MAT4144 5158 LieGroups

Introduction to Lie Groups
MAT 4144/5158
Winter 2015
Alistair Savage
Department of Mathematics and Statistics
University of Ottawa
This work is licensed under a

Creative Commons Attribution-ShareAlike 4.0 International License
Contents
Preface iii
1 Introduction and first examples 1

1.1 Category theoretic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The circle: S1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Matrix representations of complex numbers . . . . . . . . . . . . . . . . . . . 5
1.4 Quaternions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Quaternions and space rotations . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.6 Isometries of Rn and reflections . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.7 Quaternions and rotations of R4 . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.8 SU(2) × SU(2) and SO(4) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2 Matrix Lie groups 20

2.1 Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Finite groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 The general and special linear groups . . . . . . . . . . . . . . . . . . . . . . 21
2.4 The orthogonal and special orthogonal groups . . . . . . . . . . . . . . . . . 21
2.5 The unitary and special unitary groups . . . . . . . . . . . . . . . . . . . . . 22
2.6 The complex orthogonal groups . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.7 The symplectic groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.8 The Heisenberg group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.9 The groups R∗ , C∗ , S1 and Rn . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.10 The Euclidean group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.11 Homomorphisms of Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3 Topology of Lie groups 29

3.1 Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Polar Decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 Compactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4 Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4 Maximal tori and centres 39

4.1 Maximal tori . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Centres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.3 Discrete subgroups and the centre . . . . . . . . . . . . . . . . . . . . . . . . 45
i
ii CONTENTS
5 Lie algebras and the exponential map 47

5.1 The exponential map onto SO(2) . . . . . . . . . . . . . . . . . . . . . . . . 47
5.2 The exponential map onto SU(2) . . . . . . . . . . . . . . . . . . . . . . . . 48
5.3 The tangent space of SU(2) . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.4 The Lie algebra su(2) of SU(2) . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.5 The exponential of square matrices . . . . . . . . . . . . . . . . . . . . . . . 52
5.6 The tangent space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.7 The tangent space as a Lie algebra . . . . . . . . . . . . . . . . . . . . . . . 59
5.8 Complex Lie groups and complexification . . . . . . . . . . . . . . . . . . . . 61
5.9 The matrix logarithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.10 The exponential mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.11 The logarithm into the tangent space . . . . . . . . . . . . . . . . . . . . . . 67
5.12 Some more properties of the exponential . . . . . . . . . . . . . . . . . . . . 71
5.13 The functor from Lie groups to Lie algebras . . . . . . . . . . . . . . . . . . 73
5.14 Lie algebras and normal subgroups . . . . . . . . . . . . . . . . . . . . . . . 78
5.15 The Campbell–Baker–Hausdorff Theorem . . . . . . . . . . . . . . . . . . . . 79
6 Covering groups 85
6.1 Simple connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
6.2 Simply connected Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
6.3 Three Lie groups with tangent space R . . . . . . . . . . . . . . . . . . . . . 89
6.4 Three Lie groups with the cross-product Lie algebra . . . . . . . . . . . . . . 90
6.5 The Lie group-Lie algebra correspondence . . . . . . . . . . . . . . . . . . . 91
6.6 Covering groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.7 Subgroups and subalgebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
7 Further directions of study 101

7.1 Lie groups, Lie algebras, and quantum groups . . . . . . . . . . . . . . . . . 101
7.2 Superalgebras and supergroups . . . . . . . . . . . . . . . . . . . . . . . . . 101
7.3 Algebraic groups and group schemes . . . . . . . . . . . . . . . . . . . . . . 102
7.4 Finite simple groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
7.5 Representation theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Preface
These are notes for the course Introduction to Lie Groups (cross-listed as MAT 4144 and
MAT 5158) at the University of Ottawa. At the title suggests, this is a first course in the
theory of Lie groups. Students are expected to a have an undergraduate level background
in group theory, ring theory and analysis. We focus on the so-called matrix Lie groups since
this allows us to cover the most common examples of Lie groups in the most direct manner
and with the minimum amount of background knowledge. We mention the more general
concept of a general Lie group, but do not spend much time working in this generality.
After some motivating examples involving quaternions, rotations and reflections, we give
the definition of a matrix Lie group and discuss the most well-studied examples, including
the classical Lie groups. We then study the topology of Lie groups, their maximal tori,
and their centres. In the second half of the course, we turn our attention to the connection
between Lie algebras and Lie groups. We conclude with a discussion of simply connected
Lie groups and covering groups.
Acknowledgement: The author would like to thank the students of MAT 4144/5158 for
making this such an enjoyable course to teach, for asking great questions, and for pointing
out typographical errors in the notes.
Alistair Savage Ottawa, 2015.
Course website: http://alistairsavage.ca/mat4144/
iii
Chapter 1
Introduction and first examples
In this chapter we introduce the concept of a Lie group and then discuss some important
basic examples.
1.1 Category theoretic definitions

We will motivate the definition of a Lie group in category theoretic language. Although we
will not use this language in the rest of the course, it provides a nice viewpoint that makes
the analogy with usual groups precise.
Definition 1.1.1 (Category). A category C consists of
• a class of objects ob C,
• for each two objects A, B ∈ ob C, a class hom(A, B) of morphisms between them,
• for every three objects A, B, C ∈ ob C, a binary operation
hom(B, C) × hom(A, B) → hom(A, C)
called composition and written (f, g) 7→ f ◦ g
such that the following axioms hold:
• Associativity. If f ∈ hom(A, B), g ∈ hom(B, C) and h ∈ hom(C, D), then h ◦ (g ◦ f ) =

(h ◦ g) ◦ f .
• Identity. For every object X, there exists a morphism 1X ∈ hom(X, X) such that for
every morphism f ∈ hom(A, B) we have 1B ◦ f = f = f ◦ 1A .
Definition 1.1.2 (Terminal object). An object T of a category C is a terminal object if there

exists a single morphism X → T for every X ∈ ob C.
Examples 1.1.3.
1
2 CHAPTER 1. INTRODUCTION AND FIRST EXAMPLES
Objects Morphisms Terminal object(s) Product

Sets Set maps Singletons Cartesian product
Vector spaces Linear maps 0 Tensor product
Cartesian product
Topological spaces Continuous maps Single point
(product topology)
Cartesian product (induced
Smooth manifolds Smooth maps Single point
manifold structure)
Algebraic varieties Algebraic maps Single point Product variety
Recall that a group is a set G together with a maps of sets G × G → G, often called
multiplication and denoted (g, h) 7→ g · h, satisfying the following properties:
• Associativity. For all a, b, c ∈ G, we have (a · b) · c = a · (b · c).
• Identity element. ∃ e ∈ G such that e · a = a · e = a for all a ∈ G.
• Inverse element. ∀ a ∈ G ∃ b ∈ G such that a · b = b · a = e. One can show that the
element b is unique and we call it the inverse of a and denote it a−1 .
Note that G is a set and the multiplication is a map of sets. We can generalize this
definition to (almost) any other category.
Definition 1.1.4 (Group object). Suppose we have a category C with finite products and a
terminal object 1. Then a group object of C is an object G ∈ ob C together with morphisms
• m : G × G → G (thought of as the “group multiplication”),
• e : 1 → G (thought of as the “inclusion of the identity element”),
• ι : G → G (thought of as the “inversion operator”),
such that the following properties are satisfied:
• m is associative: m ◦ (m × 1G ) = m ◦ (1G × m) as morphisms G × G × G → G.
• e is a two-sided unit of m:
m ◦ (1G × e) = p1 ,
m ◦ (e × 1G ) = p2 ,
where p1 : G × 1 → G and p2 : 1 × G → G are the canonical projections.

• ι is a two-sided inverse for m: if d : G → G × G is the diagonal map, and eG : G → G
is the composition
e
G→1→ − G,
then m ◦ (1G × ι) ◦ d = eG and m ◦ (ι × 1G ) ◦ d = eG .
We can now generalize the definition of group by considering group objects in other
categories.
1.2. THE CIRCLE: S1 3
Category Group objects

Sets Groups
Topological spaces Topological groups
Smooth manifolds Lie groups
Algebraic varieties Algebraic groups
So a Lie group is just a group object in the category of smooth manifolds. It is a group
which is also a finite-dimensional (real) smooth manifold, and in which the group operations
of multiplication and inversion are smooth maps. Roughly, Lie groups are “continuous
groups”.
Examples 1.1.5. (a) Euclidean space Rn with vector addition as the group operation.
(b) The circle group S1 (complex numbers with absolute value 1) with multiplication as the
group operation.
(c) General linear group GL(n, R) with matrix multiplication.
(d) Special linear group SL(n, R) with matrix multiplication.
(e) Orthogonal group O(n, R) and special orthogonal group SO(n, R).
(f) Unitary group U(n) and special unitary group SU(n).
(g) Physics: Lorentz group, Poincaré group, Heisenberg group, gauge group of the Standard
Model.
Many of the above examples are linear groups or matrix Lie groups (subgroups of some
GL(n, R)). In this course, we will focuss on linear groups instead of the more abstract full
setting of Lie groups.
Exercises.
1.1.1. Show that the notions of group and group object in the category of sets are equivalent.
1.2 The circle: S1

Consider the plane R2 . If we use column vector notation for points of R2 , then rotation about
the origin through an angle θ is a linear transformation corresponding to (left multiplication
by) the matrix
cos θ − sin θ
Rθ = .
sin θ cos θ
This rotation corresponds to the map

x x x cos θ − y sin θ
7→ Rθ = .
y y x sin θ + y cos θ
Rotating first by θ and then by ϕ corresponds to multiplication by Rθ and then by Rϕ or,

equivalently, by Rϕ Rθ . So composition of rotation corresponds to multiplication of matrices.
Definition 1.2.1 (SO(2)). The group {Rθ | θ ∈ R} is called the special orthogonal group
SO(2). The term special refers to the fact that their determinant is one (check!) and the
term orthogonal refers to the fact that they do not change distances (or that Rθ RθT = 1 for
any θ) – we will come back to this issue later.
Another way of viewing the plane is as the set of complex numbers C. Then the point
(x, y) corresponds to the complex number x + iy. Then rotation by θ corresponds to multi-
plication by
zθ = cos θ + i sin θ
since
zθ (x + iy) = (cos θ + i sin θ)(x + iy)

= x cos θ − y sin θ + i(x sin θ + y cos θ).
Composition of rotations corresponds to multiplication of complex numbers since rotating

by θ and then by ϕ is the same as multiplying by zϕ zθ .
Note that
S1 := {zθ | θ ∈ R}
is the precisely the set of complex numbers of absolute value 1. Thus S1 is the circle of radius
1 centred at the origin. Therefore, S1 has a geometric structure (as a circle) and a group
structure (via multiplication of complex numbers). The multiplication and inverse maps are
both smooth and so S1 is a Lie group.
Definition 1.2.2 (Matrix group and linear group). A matrix group is a set of invertible
matrices that is closed under multiplication and inversion. A linear group is a group that is
isomorphic to a matrix group.
Remark 1.2.3. Some references use the term linear group to mean a group consisting of
matrices (i.e. a matrix group as defined above).
Example 1.2.4. We see that SO(2) is a matrix group. Since S1 is isomorphic (as a group) to
SO(2) (Exercise 1.2.2), S1 is a linear group.
1.3. MATRIX REPRESENTATIONS OF COMPLEX NUMBERS 5
Exercises.
1.2.1. Verify that if u, v ∈ R2 and R ∈ SO(2), then the distance between the points u and v
is the same as the distance between the points Ru and Rv.
1.2.2. Prove that the map zθ 7→ Rθ , θ ∈ R, is a group isomorphism from S1 to SO(2).
1.3 Matrix representations of complex numbers

Define
1 0 0 −1
1= , i= .
0 1 1 0
The set of matrices

a −b
C̃ := {a1 + bi | a, b ∈ R} = a, b ∈ R
b a
is closed under addition and multiplication, and hence forms a subring of M2 (R) (Exer-
cise 1.3.1).
Theorem 1.3.1. The map

C → C̃, a + bi 7→ a1 + bi,
is a ring isomorphism.
Proof. It is easy to see that it is a bijective map that commutes with addition and scalar
multiplication. Since we have
12 = 1, 1i = i1 = i, i2 = −1,
it is also a ring homomorphism.
Remark 1.3.2. Theorem 1.3.1 will allow us to convert from matrices with complex entries to
(larger) matrices with real entries.
Note that thesquaredabsolute value |a + bi|2 = a2 + b2 is the determinant of the corre-

a −b
sponding matrix . Let z1 , z2 be two complex numbers with corresponding matrices
b a
A1 , A2 . Then
|z1 |2 |z2 |2 = det A1 det A2 = det(A1 A2 ) = |z1 z2 |2 ,
and thus
|z1 ||z2 | = |z1 z2 |.
So multiplicativity of the absolute value corresponds to multiplicativity of determinants.
Note also that if z ∈ C corresponds to the matrix A, then

a − bi
z −1 =
a2 + b2
corresponds to the inverse matrix
−1
a −b 1 a b
= 2 .
b a a + b2 −b a
Of course, this also follows from the ring isomorphism above.
Exercises.
1.3.1. Show that the set of matrices

a −b
C̃ := {a1 + bi | a, b ∈ R} = a, b ∈ R
b a
is closed under addition and multiplication and hence forms a subring of Mn (R).
1.4 Quaternions
Definition 1.4.1 (Quaternions). Define a multiplication on the real vector space with basis
{1, i, j, k} by
1i = i1 = i, 1j = j1 = j, 1k = k1 = k,
ij = −ji = k, jk = −kj = i, ki = −ik = j,
12 = 1, i2 = j 2 = k 2 = −1,
and extending by linearity. The elements of the resulting ring (or algebra) H are called
quaternions.
Strictly speaking, we need to check that the multiplication is associative and distributive
over addition before we know it is a ring (but see below). Note that the quaternions are not
commutative.
We would like to give a matrix realization of quaternions like we did for the complex
numbers. Define 2 × 2 complex matrices

