras

Dynkin defines a Π-system to be a set of simple roots which satisfy three conditions.

(A) There are n linearly independent simple root vectors α i for i = 1, · · · , n.

(B) Any two distinct simple roots α i , α j are such that,

αi , α j )
2(α
= −pij pij ≥ 0, pij ∈ Z (9.6)
αi , α i )
(α

97
(C) The simple root system is indecomposable, i.e. it cannot be split into two sets of
simple roots that are mutually orthogonal to one another. A decomposable simple
root system corresponds to a Lie algebra which is the direct sum of two Lie algebras
and hence is not a simple Lie algebra.
We shall now enumerate all possible Π-systems. The classification theorem is simply done
with the help of a few Lemmas.
Lemma 2 The only Π system with 3 vectors are

α1 α2 α3 α1 α2 α3

The only angles allowed between two distinct simple roots are 90o , 120o , 135o , and
150o . The sum of the angles between three linearly independent vectors must be < 360o .
This means that one and only one angle must be 90o , since otherwise the system would be
decomposable. Suppose θ13 = 90o as in the diagrams above. The remaining angles must
satisfy θ12 + θ23 < 270o . This requires one angle to be θ12 = 120o and the remaining angle
θ23 is allowed to be either 120o or 135o , with the two cases depicted in the above figures.
Lemma 3 An indecomposable subsystem of a Π-system is a Π-system.
This is obviously true.
Corollary 2 The only Π-system with a triple link is given by,

This is an immediate consequence of Lemma 3.

Lemma 4 If a Π-system contains two vectors connected by a single line, the diagram
obtained by shrinking the line away and merging the two vectors into a single circle is
another Π-system.
To prove this lemma, let α and β be the two vectors joined by a single line, and let Γ be
the set of all other vectors. It follows from Lemma 2 that Γ contains no vector connected
to both α and β . The vector α + β has the same length as α and β . This follows from
applying the magic formula,
α, β )
2(α α, β )
2(α
= = −1 (9.7)
α, α )
(α β, β)
(β
Now if γ ∈ Γ and connected to α , then it cannot be connected to β and hence we have
(γγ , α + β ) = (γγ , α ), while if γ 0 ∈ Γ and connected to β , then it cannot be connected to α
and hence we have (γγ , α + β ) = (γγ , β ). Hence the set {α α + β } ∪ Γ is a Π-system.

98
Corollary 3 No Π-system contains more than one double link.
To prove this, consider a diagram with two double links and shrink all the single lines
in between the two double links, to end up with a system with two consecutive double links
which contradicts Lemma 2 with the help of Lemma 3.

Corollary 4 No Π-system contains a closed loop.

Again, shrink all single lines until you get a closed loop diagram with three vectors that
contradicts Lemma 2.

Lemma 5 If the configuration on the left is a Π-system, then the configuration on the
right must also be a Π-system.

γ γ
α +β
β

To prove this, we use the fact that (αα, β ) = 0 and the fact that γ , α , β have the same
α + β , α + β ) = 2(α
length. It then follows that also (α α, α ), and that its inner product with
γ is given by,

2(γγ , α + β ) 2(γγ , α + β )
= −2 = −1 (9.8)
(γγ , γ ) α + β , αβ )
(α

which indeed gives the Π-systen in the right panel of the figure.

Corollary 5 The only branches in a π-system can be tri-valent with single links emanating
from the vertex.

Lemma 6 The following four diagrams do not correspond to Π-systems.

1 2 3 4 3 2 1

1 2 3 2 1

99
 3

1 2 3 4 5 6 4 2

2 1

1 2 3 2 1

The numbers inside the root circles are the coefficients needed to construct a linear
combination that vanishes. For the last diagram, for example, we have,

α1 + 2α
(α α2 + 3α
α3 + 2α α 5 + α 6 + α 7 )2 = 0
α4 + 2α (9.9)

hence the root vectors are linearly-dependent.

As a result of all the lemmas we have,

1. A triple link can occur only by itself, as is reflected in the algebra G2 of Theorem 9.2.

2. A double link can occur either at the end of a chain of simple links as in the algebras
Bn and Cn of Theorem 9.2, or with just a single length on each side of the double
link as in the algebra F4 of Theorem 9.2.

3. The remaining diagrams which have only single links can have at most one trivalent
vertex since otherwise the single links that connected the two vertices could be shrunk
to a quadrivalent vertex. This leaves chains without vertices as in the algebras An of
Theorem 9.2, or if there is one vertex then one of the links emanating from the vertex
must connected to a single root. With the diagrams that are eliminated by Lemma
9.2, this leaves the algebras Dn , E8 , E7 , and E6 of Theorem 9.2.

100
Theorem 12 All finite-dimensional complex simple Lie algebras are given by

An ··· dimC An = n2 − 1

Bn ··· dimC Bn = n(2n + 1)

Cn ··· dimC Cn = n(2n + 1)

Dn ··· dimC Dn = n(2n − 1)

E8 dimC E8 = 248

E7 dimC E7 = 133

E6 dimC E6 = 78

F4 dimC F4 = 52

G2 dimC G2 = 14

The short roots have been indicated in blue. The root systems for the exceptional
groups are as follows.
 1
RE8 = ± ei ± ej , 1 ≤ i < j ≤ 8; (±e1 ± e2 ± e3 ± e4 ± e5 ± e6 ± e7 ± e8 )
2
 1
RE7 = ± ei ± ej , 1 ≤ i < j ≤ 6; ±(e7 − e8 ); ± (±e1 ± e2 ± e3 ± e4 ± e5 ± e6 − e7 + e8 )
2
 1
RE6 = ± ei ± ej , 1 ≤ i < j ≤ 5; ± (±e1 ± e2 ± e3 ± e4 ± e5 − e6 − e7 + e8 )
2
 1
RF4 = ± ei ± ej , 1 ≤ i < j ≤ 4; ± ei ; ± (±e1 ± e2 ± e3 ± e4 )
 2
RG2 = ± (ei − ej , 1 ≤ i < j ≤ 3; (2ei − ej − ek ), {i, j, k} = {1, 2, 3} (9.10)

101
10 Weyl’s character formulas
Consider a compact matrix group G of rank n with Lie algebra G. Two of the most
important problems in group theory are as follows.

1. Decomposition of the tensor product of two irreducible representations R1 , R2 of G,

M
R1 ⊗ R2 = mR (R1 , R2 ) R (10.1)
R

where the sum is over all possible irreducible representations R of G. The positive
integer mR (R1 , R2 ) is the multiplicity with which the representation R occurs in the
tensor product R1 ⊗ R2 .
2. If R is an irreducible representation of the group G, and we have a subgroup G0
of G, then we want to know how R decomposes into a direct sum of irreducible
representations r of the subgroup G0 ,
M
R= mr (R)r (10.2)
r

where the sum is over all irreducible representations r of G0 and the positive integers
mr (R) is the multiplicity with which the representation r occurs in the decomposition
of R with respect to G0 . This decomposition is sometimes referred to as the branching
rules for the subgroup G0 .

Both questions could already have been answered by use of the method of weights or tensors,
but the most natural approach is the method of group characters. Recall the definition of
the character χR (g) of a representation R of G,

χR (g) = trR(g) χR (e) = dim R (10.3)

Recall that characters behave well under direct sum and tensor product of representations,

χR1 ⊕R2 (g) = χR1 (g) + χR2 (g)

χR1 ⊗R2 (g) = χR1 (g) × χR2 (g) (10.4)

for all g ∈ G. Furthermore, characters are functions only of the conjugacy class of g,

χR (h g h−1 ) = χR (g) for all g, h ∈ G (10.5)

The method of characters provides a very convenient solution to both the decomposition
and the branching rule problem.

102
10.1 Characters on the maximal torus
By conjugation, we may diagonalize g to a diagonal matrix of phases,
n
X
iH(ϕ)
g=e H(ϕ) = ϕi Hi (10.6)
i=1

where as earlier, we have chosen the Cartan generators Hi to be Hermitian. Since G is

compact, every finite-dimensional represnetation R of G is equivalent to a unitary repre-
sentation. Thus, by the same token we may diagonalize R(g) in any representation ρ of
the Lie algebra G of G, and choose the representation matrices ρ(Hi ) again Hermitian,
n
X
iρ(H(ϕ))
R(g) = e ρ(H(ϕ)) = ϕi ρ(Hi ) (10.7)
i=1

Then in general a character is of the form

X
χR (g) = nλ ei(λλ,H(ϕ)) (10.8)
λ

nλ is the multiplicity for weight λ in the weight diagram of representation ρ.