1 0 0 −1 0 −i i 0
1= , i= , j= , k= .
0 1 1 0 −i 0 0 −i
Then define

a + di −b − ci
H̃ = {a1 + bi + cj + dk | a, b, c, d ∈ R} = a, b, c, d ∈ R
b − ci a − di
1.4. QUATERNIONS 7

z w
= z, w ∈ C .
−w̄ z̄
The map
H → H̃, a + bi + cj + dk 7→ a1 + bi + cj + dk, (1.1)
is an isomorphism of additive groups that commutes with the multiplication (Exercise 1.4.1).
It follows that the quaternions satisfy the following properties.
Addition:
• Commutativity. q1 + q2 = q2 + q1 for all q1 , q2 ∈ H.
• Associativity. q1 + (q2 + q3 ) = (q1 + q2 ) + q3 for all q1 , q2 , q3 ∈ H.
• Inverse law. q + (−q) = 0 for all q ∈ H.
• Identity law. q + 0 = q for all q ∈ H.
Multiplication:
• Associativity. q1 (q2 q3 ) = (q1 q2 )q3 for all q1 , q2 , q3 ∈ H.
• Inverse law. qq −1 = q −1 q = 1 for q ∈ H, q 6= 0. Here q −1 is the quaternion correspond-
ing to the inverse of the matrix corresponding to q.
• Identity law. 1q = q1 = q for all q ∈ H.
• Left distributive law. q1 (q2 + q3 ) = q1 q2 + q1 q3 for all q1 , q2 , q3 ∈ H.
• Right distributive law. (q2 + q3 )q1 = q2 q1 + q3 q1 for all q1 , q2 , q3 ∈ H.
In particular, H is a ring. We need the right and left distributive laws because H is not
commutative.
Remark 1.4.2. Note that C is a subring of H (spanned by 1 and i).
Following the example of complex numbers, we define the absolute value of the quaternion
q = a + bi + cj + dk to be the (positive) square root of the determinant of the corresponding
matrix. That is
2 a + id −b − ic
|q| = det = a2 + b 2 + c 2 + d 2 .
b − ic a − id
In other words, |q| is the distance of the point (a, b, c, d) from the origin in R4 .
As for complex numbers, multiplicativity of the determinant implies multiplicativity of
absolute values of quaternions:
|q1 q2 | = |q1 ||q2 | for all q ∈ H.
From our identification with matrices, we get an explicit formula for the inverse. If
q = a + bi + cj + dk, then
1
q −1 = (a − bi − cj − dk).
a2 + b2 + c2 + d2
If q = a + bi + cj + dk, then q̄ := a − bi − cj − dk is called the quaternion conjugate of q.

So we have
q̄q = q q̄ = a2 + b2 + c2 + d2 = |q|2 .
Because of the multiplicative property of the absolute value of quaternions, the 3-sphere
of unit quaternions
{q ∈ H | |q| = 1} = {a + bi + cj + dk | a2 + b2 + c2 + d2 = 1}
is closed under multiplication. Since H can be identified with R4 , this shows that S3 is a
group under quaternion multiplication (just like S1 is group under complex multiplication).
If X is a matrix with complex entries, we define X ∗ to be the matrix obtained from the
transpose X T by taking the complex conjugate of all entries.
Definition 1.4.3 (U(n) and SU(n)). A matrix X ∈ Mn (C) is called unitary if X ∗ X = In .

The unitary group U(n) is the subgroup of GL(n, C) consisting of unitary matrices. The
special unitary group SU(n) is the subgroup of U(n) consisting of matrices of determinant 1.
Remark 1.4.4. Note that X ∗ X = I implies that | det X| = 1.
Proposition 1.4.5. The group S3 of unit quaternions is isomorphic to SU(2).
Proof. Recall that under our identification of quaternions with matrices, absolute value
corresponds to the determinant. Therefore, the group of unit quaternions is isomorphic to
the matrix group
z w
Q= det Q = 1 .
−w̄ z̄

z w −1 y −w
Now, if Q = , w, x, y, z ∈ C, and det Q = 1, then Q = . Therefore
x y −x z

∗ −1 z̄ x̄ y −w
Q =Q ⇐⇒ = ⇐⇒ y = z̄, w = −x̄
w̄ ȳ −x z
and the result follows.
Exercises.
1.4.1. Show that the map (1.1) is an isomorphism of additive groups that commutes with
the multiplication.
1.4.2. Show that if q ∈ H corresponds to the matrix A, then q̄ corresponds to the matrix A∗ .
Show that it follows that
q1 q2 = q̄2 q̄1 .
1.5. QUATERNIONS AND SPACE ROTATIONS 9
1.5 Quaternions and space rotations

The pure imaginary quaternions
Ri + Rj + Rk := {bi + cj + dk | b, c, d ∈ R}
form a three-dimensional subspace of H, which we will often simply denote by R3 when the
context is clear. This subspace is not closed under multiplication. An easy computation
shows that
(u1 i + u2 j + u3 k)(v1 i + v2 j + v3 k)
= −(u1 v1 + u2 v2 + u3 v3 ) + (u2 v3 − u3 v2 )i − (u1 v3 − u3 v1 )j + (u1 v2 − u2 v1 )k.
If we identify the space of pure imaginary quaternions with R3 by identifying i, j, k with the
standard unit vectors, we see that
uv = −u · v + u × v,
where u × v is the vector cross product.

Recall that u × v = 0 if u and v are parallel (i.e. if one is a scalar multiple of the other).
Thus, if u is a pure imaginary quaterion, we have
u2 = −u · u = −|u|2 .
So if u is a unit vector in Ri + Rj + Rk, then u2 = −1. So every unit vector in Ri + Rj + Rk

is a square root of −1.
Let
t = t0 + t1 i + t2 j + t3 k ∈ H, |t| = 1.
Let
tI = t1 i + t2 j + t3 k
be the imaginary part of t. We have
t = t0 + tI , 1 = |t|2 = t20 + t21 + t22 + t23 = t20 + |tI |2 .
Therefore, (t0 , |tI |) is a point on the unit circle and so there exists θ such that
t0 = cos θ, |tI | = sin θ
and
tI
t = cos θ + |tI | = cos θ + u sin θ
|tI |
where
tI
u=
|tI |
is a unit vector in Ri + Rj + Rk and hence u2 = −1.
We want to associate to the unit quaternion t a rotation of R3 . However, this cannot
be simply by multiplication by t since this would not preserve R3 . However, note that
multiplication by t (on the left or right) preserves distances in R4 (we identify H with R4
here) since if u, v ∈ H, then
|tu − tv| = |t(u − v)| = |t||u − v| = |u − v|, and
|ut − vt| = |(u − v)t| = |u − v||t| = |u − v|.
It follows that multiplication by t preserves the dot product on R4 . Indeed, since it sends
zero to zero, it preserves absolute values (since |u| = |u − 0| is the distance from u to zero),
and since we can write the dot product in terms of absolute values,
1
u · v = (|u + v|2 − |u|2 − |v|2 ),
2
multiplication by t preserves the dot product. Therefore, conjugation by t
q 7→ tqt−1
is an isometry (e.g. preserves distances, angles, etc.). Note that this map also fixes the real
numbers since for r ∈ R,
trt−1 = tt−1 r = 1 · r = r ∈ R.
Therefore, it maps Ri + Rj + Rk (the orthogonal complement to R) to itself.
Recall that if u is a (unit) vector in R, then rotation through an angle about u is rotation
about the line determined by u (i.e. the line through the origin in the direction u) in the
direction given by the right hand rule.
Proposition 1.5.1. Let t = cos θ + u sin θ, where u ∈ Ri + Rj + Rk is a unit vector. Then

conjugation by t on Ri + Rj + Rk is rotation through angle 2θ about u.
Proof. Note that

t̄
t−1 = = cos θ − u sin θ
|t|2
and so conjugation by t fixes Ru since
tut−1 = (cos θ + u sin θ)u(cos θ − u sin θ)
= (u cos θ + u2 sin θ)(cos θ − u sin θ)
= (u cos θ − sin θ)(cos θ − u sin θ)
= u(cos2 θ + sin2 θ) − sin θ cos θ − u2 sin θ cos θ
=u
(and the conjugation map is linear in R). Therefore, since conjugation by t is an isometry,
it is determined by what it does to the orthogonal complement to Ru (i.e. the plane in R3
through the origin orthogonal to u). It suffices to show that the action of the conjugation
map on this plane is rotation through angle 2θ (in the direction given by the right hand
rule).
Let v ∈ R3 be a unit vector orthogonal to u (i.e. u · v = 0) and let w = u × v. Then
{u, v, w} is an orthonormal basis of R3 . We have
uv = −u · v + u × v = u × v.
Similarly,
uv = −vu = w, vw = −wv = u, wu = −uw = v.
We compute
twt−1 = (cos θ + u sin θ)w(cos θ − u sin θ)

= (w cos θ + uw sin θ)(cos θ − u sin θ)
= w cos2 θ + uw sin θ cos θ − wu sin θ cos θ − uwu sin2 θ
= w cos2 θ − 2wu sin θ cos θ + u2 w sin2 θ
= w(cos2 θ − sin2 θ) − 2v sin θ cos θ
= w cos 2θ − v sin 2θ.
Similarly,
tvt−1 = v cos 2θ + w sin 2θ.
Therefore, in the basis {v, w}, conjugation by t is given by

cos 2θ − sin 2θ
sin 2θ cos 2θ
and is thus rotation by an angle 2θ. This is rotation measured in the direction from v to w
and is thus in the direction given by the right hand rule (with respect to u).
Remark 1.5.2. In [Sti08, §1.5], the conjugation map is given by q 7→ t−1 qt. This gives a
rotation by −2θ instead of a rotation by 2θ (since it is the inverse to the conjugation used
above).
Therefore, rotation of R3 through an angle α about the axis u is given by conjugation by

α α
t = cos + u sin
2 2
and so all rotations of R3 arise as conjugation by a unit quaternion.

Note that
(−t)q(−t)−1 = tqt−1
and so conjugation by −t is the same rotation as conjugation by t. We can also see this since

α α α α α + 2π α + 2π
−t = − cos − u sin = cos + π + u sin + π = cos + u sin
2 2 2 2 2 2
which is rotation through an angle of α + 2π about u, which is the same transformation.

Are there any other quaternions that give the same rotation? We could have rotation
through an angle of −α about −u:
α α α α
cos − + (−u) sin − = cos + u sin = t.
2 2 2 2
Definition 1.5.3 (O(n) and SO(n)). The subgroup of GL(n, R) consisting of orthogonal
matrices is called the orthogonal group and is denoted O(n). That is,
O(n) = {X ∈ GL(n, R) | XX T = In }.
The special orthogonal group SO(n) is the subgroup of O(n) consisting of matrices of deter-
minant 1:
SO(n) = {X ∈ GL(n, R) | XX T = In , det X = 1}.
Remark 1.5.4. Note that XX T = In implies X T X = In and det X = ±1.
Proposition 1.5.5. The rotations of R3 form a group isomorphic to SO(3).
Proof. First note that by choosing an orthonormal basis for R3 (for instance, the standard
basis), we can identify linear transformations of R3 with 3 × 3 matrices. The dot product in
R3 is a bilinear form given by (v, w) 7→ v · w = v T w. Thus, an element of M3 (R) preserves
the dot product (equivalently, distances) if and only if for all v, w ∈ R3 ,
v T w = (Xv)T (Xw) = v(X T X)w.
This true iff X T X = I3 (take v and w to be the standard basis vectors to show that each entry
in X T X must equal the corresponding entry in I3 ). Therefore O(3) is the group of matrices
preserving the bilinear form. Since rotations preserve the bilinear form, all rotations are
elements of O(3). In fact, since rotations preserve orientation, they are elements of SO(3).
It remains to show that every element of SO(3) is a rotation (through an angle about some
axis).
Recall that rotations fix an axis (the axis of rotation). Thus, any rotation has 1 as an
eigenvalue (the corresponding eigenvector is any nonzero vector on the axis). So we first
show that any element of SO(3) has 1 as an eigenvalue.
Let X ∈ SO(3). Then
det(X − I) = det(X − I)T = det(X T − I) = det(X −1 − I) = det X −1 (I − X)

= det X −1 det(I − X) = det(I − X) = − det(X − I).
Thus
2 det(X − I) = 0 =⇒ det(X − I) = 0
and so 1 is an eigenvalue of X with some unit eigenvector u. Thus X fixes the line Ru and
its orthogonal complement (since it preserves the dot product). If we pick an orthonormal
3
basis {v, w} of this orthogonal complement, then {u, v, w} is an orthonormal basis of R (if

necessary, switch the order of v and w so that this basis is right-handed). Let A = u v w .
Then A is orthogonal (check!) and

−1 T 1 0
A XA = A XA =
0 Y
where Y ∈ M2 (C). Then 1 = det X = 1 · det Y = det Y and

T
1 0 1 0
= = (AT XA)T = AT X T A = AT X −1 A
0 YT 0 Y
−1
T −1 1 0 1 0
= (A XA) = = .
0 Y 0 Y −1
Thus Y T = Y −1 and so Y ∈ SO(2). But we showed earlier that SO(2) consists of the 2 × 2
rotation matrices. Thus there exists θ such that
 
1 0 0
A−1 XA = 0 cos θ − sin θ
0 sin θ cos θ
and so X is rotation through the angle θ about the axis u.
Corollary 1.5.6. The rotations of R3 form a subgroup of the group of isometries of R3 .

In other words, the inverse of a rotation is a rotation and the product of two rotations is a
rotation.
Note that the statement involving products is obvious for rotations of R2 but not of R3 .
Proposition 1.5.7. There is a surjective group homomorphism SU(2) → SO(3) with kernel
{±1} (i.e. {±I2 }).
Proof. Recall that we can identity the group of unit quaternions with the group SU(2). By
the above, we have a surjective map
ϕ : SU(2) −→ {rotations of R3 } ∼
= SO(3),
−1
t 7→ (q 7→ t qt)
and ϕ(t1 ) = ϕ(t2 ) iff t1 = ±t2 . In particular, the kernel of the map is {±1}. It remains to
show that this map is a group homomorphism. Suppose that
αi αi
ti = cos + ui sin .
2 2
and let ri , i = 1, 2, be rotation through angle αi about axis ui . Then ri corresponds to
conjugation by ti . That is, ϕ(ti ) = ri , i = 1, 2. The composition of rotations r2 r1 (r1 followed
by r2 – we read functions from right to left as usual) corresponds to the composition of the
two conjugations which is the map
q 7→ t1 qt−1 −1 −1 −1
1 7→ t2 t1 qt1 t2 = (t2 t1 )q(t2 t1 ) .
Therefore ϕ(t2 t1 ) = r2 r1 = ϕ(t2 )ϕ(t1 ) and so ϕ is a group homomorphism.
Corollary 1.5.8. We have a group isomorphism SO(3) ∼

= SU(2)/{±1}.
Proof. This follows from the fundamental isomorphism theorem for groups.
Remark 1.5.9. Recall that the elements of SU(2)/{±1} are cosets {±t} and multiplication is
given by {±t1 }{±t2 } = {±t1 t2 }. The above corollary is often stated as “SU(2) is a double
cover of SO(3).” It has some deep applications to physics. If you “rotate” an electron
through an angle of 2π it is not the same as what you started with. This is related to
the fact that electrons are described by representations of SU(2) and not SO(3). One can
illustrate this idea with Dirac’s belt trick .
Proposition 1.5.7 allows one to identify rotations of R3 with pairs ±t of antipodal unit
quaternions. One can thus do things like compute the composition of rotations (and find
the axis and angle of the composition) via quaternion arithmetic. This is actually done in
the field of computer graphics.
Recall that a subgroup H of a group G is called normal if gHg −1 = H for all g ∈ G.
Normal subgroups are precisely those subgroups that arise as kernels of homomorphisms. A
group G is simple if its only normal subgroups are the trivial subgroup and G itself.
Proposition 1.5.10 (Simplicity of SO(3)). The group SO(3) is simple.
Proof. See [Sti08, p. 33]. We will return to this issue later with a different proof (see
Corollary 5.14.7).
1.6 Isometries of Rn and reflections

We now want to give a description of rotations in R4 via quaternions. We first prove some
results about isometries of Rn in general. Recall that an isometry of Rn is a map f : Rn → Rn
such that
|f (u) − f (v)| = |u − v|, ∀ u, v ∈ Rn .
Thus, isometries are maps that preserve distance. As we saw earlier, preserving dot products
is the same as preserving distance and fixing the origin.
Definition 1.6.1. A hyperplane H through O is an (n − 1)-dimensional subspace of Rn ,

and reflection in H is the linear endomorphism of Rn that fixes the points of H and reverses
vectors orthogonal to H.
We can give an explicit formula for reflection ru in the hyperplane orthogonal to a

(nonzero) vector u. It is
v·u
ru (v) = v − 2 2 u, v ∈ Rn . (1.2)
|u|
Theorem 1.6.2 (Cartan–Dieudonné Theorem). Any isometry of Rn that fixes the origin O
is the product of at most n reflections in hyperplanes through O.
Note: There is an error in the proof of this result given in [Sti08, p. 36]. It states there that
ru f is the identity on Ru when it should be Rv.
Proof. We prove the result by induction on n.

1.6. ISOMETRIES OF RN AND REFLECTIONS 15
Base case (n = 1). The only isometries of R fixing O are the identity and the map
x 7→ −x, which is reflection in O (a hyperplane in R).
Inductive step. Suppose the result is true for n = k − 1 and let f be an isometry of Rk
fixing O. If f is the identity, we’re done. Therefore, assume f is not the identity. Then there
exists v ∈ Rk such that f (v) = w 6= v. Let ru be the reflection in the hyperplane orthogonal
to u = v − w. Then
w·u w · (v − w)
ru (w) = w − 2 2
u=w−2 (v − w)
|u| |v − w|2
w·v−w·w
=w−2 (v − w)
v · v − 2w · v + w · w
w·v−w·w
=w−2 (v − w) (since v · v = f (v) · f (v) = w · w)
2w · w − 2w · v
= w + (v − w) = v.
Thus ru f (v) = ru (w) = v and so v is fixed by ru f . Since isometries are linear transformations,
ru f is the identity on the subspace Rv of Rn and is determined by its restriction g to the
Rk−1 orthogonal to Rv. By induction, g is the product of ≤ k − 1 reflections. Therefore
f = ru g is the product of ≤ k reflections.
Remark 1.6.3. The full Cartan–Dieudonné Theorem is actually more general, concerning
isometries of n-dimensional vector spaces over a field of characteristic not equal to 2 with a
non-degenerate bilinear form. What we proved above is just a special case.
Definition 1.6.4 (Orientation-preserving and orientation-reversing). A linear map is called

orientation-preserving if its determinant is positive and orientation-reversing otherwise.
Reflections are linear and have determinant −1. To see this, pick a basis compatible with
the reflection, i.e. a basis {v1 , . . . , vn } where {v1 , . . . , vn−1 } span the hyperplane of reflection
and vn is orthogonal to the hyperplane of reflection. Then in this basis, the reflection is
diagonal with diagonal entries 1, . . . , 1, −1.
So we see that a product of reflections is orientation-preserving iff it contains an even
number of terms.
Definition 1.6.5 (Rotation). A rotation of Rn about O is an orientation-preserving isometry

that fixes O.
It follows that if we choose an orthonormal basis of Rn (for instance, the standard basis)
so that linear maps correspond to n × n matrices, the rotations of Rn correspond to SO(n).
Exercises.
1.6.1. Verify that the map ru defined by (1.2) is indeed reflection in the hyperlane orthogonal
to the nonzero vector U . (It helps to draw a picture.) Note that it suffices to show that the
right hand side maps u to u and fixes any vector orthogonal to u.
1.7 Quaternions and rotations of R4

It follows from the Cartan–Dieudonné Theorem that any rotation of R4 is the product of 0,
2, or 4 reflections. Recall that we identify the group of unit quaternions with SU(2) and H
with R4 .
Proposition 1.7.1. Let u ∈ SU(2) be a unit quaternion. The map
ru : H → H, ru (q) = −uq̄u
is reflection in the hyperplane through O orthogonal to u.
Proof. Note that q 7→ −q̄ is the map
a + bi + cj + dk 7→ −a + bi + cj + dk, a, b, c, d ∈ R
and is therefore reflection in the hyperplane Ri + Rj + Rk, which is an isometry. We saw

before that left or right multiplication by a unit quaternion is an isometry and thus the
composition
q 7→ −q̄ 7→ −uq̄ 7→ −uq̄u
is also an isometry. Now, for v ∈ H, we have
ru (vu) = −u(vu)u = −uūv̄u = −|u|2 v̄u = −v̄u.
Therefore, we have
ru (u) = −u, ru (iu) = iu, ru (ju) = ju, ru (ku) = ku.
So ru reverses vectors parallel to u and fixes points in the space spanned by iu, ju and ku.
Therefore, it suffices to show that iu, ju and ku span the hyperplane through O orthogonal
to u. First note that if u = a + bi + cj + dk, then
u · iu = (a, b, c, d) · (−b, a, −d, c) = 0.
Similarly, one can show that u is also orthogonal to ju and ku. So it remains to show that
iu, ju and ku span a 3-dimensional subspace of H or, equivalently, that u, iu, ju, ku span all
of H. But this is true since for any v ∈ H, we have
v = (vu−1 )u
and we can express vu−1 as an R-linear combination of 1, i, j and k.
Proposition 1.7.2. The rotations of H = R4 about O are precisely the maps of the form
q 7→ vqw, where v, w ∈ SU(2) are unit quaternions.
1.8. SU(2) × SU(2) AND SO(4) 17
Proof. We know that any rotation of H is a product of an even number of reflections. The
composition of reflections in the hyperplanes orthogonal to u1 , u2 , . . . , u2n ∈ SU(2) is the
map
q 7→ −u1 q̄u1 7→ −u2 (−u1 q̄u1 )u2 = u2 ū1 qū1 u2 7→ . . . 7→ u2n · · · ū3 u2 ū1 qū1 u2 ū3 · · · u2n .
Thus we have that this composition is the map q 7→ vqw where
v = u2n · · · ū3 u2 ū1 , w = ū1 u2 ū3 · · · u2n .
It follows that all rotations are of this form.

It remains to show that all maps of the form q 7→ vqw for v, w ∈ SU(2) are rotations of
H about O. Since this map is the composition of left multiplication by v followed by right
multiplication by w, it is enough to show that multiplication of H on either side by a unit
quaternion is an orientation-preserving isometry (i.e. a rotation). We have already shown
that such a multiplication is an isometry, so we only need to show it is orientation-preserving.
Let
v = a + bi + cj + dk, a2 + b2 + c2 + d2 = 1,
be an arbitrary unit quaternion. Then in the basis {1, i, j, k} of H = R4 , left multiplication
by v has matrix  
a −b −c −d
 b a −d c 
 
c d a −b 
d −c b a
and one can verify by direct computation that this matrix has determinant one. The proof
that right multiplication by w is orientation-preserving is analogous.
1.8 SU(2) × SU(2) and SO(4)

Recall that the direct product G × H of groups G and H is the set
G × H = {(g, h) | g ∈ G, h ∈ H}
with multiplication
(g1 , h1 ) · (g2 , h2 ) = (g1 g2 , h1 h2 ).
The unit of G × H is (1G , 1H ) where 1G is the unit of G and 1H is the unit of H and the
inverse of (g, h) is (g −1 , h−1 ).
If G is a group of n × n matrices and H is a group of m × m matrices, then the map

g 0
(g, h) 7→
0 h
is an isomorphism from G × H to the group

g 0
g ∈ G, h ∈ H
0 h
of block diagonal matrices (n + m) × (n + m) matrices since

g1 0 g2 0 g1 g2 0
= .
0 h1 0 h2 0 h1 h2
For each pair of unit quaternions (v, w) ∈ SU(2) × SU(2), we know that the map
q 7→ vqw−1
is a rotation of H = R4 (since w−1 is also a unit quaterion) and is hence an element of SO(4).
Proposition 1.8.1. The map
ϕ : SU(2) × SU(2) → SO(4), (v, w) 7→ (q 7→ vqw−1 ),
is a surjective group homomorphism with kernel {(1, 1), (−1, −1)}.
Proof. For q ∈ H,
ϕ((v1 , w1 )(v2 , w2 ))(q) = ϕ(v1 v2 , w1 w2 )(q) = (v1 v2 )q(w1 w2 )−1

= v1 (v2 qw2−1 )w1−1 = ϕ(v1 , w1 )ϕ(v2 , w2 )(q).
Therefore
ϕ((v1 , w1 )(v2 , w2 )) = ϕ(v1 , w1 )ϕ(v2 , w2 ),
and so ϕ is a group homomorphism. It is surjective since we showed earlier that any rotation
of R4 is of the form q 7→ vqw−1 for v, w ∈ SU(2).
Now suppose that (v, w) ∈ ker ϕ. Then q 7→ vqw−1 is the identity map. In particular
v1w−1 = 1
and thus v = w. Therefore ϕ(v, w) = ϕ(v, v) is the map q 7→ vqv −1 . We saw in our
description of rotations of R3 using quaternions that this map fixes the real axis and rotates
the space of pure imaginary quaternions. We also saw that it is the identity map if and only
if v = ±1.
Remark 1.8.2. Our convention here is slightly different from the one used in [Sti08]. That
reference uses the map q 7→ v −1 qw. This difference is analogous to one noted in the de-
scription of rotations of R3 by quaternions (see Remark 1.5.2). Essentially, in [Sti08], the
composition of rotations r1 and r2 is thought of as being r1 r2 instead of r2 r1 . We use the
latter composition since it corresponds to the order in which you apply functions (right to
left).
Corollary 1.8.3. We have a group isomorphism (SU(2) × SU(2))/{±1} ∼ = SO(4). Here

1 = (1, 1) is the identity element of the product group SU(2) × SU(2) and −1 = (−1, −1).
Proof. This follows from the fundamental isomorphism theorem for groups.
Lemma 1.8.4. The group SO(4) is not simple.