Now any unitary matrix may be diagonalized by a unitary transformation. For simplic-
ity, we shall assume G = U (n), but the derivation can be easily generalized to SU (n) or
any of the other compact Lie groups. Therefore we may decompose g as follows,
 iϕ 
e 1 0 0 ··· 0
 0 eiϕ2 0 · · · 0 
+
 iϕ3

g = U DU D=  0 0 e · · · 0 
 (10.9)
· · · · · · · · · · · · · · · 
0 0 0 · · · eiϕn

The character in a representation R is then defined by,

 
eiϕ1 0
χR (g) = χR 
 ... 
 (10.10)
iϕn
0 e

Since characters are class functions, they only depend on the diagonal matrix D and not on
U . This is a tremendous simplification. We have at our disposal the orthogonality relations
for characters of irreducible representations (primitive characters), given for G = U (n) by,
Z
dµ(g) χ?R1 (g)χR2 (g) = N δR1 ,R2 (10.11)
G

103
 Now we would like to express χR (g) in terms of polynomials in eiϕi . The problem is that
the measure dµ(g) depends on D but also on the other variables U . There is a standard
change of variables that yields us the Jacobian for the change of variables

g → (U, ϕ1 , · · · , ϕn ) (10.12)

To do so, regard dµ(g) as coming from the invariant metric on G = U (n),

 
ds2 = −tr g † dg g † dg (10.13)

To decompose the mesure dµ(g) it suffices to decompose the metric, using,

g † dg = −U † D? (dU ) U † DU + U † D? (dD)U + U † (dU ) (10.14)

In evaluating the metric, the cross terms of dD and dU cancel, and one is left with,
 
2 ? 2 † 2 ? † †
ds = −tr(D dD) − 2 tr(U dU ) + 2 tr D (dU U )D(dU U ) (10.15)

The metric is the sum of a metric on the variables ϕi which is independent of U , and a
metric on U which depends on ϕi . Therefore, the measure factorizes and is of the form,
n
Y
?
dµ(g) = ∆(ϕ)∆ (ϕ) dϕi dµ(U ) (10.16)
i=1

Each factor dU U † is accompanied by a linear function in one of the phases eiϕi and the
dependence on positive and negative phases factroizes. The measure dµ(U ) contains n2 − n
factors, so that ∆ must be a polynomial in eiϕ1 , · · · , eiϕn of degree 21 n(n − 1).
∆(ϕ) must vanish whenever ϕi = ϕj for i 6= j. This tells us what ∆ should be up to a
ϕ-independent constant,
Y
∆(ϕi ) = (eiϕi − eiϕj ) (10.17)
i<j

which is of degree 21 n(n − 1) and is the only function with those properties.

10.2 Weyl’s first formula

The character of any irreducible representation of U (n) has the following form,

XR (γi )
χR (g) = γi = eiϕi (10.18)
∆(γi )

104
 1 1 ... 1
γ1 γ2 γn
γ12 γ22 . . . γn2
Y
∆(γi ) = (γi − γj ) = (10.19)
i<j
..
.
γ n−1 . . . γnn−1
and χR (γi ) is an anti-symmetric polynomial in γi ,
γ1`1 γ2`1 . . . γn`1
..
γ1`2 γ2`2 .
XT (γi ) = .. (10.20)
··· ··· .
γ1`n γ2`n
γn`n
Here
`k = mk + (n − k) k = 1, . . . , n (10.21)
with
k
X
mk = qi (10.22)
i=1

and the q’s are the components as follows:

X
µ= qiλ i (10.23)
i

where λ i are the weights of the fundamental representations:

λj , α i )
2(λ
= δij (10.24)
αi , αi )
(α
By letting ϕi → 0, and taking the limit, one deduces a formula for dimensions.
Y
(`i − `j )
1≤i<j≤n
dim R = Y (10.25)
i!
1≤i≤n−1

where we may express `i − `j directly in terms of The q-parameters by,

j
X
`i − `j = j − i + qk (10.26)
k=i+1

For SU (3), we set n = 3 and we find,

1
dim R = (1 + q2 )(1 + q3 )(2 + q2 + q3 ) (10.27)
2
in agreement with the formula derived from tensor analysis and Young tableaux.

105