1.8. SU(2) × SU(2) AND SO(4) 19
Proof. The subgroup SU(2) × {1} = {(v, 1) | v ∈ SU(2)} is the kernel of the group homo-
morphism (v, w) 7→ (1, w), projection onto the second factor, and is thus a normal subgroup
of SU(2) × SU(2). It follows that its image (under the map ϕ above) is a normal subgroup of
SO(4). This subgroup consists of the maps q 7→ vq. This subgroup is nontrivial and is not
the whole of SO(4) because since it does not contain the map q 7→ qw−1 for w 6= 1 (recall
that maps q 7→ v1 qw1−1 and q 7→ v2 qw2−1 are equal if and only if (v1 , w1 ) = (±v1 , ±w1 ) by the
above corollary).
Chapter 2
Matrix Lie groups
In this chapter, we will study what is probably the most important class of Lie groups,
matrix Lie groups. In addition to containing the important subclass of classical Lie groups,
focussing on matrix Lie groups (as opposed to abstract Lie groups) allows us to avoid much
discussion of abstract manifold theory.
2.1 Definitions
2 2
Recall that we can identify Mn (R) with Rn and Mn (C) with Cn by interpreting the n2
entries
a11 , a12 , . . . , a1n , a21 , . . . , a2n , . . . , an1 , . . . , ann
as the coordinates of a point/matrix.
Definition 2.1.1 (Convergence of matrices). Let (Am )m∈N be a sequence of elements of

Mn (C). We say that Am converges to a matrix A if each entry of Am converges (as m → ∞)
to the corresponding entry of A. That is, Am converges to A if
lim |(Am )k` − Ak` | = 0, ∀ 1 ≤ k, ` ≤ n.

m→∞
2
Note that convergence in the above sense is the same as convergence in Cn .
Definition 2.1.2 (Matrix Lie group). A matrix group is any subgroup of GL(n, C). A
matrix Lie group is any subgroup G of GL(n, C) with the following property: If (Am )m∈N is
a sequence of matrices in G, and Am converges to some matrix A then either A ∈ G, or A
is not invertible.
Remark 2.1.3. The convergence condition on G in Definition 2.1.2 is equivalent to the condi-
tion that G be a closed subset of GL(n, C). Note that this does not imply that G is a closed
subset of Mn (C). So matrix Lie groups are closed subgroups of GL(n, C).
Definition 2.1.4 (Linear group). A linear group is any group that is isomorphic to a matrix
group. A linear Lie group is any group that is isomorphic to a matrix Lie group. We will
sometimes abuse terminology and refer to linear Lie groups as matrix Lie groups.
20
2.2. FINITE GROUPS 21
Before discussing examples of matrix Lie groups, let us give a non-example. Let G be
the set of all n × n invertible matrices with (real) rational entries. Then, for example,
Pm 1   
k=0 k! 0 ··· ··· 0 e 0 ··· ··· 0
.. ..   ... .. 
 0 1 . .  0 1 .


lim  .
. . . . . . . .
.
 . .
 =  .. . . . . . . .
.

,
m→∞ 
. . . . .   . . . 
 .. ..   .. .. 
 . . 1 0   . . 1 0
0 ··· ··· 0 1 0 ··· ··· 0 1
which is invertible but not in G. Therefore G is not a matrix Lie group.

We now discuss important examples of matrix Lie groups.
2.2 Finite groups

Suppose G is a finite subgroup of GL(n, C). Then any convergent sequence (Am )m∈N in G is
eventually stable (i.e. there is some A ∈ G such that Am = A for m sufficiently large). Thus
G is a matrix Lie group.
Recall that Sn is isomorphic to a subgroup of GL(n, C) as follows. Consider the standard
basis X = {e1 , . . . , en } of Cn . Then permutations of this set correspond to elements of
GL(n, C). Thus Sn is a linear Lie group.
Any finite group G acts on itself (regarded as a set) by left multiplication. This yields
an injective group homomorphism G → SG and so G is isomorphic to a subgroup of SG and,
hence, all finite groups are linear Lie groups.
2.3 The general and special linear groups

The complex general linear group GL(n, C) is certainly a subgroup of itself. If (Am )m∈N
is a sequence of matrices in GL(n, C) converging to A, then A is in GL(n, C) or A is not
invertible (by the definition of GL(n, C)). Therefore GL(n, C) is a matrix Lie group.
GL(n, R) is a subgroup of GL(n, C) and if Am is a sequence of matrices in GL(n, R)
converging to A, then the entries of A are real. Thus either A ∈ GL(n, R) or A is not
invertible. Therefore the real general linear group GL(n, R) is a matrix Lie group.
The real special linear group SL(n, R) and complex special linear group SL(n, C) are both
subgroups of GL(n, C). Suppose (Am )m∈N is a sequence of matrices in SL(n, R) or SL(n, C)
converging to A. Since the determinant is a continuous function of the entries of a matrix,
det A = 1. Therefore SL(n, R) and SL(n, C) are both matrix Lie groups.
2.4 The orthogonal and special orthogonal groups

Recall that we have an inner product (dot product) on Rn defined as follows. If u =
(u1 , . . . , un ) and v = (v1 , . . . , vn ), then
u · v = uT v = u1 v1 + · · · + un vn .
22 CHAPTER 2. MATRIX LIE GROUPS
(In the above, we view u and v as column vectors.) We have that |u|2 = u · u and we know
that a linear transformation preserves length (or distance) if and only if it preserves the inner
product. We showed that an n × n matrix X preserves the dot product if and only if
X ∈ O(n) = {A ∈ GL(n, R) | AT A = In }.
Since we defined rotations to be orientation preserving isometries, we have that X is a

rotation if and only if
X ∈ SO(n) = {A ∈ O(n) | det A = 1}.
One easily checks (Exercise 2.4.1) that the orthogonal group O(n) and special orthogonal
group SO(n) are closely under multiplication and inversion and contain the identity matrix.
For instance, if A, B ∈ O(n), then
(AB)T (AB) = B T AT AB = B T IB = B T B = I.
Thus SO(n) is a subgroup of O(n) and both are subgroups of GL(n, C).
Suppose (Am )m∈N is a sequence of matrices in O(n) converging to A. Since multiplication
of matrices is a continuous function (Exercise 2.4.2), we have that AT A = In and hence
A ∈ O(n). Therefore O(n) is a matrix Lie group. Since the determinant is a continuous
function, we see that SO(n) is also a matrix Lie group.
Remark 2.4.1. Note that we do not write O(n, R) because, by convention, O(n) always
consists of real matrices.
Exercises.
2.4.1. Verify that SO(n) and O(n) are both subgroups of GL(n, C).
2.4.2. Show that multiplication of matrices is a continuous function in the entries of the
matrices.
2.5 The unitary and special unitary groups

The unitary group
U(n) = {A ∈ GL(n, C) | AA∗ = In }, where (A∗ )jk = Akj
is a subgroup of GL(n, C). The same argument used for the orthogonal and special orthogonal
groups shows that U(n) is a matrix Lie group, and so is the special unitary group
SU(n) = {A ∈ U (n) | det A = 1}.

2.6. THE COMPLEX ORTHOGONAL GROUPS 23
Remark 2.5.1. A unitary matrix can have determinant eiθ for any θ, whereas an orthogonal
matrix can only have determinant ±1. Thus SU(n) is a “smaller” subset of U(n) than SO(n)
is of O(n).
Remark 2.5.2. For a real or complex matrix A, the condition AA∗ = I (which reduces to
AAT = I if A is real) is equivalent to the condition A∗ A = I. This follows from the standard
result in linear algebra that if A and B are square and AB = I, then B = A−1 (one doesn’t
need to check that BA = I as well).
Exercises.
2.5.1. For an arbitrary θ ∈ R, write down an element of U(n) with determinant eiθ .
2.5.2. Show that the unitary group is precisely the group of matrices preserving the bilinear
form on Cn given by
hx, yi = x1 ȳn + · · · + xn ȳn .
2.6 The complex orthogonal groups

The complex orthogonal group is the subgroup
O(n, C) = {A ∈ GL(n, C) | AAT = In }
of GL(n, C). As above, we see that O(n, C) is a matrix Lie group and so is the special
complex orthogonal group
SO(n, C) = {A ∈ O(n, C) | det A = 1}.
Note that O(n, C) is the subgroup of GL(n, C) preserving the bilinear form on Cn given
by
hx, yi = x1 y1 + · · · + xn yn .
Note that this is not an inner product since it is symmetric rather than conjugate-symmetric.
2.7 The symplectic groups

We have a natural inner product on Hn given by
h(p1 , . . . , pn ), (q1 , . . . , qn )i = p1 q̄1 + . . . pn q̄n .
Note that Hn is not a vector space over H since quaternions do not act properly as “scalars”
since their multiplication is not commutative (H is not a field).
Since quaternion multiplication is associative, matrix multiplication of matrices with

quaternionic entries is associative. Therefore we can use them to define linear transformations
of Hn . We define the (compact) symplectic group Sp(n) to be the subset of Mn (H) consisting
of those matrices that preserve the above bilinear form. That is,
Sp(n) = {A ∈ Mn (H) | hAp, Aqi = hp, qi ∀ p, q ∈ Hn }.
The proof of the following lemma is an exercise (Exercise 2.7.1). It follows from this lemma
that Sp(n) is a matrix Lie group.
Lemma 2.7.1. We have
Sp(n) = {A ∈ Mn (H) | AA? = In }
where A? denotes the quaternion conjugate transpose of A.
Remark 2.7.2. Note that, for a quaternionic matrix A, the condition AA? = I is equivalent
to the condition A? A = I. This follows from the fact that Sp(n) is a group, or from the
realization of quaternionic matrices as complex matrices (see Section 1.4 and Remark 2.5.2).
Because H is not a field, it is often useful to express Sp(n) in terms of complex matrices.
Recall that we can identify quaternions with matrices of the form

α −β
, α, β ∈ C.
β̄ ᾱ
Replacing the quaternion entries in A ∈ Mn (H) with the corresponding 2 × 2 complex

matrices produces a matrix in A ∈ M2n (C) and this identification is a ring isomorphism.
Preserving the inner product means preserving length in the corresponding real space
R4n . For example, Sp(1) consists of the 1 × 1 quaternion matrices, multiplication by which
preserves length in H = R4 . We saw before that these are simply the unit quaternions, which
we identified with SU(2). Therefore,

a + id −b − ic 2 2 2 2
Sp(1) = a + b + c + d = 1 = SU(2).
b − ic a − id
We would like to know, for general values of n, which matrices in M2n (C) correspond to
elements of Sp(n).
Define a skew-symmetric bilinear form B on R2n or C2n by
n
X
B((x1 , y1 , . . . , xn , yn ), (x01 , y10 , . . . , x0n , yn0 )) = xk yk0 − yk x0k .
k=1
Then the real symplectic group Sp(n, R) (respectively, complex symplectic group Sp(n, C)) is
the subgroup of GL(2n, R) (respectively, GL(2n, C)) consisting of matrices preserving B.
Let
0 1
J= ,
−1 0
2.8. THE HEISENBERG GROUP 25
and let  
J 0
J2n = 
 .. .

.
0 J
Then
B(x, y) = xT J2n y
and
Sp(n, R) = {A ∈ GL(2n, R) | AT J2n A = J2n },

Sp(n, C) = {A ∈ GL(2n, C) | AT J2n A = J2n }.
Note that we use the transpose (not the conjugate transpose) in the definition of Sp(n, C).
For a symplectic (real or complex) matrix, we have
det J2n = det(AT J2n A) = (det A)2 det J2n =⇒ det A = ±1.
In fact one can show that det A = 1 for A ∈ Sp(n, R) or A ∈ Sp(n, C). We will come back
to this point later.
The proof of the following proposition is an exercise (see [Sti08, Exercises 3.4.1–3.4.3]).
Proposition 2.7.3. We have
Sp(n) = Sp(n, C) ∩ U(2n).
Remark 2.7.4. Some references write Sp(2n, C) for what we have denoted Sp(n, C).
Example 2.7.5 (The metaplectic group). There do exist Lie groups that are not matrix Lie
groups. One example is the metaplectic group Mp(n, R), which is the unique connected
double cover (see Section 6.6) of the symplectic Lie group Sp(n, R). It is not a matrix Lie
group because it has no faithful finite-dimensional representations.
Exercises.
2.7.1. Prove Lemma 2.7.1.
2.8 The Heisenberg group

The Heisenberg group H is the set of all 3 × 3 real matrices of the form
 
1 a b
A = 0 1 c  , a, b, c ∈ R.
0 0 1
It is easy to check that I ∈ H and that H is closed under multiplication. Also

 
1 −a ac − b
A−1 = 0 1 −c 
0 0 1
and so H is closed under inversion. It is clear that the limit of matrices in H is again in H
and so H is a matrix Lie group.
The name “Heisenberg group” comes from the fact that the Lie algebra of H satisfies the
Heisenberg commutation relations of quantum mechanics.
2.9 The groups R∗, C∗, S1 and Rn

The groups R∗ and C∗ under matrix multiplication are isomorphic to GL(1, R) and GL(1, C),
respectively, and so we view them as matrix Lie groups. The group S1 of complex numbers
with absolute value one is isomorphic to U(1) and so we also view it as a matrix Lie group.
The group Rn under vector addition is isomorphic to the group of diagonal real matrices
with positive diagonal entries, via the map
 
e x1 0
(x1 , . . . , xn ) 7→ 
 .. .

.
xn
0 e
One easily checks that this is a matrix Lie group and thus we view Rn as a matrix Lie group
as well.
2.10 The Euclidean group

The Euclidean group E(n) is the group of all one-to-one, onto, distance-preserving maps
from Rn to itself:
E(n) = {f : Rn → Rn | |f (x) − f (y)| = |x − y| ∀ x, y ∈ Rn }.
Note that we do not require these maps to be linear.

The orthogonal group O(n) is the subgroup of E(n) consisting of all linear distance-
preserving maps of Rn to itself. For x ∈ Rn , define translation by x, denoted Tx , by
Tx (y) = x + y.
The set of all translations is a subgroup of E(n).
Proposition 2.10.1. Every element of E(n) can be written uniquely in the form
Tx R, x ∈ Rn , R ∈ O(n).
Proof. The proof of this proposition can be found in books on Euclidean geometry.
2.11. HOMOMORPHISMS OF LIE GROUPS 27
One can show (Exercise 2.10.2) that the Euclidean group is isomorphic to a matrix Lie
group.
Exercises.
2.10.1. Show that, for x1 , x2 ∈ Rn and R1 , R2 ∈ O(n), we have
(Tx1 R1 )(Tx2 R2 ) = Tx1 +R1 x2 (R1 R2 )
and that, for x ∈ Rn and R ∈ O(n), we have
(Tx R)−1 = T−R−1 x R−1 .
This shows that E(n) is a semidirect product of the group of translations and the group O(n).
More precisely, E(n) is isomorphic to the group consisting of pairs (Tx , R), for x ∈ Rn and
R ∈ O(n), with multiplication
(x1 , R1 )(x2 , R2 ) = (x1 + R1 x2 , R1 R2 ).
2.10.2. Show that the map E(n) → GL(n + 1, R) given by

 
x1
 .. 
Tx R 7→ 
 R .
 xn 
0 ··· 0 1
is a one-to-one group homomorphism and conclude that E(n) is isomorphic to a matrix Lie
group.
2.11 Homomorphisms of Lie groups

Whenever one introduces a new type of mathematical object, one should define the allowed
maps between objects (more precisely, one should define a category).
Definition 2.11.1 (Lie group homomorphism/isomorphism). Let G and H be (matrix) Lie

groups. A map Φ from G to H is called a Lie group homomorphism if
(a) Φ is a group homomorphism, and
(b) Φ is continuous.
If, in addition, Φ is one-to-one and onto and the inverse map Φ−1 is continuous, then Φ is a
Lie group isomorphism.
Examples 2.11.2. (a) The map R → U (1) given by θ 7→ eiθ is a Lie group homomorphism.
(b) The map U (1) → SO(2) given by

iθ cos θ − sin θ
e 7→
sin θ cos θ
is a Lie group isomorphism (you should check that this map is well-defined and is indeed
an isomorphism).
(c) Composing the previous two examples gives the Lie algebra homomorphism R → SO(2)
defined by
cos θ − sin θ
θ 7→ .
sin θ cos θ
(d) The determinant is a Lie group homomorphism GL(n, C) → C∗ .
(e) The map SU(2) → SO(3) of Proposition 1.5.7 is continuous and is thus a Lie group
homomorphism (recall that this map has kernel {±1}).
Chapter 3
Topology of Lie groups
In this chapter, we discuss some topological properties of Lie groups, including connectedness
and compactness.
3.1 Connectedness
An important notion in theory of Lie groups is that of path-connectedness.
Definition 3.1.1 (Path-connected). Let A be a subset of Rn . A path in A is a continuous

map γ : [0, 1] → A. We say A is path-connected if, for any two points a, b ∈ A, there is a
path γ in A such that γ(0) = a and γ(1) = b.
Remark 3.1.2. If γ : [a, b] → A, a < b, is a continuous map, we can always rescale to obtain
a path γ̃ : [0, 1] → A. Therefore, we will sometimes also refer to these more general maps as
paths.
Remark 3.1.3. For those who know some topology, you know that the notion of path-
connectedness is not the same, in general, as the notion of connectedness. However, it turns
out that a matrix Lie group is connected if and only if it is path-connected. (This follows
from the fact that matrix Lie groups are manifolds, and hence are locally path-connected.)
Thus we will sometimes ignore the difference between the two concepts in this course.
The property of being connected by some path is an equivalence relation on the set of
points of a matrix Lie group. The equivalence classes are the (connected) components of the
matrix Lie group. Thus, the components have the property that two elements of the same
component can be joined by a continuous path but two elements of different components
cannot.
The proof of the following proposition is an exercise (see [Sti08, Exercises 3.2.1–3.2.3]).
Proposition 3.1.4. If G is a matrix Lie group, then the component of G containing the
identity is a subgroup of G.
Proposition 3.1.5. The group GL(n, C) is connected for all n ≥ 1.
29
30 CHAPTER 3. TOPOLOGY OF LIE GROUPS
We will give two proofs, one explicit and one not explicit (but shorter).
First proof. We will show that any matrix in GL(n, C) can be connected to the identity by
some path. Then any two elements of GL(n, C) can be connected by a path through the
identity.
Let A ∈ GL(n, C). Recall from linear algebra that every matrix is similar to an upper
triangular matrix (for instance, its Jordan canonical form). Thus, there exists C ∈ GL(n, C)
such that
A = CBC −1 ,
where B is of the form  
λ1 ?
B=
 ... .

0 λn
Since det A = det B = λ1 · · · λn and A is invertible, all the λi must be nonzero. Let B(t),
0 ≤ t ≤ 1, be obtained from B by multiplying all the entries above the diagonal by (1 − t),
and let A(t) = CB(t)C −1 . Then A(t) is a continuous path starting at A and ending at
CDC −1 where  
λ1 0
D=
 ... .

0 λn
This path is contained in GL(n, C) since
det A(t) = λ1 . . . λn = det A, for all t.
For 1 ≤ i ≤ n, choose a path λi : [1, 2] → C∗ such that λi (1) = λi and λi (2) = 1. This is
possible since C∗ is path-connected. Then define A(t) on the interval 1 ≤ t ≤ 2 by
 
λ1 (t) 0
A(t) = C 
 ..  −1
C .
.
0 λn (t)
This is a continuous path starting at CDC −1 , when t = 1, and ending at CIC −1 = I, when
t = 2. Since the λk (t) are always nonzero, A(t) lies in GL(n, C) for all t. Thus we have
constructed a path from A to I.
Second proof. Let A, B ∈ GL(n, C). Consider the plane
(1 − z)A + zB, z ∈ C.
Matrices of this form are not in GL(n, C) precisely when
det((1 − z)A + zB) = 0. (3.1)
Since (1−z)A+zB is an n×n complex matrix whose entries are linear in z, its determinant is
a polynomial of degree at most n in z. Therefore, by the Fundamental Theorem of Algebra,
3.1. CONNECTEDNESS 31
the left hand side of (3.1) has at most n roots. These roots correspond to at most n points
in the plane (1 − z)A + zB, z ∈ C, not including the points A or B. Therefore, we can find
a path from A to B which avoids these n points (since the plane minus a finite number of
points is path-connected).
The proof of the following proposition is left as an exercise (Exercise 3.1.1).
Proposition 3.1.6. The group SL(n, C) is connected for all n ≥ 1.
Proposition 3.1.7. For all n ≥ 1, the group O(n) is not connected.
Proof. Let A, B ∈ O(n) such that det A = 1 (for instance, A = I) and det B = −1 (for
instance, B = diag(1, . . . , 1, −1)). The determinant map det : Mn (R) → R is a polynomial
in the entries and is thus a continuous map. Suppose that γ is a path from A to B. Then
det ◦γ is a composition of continuous maps and hence is a continuous map [0, 1] → R. But
the image of det ◦γ lies in {±1} and we have det ◦γ(0) = 1, det ◦γ(1) = −1. Since there is
no continuous map R → {±1} starting at 1 and ending at −1, we have a contradiction.
Proposition 3.1.8. The group SO(n) is connected for n ≥ 1.
Proof. Since SO(1) = {(1)} is simply a point, it is connected. Also, we have seen that SO(2)
is a circle and is therefore also connected. Assume that SO(n − 1) is connected for some
n ≥ 2. We show that every A ∈ SO(n) can be joined to the identity by a path. We can
then conclude that SO(n) is connected since any two elements could be connected by a path
through the identity.
Let {e1 , . . . , en } be the standard basis of Rn . Let R be a rotation in a plane containing e1
and Ae1 such that RAe1 = e1 . Since SO(2) (the group of rotations in a plane) is connected,
there is a path R(t), 0 ≤ t ≤ 1, such that R(0) = I and R(t) = R. Then
γ(t) = R(t)A, 0 ≤ t ≤ 1,
is a path in SO(n) with γ(0) = A and γ(1) = RA. Since R and A are both orthogonal
matrices, so is RA. Thus RAej is orthogonal to RAe1 = e1 for all 2 ≤ j ≤ n. Therefore

1 0
RA = ,
0 A1
with A1 ∈ SO(n − 1). By induction, there is a continuous path from A1 to In−1 and hence
a path from RA to In . Following γ by this path yields a path from A to In .
Corollary 3.1.9. The group O(n) has 2 connected components.
Proof. It suffices to show that the set Y of matrices in O(n) with determinant equal to −1
is connected. Let A and B be two such matrices. Then XA, XB ∈ SO(n), where
 
−1 0
 1 
X= .
 
. .
 . 
0 1
Since SO(n) is connected, there is a path γ : [0, 1] → SO(n) with γ(0) = XA and γ(1) = XB.
Then Xγ(t) is a path from A to B in Y .
Proposition 3.1.10. The groups U(n) and SU(n) are connected for n ≥ 1.
Proof. We first show that every U ∈ U(n) can be joined to the identity by a continuous
path (hence showing that U(n) is connected). Recall from linear algebra that every unitary
matrix has an orthonormal basis of eigenvectors with eigenvalues of the form eiθ . Therefore,
we have  
eiθ1 0
U = U1 
 ...  −1
 U1 , (3.2)
0 eiθn
with U1 unitary (since its columns form an orthonormal basis) and θi ∈ R. Conversely, one
can easily check that any matrix of the form (3.2) is unitary. Therefore,
 
ei(1−t)θ1 0
U (t) = U1 
 ..  −1
 U1 , 0 ≤ t ≤ 1,
.
0 e1(1−t)θn
defines a continuous path in U(n) joining U to I.

Proving that SU(n) is connected involves a modification similar to the modification one
makes to the argument proving GL(n, C) is connected to prove that SL(n, C) is connected.
Proposition 3.1.11. The group Sp(n) is connected for n ≥ 1.
Proof. We prove this by induction (similar to the proof that SO(n) is connected). The only
major difference is the base case of Sp(2).
Recall that

q1 −q2 2 2
Sp(2) = q1 , q2 ∈ H, |q1 | + |q2 | = 1 .
q̄2 q̄1
Therefore
q1 = u1 cos θ, q2 = u2 sin θ
for some u1 , u2 ∈ H with |u1 | = |u2 | = 1.
We will show later that any unit quaternion is the exponential of a pure imaginary
quaternion. Thus there exist pure imaginary quaternions v1 and v2 such that ui = evi ,
i = 1, 2. Therefore
q1 (t) = etv1 cos θt, q2 (t) = etv2 sin θt

q1 −q2
gives a continuous path from I to in Sp(2). Thus Sp(2) is connected.
q̄2 q̄1
We leave the inductive step as an exercise.
Corollary 3.1.12. If A ∈ Sp(n, C) ∩ U(2n) (i.e. A is an element of Sp(n) written as a

complex matrix), then det A = 1.
3.2. POLAR DECOMPOSITIONS 33
Proof. Recall that we showed that all such matrices have determinant ±1. Since Sp(n) is
connected, it follows that they in fact all have determinant 1, since the determinant is a
continuous function and the identity matrix has determinant one.
The above corollary explains why we do not consider a “special symplectic group”.
Remark 3.1.13. One can also show that Sp(n, R) and Sp(n, C) are also connected.
Exercises.
3.1.1. Prove Proposition 3.1.6.
3.1.2. Prove that the Heisenberg group is connected and that the Euclidean group has 2
connected components.
3.2 Polar Decompositions

From now on, we will use the notation hv, wi to denote our standard inner product on Rn ,
Cn or Hn . That is,
hv, wi = v1 w̄1 + · · · + vn w̄n ,
where ¯ denotes the identity operation, complex conjugation, or quaternionic conjugation if
v and w lie in Rn , Cn , or Hn respectively.
The goal of this section is to discuss polar decompositions for SL(n, R) and SL(n, C) (and
other groups). These decompositions are useful for proving that SL(n, R) and SL(n, C) have
certain topological properties. One should think of these decompositions as analogues of the
unique decomposition of a nonzero complex number z as z = up where |u| = 1 and p is a
positive real number.
Recall that a matrix A is symmetric if AT = A. If v and w are eigenvectors of A with
distinct eigenvalues λ and µ (without loss of generality, we can assume µ 6= 0), then
λµ(v · w) = Av · Aw = (Av)T Aw = v T AT Aw = v T A2 w = µ2 v T w = µ2 v · w =⇒ v · w = 0.
Since symmetric matrices are diagonalizable, we see that any symmetric matrix A has an
orthonormal basis of eigenvectors. So we can write
A = RDR−1 = RDRT , (3.3)
where R is an orthogonal matrix (whose columns are an orthonormal basis of eigenvectors

for A) and D is a diagonal matrix. Conversely, one easily checks that any matrix of the form
(3.3) is symmetric.
Definition 3.2.1 (Positive real symmetric matrix). An n × n real symmetric matrix P is

positive (or positive-definite) if hx, P xi > 0 for all nonzero vectors x ∈ Rn . Equivalently, a
real symmetric matrix is positive if all of its eigenvalues are positive.
By the above, given a real symmetric positive matrix P , there exists an orthogonal matrix
R such that
P = RDR−1 ,
where D is diagonal with positive diagonal entries λ1 , . . . , λn . We can then construct a square
root of P as
P 1/2 = RD1/2 R−1 , (3.4)
1/2 1/2
where D is the diagonal matrix with positive diagonal entries λ1 , . . . , λn . So P 1/2 is also
symmetric and positive. In fact, one can show that P 1/2 is the unique positive symmetric
matrix whose square is P (Exercise 3.2.1).
Proposition 3.2.2 (Polar decomposition of SL(n, R)). For every A ∈ SL(n, R), there exists
a unique pair (R, P ) such that R ∈ SO(n), P is real, symmetric and positive, and A = RP .
The matrix P satisfies det P = 1.
Proof. If such a pair existed, we would have

AT A = (RP )T (RP ) = P T RT RP = P IP = P 2 .
Since (AT A)T = AT (AT )T = AT A, the matrix AT A is symmetric. It is also positive since
for all x ∈ Rn , x 6= 0, we have
hx, AT Axi = hAx, Axi > 0
since Ax 6= 0 (because A is invertible). Therefore we can define
P = (AT A)1/2 ,
and this P is real, symmetric, and positive. Since we want A = RP , we define
R = AP −1 = A((AT A)1/2 )−1 .
For the existence part of the proposition, it remains to show that R ∈ SO(n). Since
RT R = A((AT A)1/2 )−1 ((AT A)1/2 )−1 AT (recall that AT A is symmetric)
= A(AT A)−1 AT = AA−1 (AT )−1 AT = I,
we see that R ∈ O(n). Also, we have
1 = det A = det R det P.
Since P is positive (in particular, all its eigenvalues are positive), we have det P > 0. There-
fore, det R > 0 and so det R = 1 (since R is orthogonal, det R = ±1). It then follows that
det P = 1 as well.
To prove uniqueness, note that we saw above that we must have P 2 = AT A. By the
uniqueness of the square root of a real, symmetric, positive matrix, P is unique. Then
R = AP −1 is also unique.
3.2. POLAR DECOMPOSITIONS 35
Definition 3.2.3 (Positive self-adjoint complex matrix). If P is an n × n self-adjoint (or

hermitian) complex matrix (i.e. P ∗ = P ), then we say P is positive (or positive-definite) if
hx, P xi > 0 for all nonzero x ∈ Cn .
Proposition 3.2.4 (Polar decomposition of SL(n, C)). For every A ∈ SL(n, C), there exists
a unique pair (U, P ) with U ∈ SU(n), P self-adjoint and positive, and A = U P . The matrix
P satisfies det P = 1.
Proof. The proof is analogous to the proof of the polar decomposition for SL(n, R).
Remark 3.2.5. (a) Any complex matrix A can be written in the form A = U P where U is
unitary and P is a positive-semidefinite (hx, Axi ≥ 0 for all x ∈ Cn ) self-adjoint matrix.
We just do not have the uniqueness statement in general.
(b) Our decomposition gives a polar decomposition
det A = det P det U = reiθ
of the determinant of A since det P is a nonnegative real number and det U is a unit
complex number.
(c) We have similar (unique) polar decompositions for
GL(n, R), GL(n, R)+ = {A ∈ GL(n, R) | det A > 0}, and GL(n, C).
GL(n, R) : A = U P, U ∈ O(n), P real, symmetric, positive

GL(n, R)+ : A = U P, U ∈ SO(n), P real, symmetric, postive
GL(n, C) : A = U P, U ∈ U(n), P self-adjoint, positive
The proofs of these are left as an exercise. Note that the only difference between the
polar decomposition statements for GL(n, R)+ and SL(n, R) is that we do not conclude
that det P = 1 for GL(n, R)+ .
Exercises.
3.2.1. Show that P 1/2 given by (3.4) is the unique positive symmetric matrix whose square
is P . Hint: show that any matrix that squares to P has the same set of eigenvectors as P .
3.2.2. Prove the results stated in Remark 3.2.5(c).
3.2.3. Prove that a real symmetric matrix is positive if and only if all of its eigenvalues are
positive (see Definition 3.2.1).
3.3 Compactness
Definition 3.3.1 (Comact). A matrix Lie group G is compact if the following two conditions
are satisfied:
(a) The set G is closed in Mn (C): If Am is a sequence of matrices in G, and Am converges
to a matrix A, then A is in G.
(b) The set G is bounded: There exists a constant C ∈ R such that for all A ∈ G, |Aij | ≤ C
for all 1 ≤ i, j ≤ n.
Remark 3.3.2. (a) The conditions in the above definition say that G is a closed bounded
2 2 2
subset of Cn (when we identify Mn (C) with Cn ). For subsets of Cn , this is equivalent
to the usual, more general, definition of compact (that any open cover has a finite
subcover).
(b) All of our examples of matrix Lie groups except GL(n, R) and GL(n, C) satisfy the
closure condition above. Thus, we are most interested in the boundedness condition.
Proposition 3.3.3. The groups O(n), SO(n), U(n), SU(n) and Sp(n) are compact.
Proof. We have already noted that these groups satisfy the closure condition. The column
vectors of any matrix in the first four groups in the proposition have norm one (and also in
the last if we consider the complex form of elements of Sp(n)) and hence |Aij | ≤ 1 for all
1 ≤ i, j ≤ n.
Proposition 3.3.4. The following groups are noncompact:
GL(n, R), GL(n, C), n ≥ 1,

SL(n, R), SL(n, C), O(n, C), SO(n, C), n ≥ 2,
H, Sp(n, R), Sp(n, C), E(n), Rn , R∗ , C∗ , n ≥ 1.
Proof. The groups GL(n, R) and GL(n, C) violate the closure condition since a limit of
invertible matrices may be not invertible.
Since  
a
 1 
 a 

 1 


 ... 

1
has determinant one for any nonzero a, we see that the groups SL(n, R) and SL(n, C) are
not bounded for n ≥ 2.
We have
z −w
∈ SO(2, C)
w z
for z, w ∈ C with z 2 + w2 = 1. We can take z with |z| arbitrarily large and let w be any
solution to z 2 + w2 = 1 (such solutions always exist over the complex numbers) and thus
3.4. LIE GROUPS 37
SO(2) is unbounded. By considering block matrices, we see that SO(n) is unbounded for
n ≥ 2.
We leave the remaining cases as an exercise (Exercise 3.3.1).
Exercises.
3.3.1. Complete the proof of Proposition 3.3.4.
3.4 Lie groups

Technically speaking, we gave the definition of an arbitrary Lie group (as opposed to a
matrix Lie group) in Section 1.1. But in this section we will give the definition more directly.
Although we will focus on matrix Lie groups in this course, it is useful to see the more general
definition, which is the approach one must take if one wishes to study Lie groups in more
detail (beyond this course). Our discussion here will be very brief. Further details can be
found in [Hal03, Appendix C].
Definition 3.4.1 (Lie group). A (real) Lie group is a (real) smooth manifold G which is
also a group and such that the group product G × G → G and the inverse map g 7→ g −1 are
smooth maps.
Roughly speaking, a smooth manifold is an object that looks locally like Rn . For example,
the torus is a two-dimensional manifold since it looks locally (but not globally) like R2 .
Example 3.4.2. Let

G = R × R × S1 ,
with group product
(x1 , y1 , u1 ) · (x2 , y2 , u2 ) = (x1 + x2 , y1 + y2 , eix1 y2 u1 u2 ).
One can check that this operation does indeed make G into a group (Exercise 3.4.1). The
multiplication and inversion maps are both smooth, and so G is a Lie group. However, one
can show that G is not isomorphic to a matrix Lie group (see [Hal03, §C.3]).
Theorem 3.4.3 ([Sti08, Corollary 2.33]). Every matrix Lie group is a smooth embedded
submanifold of Mn (C) and is thus a Lie group.
Usually one says that a map Φ between Lie groups is a Lie group homomorphism if Φ
is a group homomorphism and Φ is smooth. However, in Definition 2.11.1, we only required
that Φ be continuous. This is because of the following result. A proof for the case of matrix
Lie groups can be found in [Sti08, Corollary 2.34].
Proposition 3.4.4. Suppose G and H are Lie groups and Φ : G → H is a group homomor-
phism from G to H. If Φ is continuous, then it is also smooth.
Exercises.
3.4.1. Show that G, as defined in Example 3.4.2, is a group.
Chapter 4
Maximal tori and centres
From now on, we will sometimes use the term Lie group. For the purposes of this course,
you can replace this term by matrix Lie group. In this chapter we discuss some important
subgroups of Lie groups.
4.1 Maximal tori

Definition 4.1.1 (Torus). A torus (or k-dimensional torus) is a group isomorphic to
Tk = S1 × S1 × · · · × S1 (k-fold Cartesian product).
A maximal torus of a Lie group G is a torus subgroup T of G such that if T 0 is another torus
subgroup containing T then T = T 0 .
Remark 4.1.2. (a) Note that tori are abelian.
(b) For those who know something about Lie algebras, maximal tori correspond to Cartan
subalgebras of Lie algebras.
Example 4.1.3. The group SO(2) = S1 = T1 is its own maximal torus.
Recall that SO(3) is the group of rotations of R3 . Let e1 , e2 , e3 be the standard basis
vectors. Then the matrices
 
cos θ − sin θ 0
0 R θ 0
Rθ =  sin θ cos θ 0 = , θ ∈ R,
0 1
0 0 1
form a copy of T1 in SO(3).
Proposition 4.1.4. The set {Rθ0 | θ ∈ R} forms a maximal torus in SO(3).
Proof. Suppose T is a torus in SO(3) containing this T1 . Since all tori are abelian, any A ∈ T
commutes with all Rθ0 ∈ T1 . Thus is suffices to show that, for A ∈ SO(3),
ARθ0 = Rθ0 A for all Rθ0 ∈ T1 (4.1)
39
40 CHAPTER 4. MAXIMAL TORI AND CENTRES
implies that A ∈ T1 . Suppose A ∈ SO(3) satisfies (4.1). First we show that
A(e1 ), A(e2 ) ∈ Span{e1 , e2 }.
Suppose
A(e1 ) = a1 e1 + a2 e2 + a3 e3 .
By (4.1), A commutes with  
−1 0 0
Rπ0 =  0 −1 0 .
0 0 1
Since
ARπ0 (e1 ) = A(−e1 ) = −a1 e1 − a2 e2 − a3 e3 ,

Rπ0 A(e1 ) = Rπ0 (a1 e1 + a2 e2 + a3 e3 ) = −a1 e1 − a2 e2 + a3 e3 ,
we have a3 = 0 and so A(e1 ) ∈ Span{e1 , e2 } as desired. A similar argument shows that

A(e2 ) ∈ Span{e1 , e2 }.
Thus, the restriction of A to the (e1 , e2 )-plane is an isometry of that plane that fixes
the origin. Therefore it is a rotation or a reflection. However, no reflection commutes with
all rotations (Exercise 4.1.1). Thus A must be a rotation in the (e1 , e2 )-plane. Since A
preserves the dot product, it must leave invariant Re3 , which is the orthogonal complement
to Span{e1 , e2 }. So A is of the form

Rθ 0
A=
0 c
for some c ∈ R. Since det A = det Rθ = 1, it follows that c = 1, and hence A ∈ T1 as

desired.
Definition 4.1.5 (Centre). The centre of a Lie group G is the subgroup (see Exercise 4.1.2)
Z(G) = {A ∈ G | AB = BA ∀ B ∈ G}.
Corollary 4.1.6. The Lie group SO(3) has trivial centre:
Z(SO(3)) = {1}.
Proof. Suppose A ∈ Z(SO(3)). Then A commutes with all elements of SO(3) and hence all
elements of T1 . By the above argument, this implies that A fixes e3 . Interchanging basis
vectors in this argument shows that A also fixes e1 and e2 and hence A is the identity.
We now find maximal tori in all of the compact connected matrix Lie groups we have
seen. We focus on connected groups because one can easily see (since Tk is connected) that
a maximal torus must always be contained in the identity component of a Lie group.
First of all, we note that there are at least three natural matrix groups isomorphic to T1 :
(a) The group {Rθ | θ ∈ R} consisting of 2 × 2 matrices.

4.1. MAXIMAL TORI 41
(b) The group

eiθ = cos θ + i sin θ, θ ∈ R,
consisting of complex numbers (or 1 × 1 complex matrices).
(c) The group
eiθ = cos θ + i sin θ, θ ∈ R,
consisting of quaternions (or 1 × 1 quaternionic matrices).
These give rise to three natural matrix groups isomorphic to Tk :
(a) The group of 2k × 2k matrices of the form
 
Rθ1
Rθ1 ,...,θk = 
 .. .

.
Rθk
(b) The group of k × k unitary matrices of the form

 
eiθ1
Zθ1 ,...,θk = 
 ... .

eiθk
(c) The group of k × k symplectic matrices of the form

 
eiθ1
Qθ1 ,...,θk = 
 .. .

.
eiθk
Theorem 4.1.7. The following are maximal tori.

(a) In SO(2m):
Tm = {Rθ1 ,...,θm | θ1 , . . . , θm ∈ R}.
(b) In SO(2m + 1):

Rθ1 ,...,θm 0
m
T = Rθ0 1 ,...,θm = θ1 , . . . , θm ∈ R .
0 1
(c) In U(n):
Tn = {Zθ1 ,...,θn | θ1 , . . . , θn }.
(d) In SU(n):
Tn−1 = {Zθ1 ,...,θn | θ1 , . . . , θn ∈ R, θ1 + · · · + θn = 0}.
Note that this is a Tn−1 since is it equal to
  

 eiθ1 


 ..  

 . 
θ1 , . . . , θn−1 ∈R .
iθ
 

 e n−1  

−θ1 −···−θn−1
 
 e 
(e) In Sp(n):
Tn = {Qθ1 ,...,θn | θ1 , . . . , θn ∈ R}.
Proof. We prove each case separately. As for the case of SO(3) dealt with above, in each case
we will show that if an element A of the group commutes with all elements of the indicated
torus, then A must be included in this torus. Then the result follows as it did for SO(3).
(a) Let e1 , e2 , . . . , e2m denote the standard basis of R2m . Suppose that A ∈ SO(2m)
commutes will all elements of Tm . We first show that
A(e1 ), A(e2 ) ∈ Span{e1 , e2 },
A(e3 ), A(e4 ) ∈ Span{e3 , e4 },
..
.
A(e2m−1 ), A(e2m ) ∈ Span{e2m−1 , e2m }.
We do the case of e1 , since the rest are analogous. Recall that we assume that A commutes
with all elements of Tm . Therefore,
ARπ,0,...,0 (e1 ) = Rπ,0,...,0 A(e1 ).
If
A(e1 ) = a1 e1 + · · · + a2m e2m ,
then
ARπ,0,...,0 (e1 ) = A(−e1 ) = −a1 e1 − · · · − a2m e2m ,
but
Rπ,0,...,0 A(e1 ) = Rπ,0,...,0 (a1 e1 + · · · + a2m e2m ) = −a1 e1 − a2 e2 + a3 e3 + · · · + e2m e2m .
Thus
a3 = a4 = · · · = a2m
as desired. Therefore, A is a product of rotations or reflections in the planes
Span{e1 , e2 }, . . . , Span{e2m−1 , e2m }.
However, the case of reflections is ruled out as it was for SO(3) and hence A is a product of
rotations in these planes and therefore is an element of Tm .
0
(b) This is a generalization of the argument for SO(3), using maps such as Rπ,0,...,0 in
0
place of Rπ . The details are left as an exercise.
(c) Let e1 , . . . , en be the standard basis of Cn . Suppose that A ∈ U(n) commutes with
all elements of Tn . In particular, A commutes with Zπ,0,...,0 . Let
A(e1 ) = a1 e1 + · · · + an en .
Then
AZπ,0,...,0 (e1 ) = A(−e1 ) = −a1 e1 − · · · − an en ,
Zπ,0,...,0 A(e1 ) = Zπ,0,...,0 (a1 e1 + · · · + an en ) = −a1 e1 + a2 e2 + · · · + an en .
Thus a2 = · · · = an = 0 and so A(e1 ) = c1 e1 for some c1 ∈ C. A similar argument shows
that A(ek ) = ck ek for each k. Since A is unitary, we must have |ck | = 1 for each k and thus
A ∈ Tn .
4.1. MAXIMAL TORI 43
(d) The case n = 1 is trivial and so we begin with the case n = 2. Suppose that

a b
A=
c d
commutes will all elements of T1 . Then A commutes with

i 0
Zπ/2,−π/2 = ,
0 −i
and thus
ai −bi ai bi
= AZπ/2,−π/2 = Zπ/2,−π/2 A = .
ci −di −ci −di
Therefore b = c = 0. Then we must have |a| = |d| = 1 since A is unitary and a = d−1 since
A has determinant one. Therefore A ∈ T1 . For n > 2, we leave it as an exercise to show
that if A ∈ SU(n) commutes with Zπ,π,0,...,0 and Zπ,0,π,0,...,0 then Ae1 = c1 e1 for some c1 ∈ C.
(Note that we could not use Zπ,0,...,0 as we did for the U(n) case since det Zπ,0,...,0 = −1.)
Similarly, one can show that Aek = ck ek for each k. Then the result follows as it did for the
U(n) case.
(e) Suppose A ∈ Sp(n) commutes with all elements of Tn . By the argument used for
U(n), one can show that for all k, Aek = ck ek for some ck ∈ H. But then the fact that A
must commute with all elements of Tn shows that each ck must commute with all eiθ , θ ∈ R.
In particular, ck must commute with i and thus ck ∈ R + Ri. We must also have |ck | = 1
since A ∈ Sp(n) and so A ∈ Tn .
Remark 4.1.8. (a) Note that in our proofs, we only used the fact that A commuted with
all elements of the torus in question. Thus it follows that these maximal tori are also
maximal abelian subgroups of the respective Lie groups.
(b) Since Tk is connected, any maximal torus must lie in the identity component of the Lie
group. Therefore, the maximal tori of SO(2m) and SO(2m + 1) found above are also
maximal tori of O(2m) and O(2m + 1).
Remark 4.1.9. A noncompact Lie group need not have any nontrivial tori (for example, Rn ).
However, the maximal tori in compact groups are well-behaved. In particular, we have the
following results which we state without proof (at least for now) because the proofs use
material we have not covered. They are useful facts that motivate some of the terminology
we will use. Let G be a compact Lie group and T a maximal torus.
(a) Every conjugate of a maximal torus is a maximal torus.
(b) Every element g ∈ G is conjugate to an element of T , hence lies in a maximal torus.
(c) All maximal tori in G are conjugate. So the maximal tori form a single conjugacy class
in G.
(d) It follows that all maximal tori in G have the same dimension, called the rank of G.
Exercises.
4.1.1. Show that no reflection of the plane commutes with all rotations.
4.1.2. Show that Z(G) is indeed a subgroup of G. In fact, show that it is a Lie subgroup of
G.
4.2 Centres
We now determine the centres of the compact classical groups SO(n), U(n), SU(n) and Sp(n).
Any element of the centre of a Lie groups commutes will all elements of the Lie group and
thus with all elements of a maximal torus. Therefore, by the arguments of the previous
section, the centres are contained in the maximal tori we described there. Note also that
any scalar multiple of the identity n × n matrix commutes will all n × n matrices.
Theorem 4.2.1. The compact classical groups have the following centres.
(a) Z(SO(2)) = SO(2), Z(SO(2m)) = {±1}, m ≥ 2.
(b) Z(SO(2m + 1)) = {1}, m ≥ 1.
(c) Z(U(n)) = {ω1 | |ω| = 1}, n ≥ 1.
(d) Z(SU(n)) = {ω1 | ω n = 1}, n ≥ 1.
(e) Z(Sp(n)) = {±1}, n ≥ 1.
Here 1 = I is the identity matrix.
Proof. All of the subgroups mentioned in the proposition consist of multiples of the identity
matrix. Therefore, by the above comment, they are contained in the centres. It thus suffices
to show that all elements of the centres lie in these subgroups.
(a) We know that SO(2) is abelian and so the first statement follows. Suppose A ∈
Z(SO(2m)), m ≥ 2. As noted, we can assume that A lies in the maximal torus Tm . Thus
A = Rθ1 ,...,θm for some θ1 , . . . , θm ∈ R. Let

1 0
M= .
0 −1
Then Rθ commutes with M if and only if

cos θ sin θ cos θ − sin θ cos θ − sin θ cos θ − sin θ
= M =M = .
sin θ − cos θ sin θ cos θ sin θ cos θ − sin θ − cos θ
So Rθ commutes with M if and only if sin θ = 0, which implies cos θ = ±1.

4.3. DISCRETE SUBGROUPS AND THE CENTRE 45
Forming a matrix with (an even number of) copies of M on the diagonal (and all other
entries zero), we see that A will commute with such matrices if and only if each sin θk = 0
and cos θk = ±1. Therefore, A is a diagonal matrix with diagonal entries ±1. Suppose that
both +1 and −1 occur. Then let B be the matrix with Rθ (with sin θ 6= 0) on the diagonal
at the position of an adjacent +1 and −1 (and otherwise only 1’s on the diagonal). Then A
does not commute with B. Therefore A = I or A = −I as desired.
(b) We follow the argument for SO(2m). However, since −I 6= SO(2m + 1), we conclude
that A = I.
(c) If n = 1, then U(n) is isomorphic to S1 = {eiθ | θ ∈ R}, which is abelian and thus
U(1) is its own centre. Therefore we n ≥ 2. Let A ∈ Z(U(n)). Then A = Zθ1 ,...,θn
assume
0 i
for some θ1 , . . . , θn ∈ R. Note that ∈ U(2) (in fact, it is in SU(2), something we will
i 0
use in the next part) and
iθ iθ
0 ieiθ1 e 1 0 0 i 0 i e 1 0 0 ieiθ2
= = =
ieiθ2 0 0 eiθ2 i 0 i 0 0 eiθ2 ieiθ1 0
if and only ifeiθ1

= eiθ2 . Therefore, since A must commute with all matrices obtained
0 i
by placing a somewhere on the diagonal, 1’s elsewhere on the diagonal, and zeros
i 0
everywhere else, we see that eiθ1 = · · · = eiθn . Therefore A = eiθ1 I as desired.
(d) Let A ∈ Z(SU(n)). The argument for U(n) shows that A = ωI for some ω ∈ C,
|w| = 1. We must also have
1 = det A = ω n ,
as claimed.
0 1 0 i
Note: [Sti08] uses the matrix instead of for the proof of the U(n) and
1 0 i 0
SU(n) cases. While this works for U(n), it is a problem for SU(n) since this matrix does not
have determinant one.
(e) Let A ∈ Z(Sp(n)). An argument similar to the one used for U(n) shows that A = qI
for some q ∈ H with |q| = 1. However, since this matrix must commute with all elements of
Sp(n), it must, in particular, commute with I, iI, jI and kI. Thus, q must commute with
all quaternions. But the only quaternions that commute with all other quaternions are the
real numbers. Thus q ∈ R. But then |q| = 1 implies that q = ±1 as desired.
4.3 Discrete subgroups and the centre

Recall that a subgroup H of a group G is normal if
gHg −1 = H for all g ∈ G,
and G is simple if it has no nontrivial (i.e. not equal to {1} or G) normal subgroups. It is
easy to see that the centre Z(G) of a group G is a normal subgroup of G. Thus if G has
nontrivial centre, it is not simple. Therefore, by our computation of centres in the previous
section, the groups SO(2m), U(n), SU(n) and Sp(n) are not simple. This leaves SO(2m + 1)
as the only possibility.
Given any two matrices A, B ∈ Mn (C), we can consider the distance between them as
2
points in Cn . Thus, the distance d(A, B) between A and B is
s X
|aij − bij |2 .
1≤i,j≤n
Definition 4.3.1 (Discrete subgroup). A subgroup H of a Lie group G is called discrete if

there is a positive lower bound to the distance between any two distinct elements of H. In
other words H is discrete if there exists a c > 0 such that
d(A, B) > c for all A, B ∈ H, A 6= B.
It follows immediately that any finite subgroup of G is discrete – just take c to be the
minimum of the (finite number of) distances between elements of H. Thus, the centres of
SO(n), SU(n), and Sp(n) are discrete. However, the centre of U(n) is not.
Theorem 4.3.2 (Schreier’s Theorem). If G is a path-connected Lie group then any discrete
normal subgroup is contained in the centre of G.
Proof. Let H be a discrete normal subgroup of G. Then
BAB −1 ∈ H for all A ∈ H, B ∈ G.
Fix A ∈ H. Then we have a continuous map
ϕ : G → H, ϕ(B) = BAB −1 .
Note that ϕ(1) = A. Let C be an arbitrary element of G. Since G is path-connected, there

is a path from 1 to C. Since ϕ is continuous, it maps this path to a path in H. But H is
discrete and so the image must be the single point A. In other words ϕB = BAB −1 = A or
BA = AB for all B ∈ G. Therefore A ∈ Z(G). So H ⊆ Z(G).
We have seen that SO(n), SU(n), and Sp(n) are all path-connected. Therefore, their dis-
crete normal subgroups are contained in their centres, which we have computed. Conversely,
any subgroup of the centre is normal and so, in fact, we now know all the discrete normal
subgroups of these Lie groups. In particular, since Z(SO(2m + 1)) = {1}, the Lie group
SO(2m + 1) has no nontrivial discrete normal subgroups.
We will see later that there are no normal subgroups of these Lie groups that are not
discrete. In particular, it will follow that SO(2m+1) is simple and that the others are simple
modulo their centres (i.e. G/Z(G) is simple).
Chapter 5
Lie algebras and the exponential map
One of the most important methods for studying Lie groups is to study their associated Lie
algebras. In this chapter we will investigate the notion of a Lie algebra and the connection
between Lie algebras and Lie groups.
5.1 The exponential map onto SO(2)
Recall the power series expansions
x x2 x3
ex = 1 + + + + ...,
1! 2! 3!
x 2 x4
cos x = 1 − + − ...,
2! 4!
x x3
sin x = − + ....
1! 3!
Since these series are absolutely convergent for all x, we can substitute iθ for x and rearrange
terms to conclude that
eiθ = cos θ + i sin θ.
Therefore the exponential function maps the imaginary axis Ri onto the circle S1 of points
of absolute value one in the plane of complex numbers. We can view this imaginary axis as
the tangent space to the identity of the Lie group S1 .
47
48 CHAPTER 5. LIE ALGEBRAS AND THE EXPONENTIAL MAP
1 + Ri
C i
x
1
While the tangent space is naturally the set 1 + Ri, we are only interested in coordinates
relative to the point of tangency (i.e. we treat this point as the origin), and these relative
coordinates are of the form Ri.
Note that the point iθ of height θ is mapped to the cos θ + i sin θ at arc length θ and so
the exponential map preserves the length of sufficiently small arcs.
5.2 The exponential map onto SU(2)

For any pure imaginary quaternion v, we can define the usual exponential series
v v2 v3
ev = 1 + + + + ··· .
1! 2! 3!
For sufficiently large n, |v|n /n! < 2−n , and so this series is absolutely convergent in H (for
the same reason as in C).
Theorem 5.2.1 (Exponentiation theorem for H). Let u be a unit vector in Ri + Rj + Rk

(i.e. a unit pure imaginary quaterion) and θ ∈ R. Then
eθu = cos θ + u sin θ
and the exponential function maps Ri + Rj + Rk onto S3 = SU(2).
Proof. Recall that if u is a unit vector in Ri + Rj + Rk, then u2 = −1. Therefore
θu θ2 θ3 u
eθu = 1 + − − + ···
1! 2! 3!
5.3. THE TANGENT SPACE OF SU(2) 49
θ2 θ4 θ3 θ5

θ
= 1− + − ··· + u − + − ···
2! 4! 1! 3! 5!
= cos θ + u sin θ.
Now, an arbitrary point of S3 (viewed as the set of unit quaternions) can be written as
a + bi + cj + dk = a + u|v|,
√
where u is a unit vector parallel to v = bi + cj + dk and |v| = b2 + c2 + d2 . Then
a2 + |v|2 = a2 + b2 + c2 + d2 = 1,
since we considered a unit quaternion. Therefore, there is a real number θ such that
a = cos θ, |v| = sin θ.
Thus any point of S3 is of the form cos θ+u sin θ for some unit vector u and so the exponential
map above is onto.
5.3 The tangent space of SU(2)

We will see that Ri + Rj + Rk is the tangent space to SU(2), just like Ri is the tangent space
to S1 . We would like to find a way to compute the tangent space of a matrix Lie group in
general. The key idea is that one should think of tangent vectors as “velocity vectors” of
paths (or smoothly moving points) in our space (in this case, in our Lie group). This is same
idea used in many areas of geometry: algebraic geometry, differential geometry, etc.
Suppose q(t) is some smooth path in SU(2) passing through the identity at t = 0. So
q(t) is defined in some neighbourhood of zero (i.e. some interval containing (−, ) for some
> 0), takes values in SU(2) with q(0) = 1, and is differentiable.
Then the “velocity” q 0 (0) at t = 0 is a tangent vector to SU(2) at the identity 1. Con-
versely, all tangent vectors arise in this way.
Since q(t) is a unit quaternion for all t in the domain of q, we have
q(t)q(t) = 1.
Differentiating, we get
q 0 (t)q(t) + q(t)q 0 (t) = 0. (5.1)
Of course, we just blindly applied the product rule to a function taking quaternionic
values. We should justify this. We simply follow the usual proof of the product rule:
d q(a + h)q(a + h) − q(a)q(a)

q(t)q(t) = lim
dt t=a
h→0 h
(q(a + h) − q(a))q(a + h) + q(a)(q(a + h) − q(a))
=
h
= q 0 (a)q(a) + q(a)q 0 (a).
Setting t = 0 in (5.1), we obtain

q 0 (0) + q 0 (0) = 0.
This implies that q 0 (0) is a pure imaginary quaternion. Therefore, every tangent vector to
SU(2) at 1 is a pure imaginary quaternion.
Conversely, suppose that p is a pure imaginary quaternion. Then
pt ∈ Ri + Rj + Rk, for all t ∈ R,
and hence q(t) = ept is a path in SU(2) by Theorem 5.2.1. We have q(0) = 1, and q(t) is
smooth since
q 0 (t) = pept ,
as can be seen from differentiating the power series for ept . Also,
q 0 (0) = pe0 = p · 1 = p,
and thus p is a tangent vector to SU(2) at 1. Therefore the tangent space of SU(2) at 1 is
Ri + Rj + Rk.
The Lie bracket. Recall that, for x, y ∈ Ri, we have ex ey = ex+y . Since the exponential
map sends Ri onto SO(2), this means that we can compute the product of elements on SO(2)
using addition in Ri.
The case of SU(2) is different. Since addition Ri + Rj + Rk is commutative, it cannot
describe the noncommutative product on SU(2). What we need is some operation on the tan-
gent space that “measures” this noncommutativity. This operation cannot be multiplication
of quaternions since Ri + Rj + Rk is not closed under multiplication.
Let U and V be two tangent vectors to SU(2) at 1 and let u(s) and v(t) be paths in
SU(2) such that u(0) = v(0) = 1 and u0 (0) = U , v 0 (0) = V . Note that
ws (t) = u(s)v(t)u(s)−1
gives us information about how the elements u(s) and v(s) commute (for instance, this equals
v(t) if and only if they commute). Note that the above equation defines a path (we think of
s as being fixed and t as varying). We have
ws (0) = u(s)v(0)u(s)−1 = 1,
and
ws0 (0) = u(s)v 0 (0)u(s)−1 = u(s)V u(s)−1
(use the definition of the derivative to see this).
Now, since ws (t) is a path through 1 for each s, ws0 (0) is a tangent vector at 1 for each s.
Therefore
x(s) = u(s)V u(s)−1
is a smooth path in Ri + Rj + Rk. Therefore, x0 (0) is also an element of Ri + Rj + Rk and
d
x0 (0) = u(s)V u(s)−1 = u0 (0)V u(0)−1 + u(0)V (−u0 (0)) = U V − V U.
ds s=0
5.4. THE LIE ALGEBRA SU(2) OF SU(2) 51
Here we have used the fact that d

ds s=0
u(s)−1 = −u0 (0) if u(0) = 1 (see [Sti08, Exer-
cise 4.3.2]).
We see that the map
(U, V ) 7→ [U, V ]
is a binary operation on the tangent space Ri + Rj + Rk to SU(2) at the identity, which is
related to the (non)commutativity of SU(2).
5.4 The Lie algebra su(2) of SU(2)

Definition 5.4.1 (Lie algebra). A Lie algebra is a vector space g with a bilinear operation
(the Lie bracket) [·, ·] : g × g → g that
• is alternating: [X, X] = 0 for all X ∈ g (equivalently, if the field is not of characteristic

2, [X, Y ] = −[Y, X] for all X, Y ∈ g), and
• satisfies the Jacobi identity
[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0.
If g1 and g2 are Lie algebras, a homomorphism (resp. isomorphism) of Lie algebras is a

linear map (resp. vector space isomorphism) ϕ : g1 → g2 that commutes with the Lie bracket:
ϕ([X, Y ]1 ) = [ϕ(X), ϕ(Y )]2
where [·, ·]i denotes the Lie bracket on [·, ·]i , i = 1, 2.
Definition 5.4.2 (su(2)). It follows from Exercise 5.4.1 that Ri + Rj + Rk, with Lie bracket
given by [U, V ] = U V − V U (i.e. the tangent space to SU(2) at the identity), is a Lie algebra.
We denote this Lie algebra by su(2).
One can show that R3 is a Lie algebra with bracket equal to the cross-product (see [Sti08,
Exercises 1.4.1 and 1.4.4]).
Definition 5.4.3 (Ideal of a Lie algebra). An ideal of a Lie algebra g is a subspace a of g

with the property that [X, Y ] ∈ a for all X ∈ g and Y ∈ a.
The proof of the following lemma is left as an exercise (Exercise 5.4.2).
Lemma 5.4.4. If ϕ : g1 → g2 is a homomorphism of Lie algebras, then ker ϕ is an ideal of

g1 .
The following notion will be useful for us later.
Definition 5.4.5 (Simple Lie algebra). A Lie algebra is simple if it has dimension at least
two and it has no proper nonzero ideals.
Exercises.
5.4.1. Show that any vector space of matrices closed under matrix multiplication (more
generally, any associative algebra) is a Lie algebra with bracket given by [A, B] = AB − BA.
5.4.3. Show that Ri + Rj + Rk (with Lie bracket as in Definition 5.4.2) is isomorphic to R3

with the cross-product. Note that the isomorphism is not given by the “identity” map.
5.5 The exponential of square matrices

2 2
Recall that we identify Mn (C) with Cn or R2n and thus we have a matrix absolute value
sX
kAk = |aij |2 , A = (aij ).
i,j
This is the distance from the origin to the point A. Similarly, if A ∈ Mn (H), then kAk is
2
the distance from the origin to A in R4n (where we identify H with R4 ).
If A, A1 , A2 , . . . are n × n matrices, we say
lim Am = A
m→∞
if
∀ ∃ M such that m > M =⇒ kAm − Ak < .
Equivalently, limm→∞ Am = A if the limits of the entries of the Am exists and equal the
corresponding entries of A.
Proposition 5.5.1 (Submultiplicative property). If A and B are n × n real matrices, then

kABk ≤ kAkkBk.
Proof. Let A = (aij ) and B = (bij ) and C = AB = (cij ). Then
|cij | = |ai1 b1j + ai2 b2j + · · · + ain bnj |

= |(ai1 , ai2 , . . . , ain ) · (b1j , b2j , . . . , bnj )|
≤ |(ai1 , ai2 , . . . , ain )||(b1j , b2j , . . . , bnj )| (Cauchy-Schwartz Inequality).
Thus
X
kABk2 = |cij |2
i,j
X
≤ (|ai1 |2 + · · · + |ain |2 )(|bi1 |2 + · · · + |bin |2 )
i,j
5.5. THE EXPONENTIAL OF SQUARE MATRICES 53
X X
= (|ai1 |2 + · · · + |ain |2 ) (|bi1 |2 + · · · + |bin |2 )
i j
2 2
= kAk kBk .
Therefore kABk ≤ kAkkBk.
Proposition 5.5.2 (Convergence of the matrix exponential). If A is any real n × n matrix,

then
A A2 A3
In + + + + ··· ,
1! 2! 3!
2
is absolutely convergent in Rn and hence defines a function of A. This function is continu-
ous.
Proof. The series

kAk kA2 k kA3 k
kIn k + + + + ···
1! 2! 3!
is a series of positive real numbers, whose terms (except for the first) are less than or equal
to the corresponding terms of
kAk kAk2 kAk3

1+ + + + ···
1! 2! 3!
by the submultiplicative property. This is the power series for ekAk and thus it converges.
Therefore the series in the proposition converges absolutely.
Since each Am is continuous, the partial sums of the series are continuous. Since the
series converges uniformly (exercise) on every set of the form {A | kAk ≤ R}, R > 0, the
sum is continuous.
Remark 5.5.3. Recall that complex numbers can be viewed as 2 × 2 real matrices, and one
can check that if Z is the matrix corresponding to the complex number z, then eZ is the
matrix corresponding to the complex number ez . Thus, the exponential of a complex matrix
may be represented by the exponential of a real matrix and so the series above also converges
for complex matrices. Similarly, it converges for quaternionic matrices.
Definition 5.5.4 (Exponential of a matrix). The exponential of an n × n (real, complex, or

quaternionic) matrix A is
A A2 A3
exp A = eA = In + + + + ··· .
1! 2! 3!
Remark 5.5.5. One can define an exponential in the more general setting of Riemannian
manifolds, where it maps lines (through the origin) in the tangent space to geodesics through
the tangent point. One can also define an exponential for arbitrary Lie groups by considering
one-parameter subgroups.
Proposition 5.5.6 (Properties of the matrix exponential). Suppose X and Y are arbitrary
n × n (real, complex, or quaternionic) matrices. Then we have the following.
(a) e0 = I.
∗
(b) (eX )∗ = eX .
(c) eX is invertible and (eX )−1 = e−X .
(d) e(α+β)X = eαX eβX for all α and β in C.
(e) If XY = Y X, then eX+Y = eX eY = eY eX .

−1
(f ) If C is invertible, then eCXC = CeX C −1 .
(g) keX k ≤ ekXk .
Proof. Part (a) is obvious. Part (b) follows by taking term-by-term adjoints (the adjoint of
a sum is the sum of adjoints and (X m )∗ = (X ∗ )m ). Parts (c) and (d) are special cases of
part (e). For part (e), we compute
X2 Y2

X Y
e e = I +X + + ··· I +Y + + ···
2! 2!
∞ X m ∞ m
X X k Y m−k X 1 X m!
= = X k Y m−k .
m=0 k=0
k! (m − k)! m=0
m! k=0
k!(m − k)!
Now, when X and Y commute, we have

m
m
X m!
(X + Y ) = X k Y m−k ,
k=0
k!(m − k)!
and so ∞
X 1
eX eY = (X + Y )m = eX+Y .
m=0
m!
Part (f) follows from the fact that (CXC −1 )m = CX m C −1 and part (g) follows from our
proof of the convergence of the exponential of matrices.
Proposition 5.5.7. Let X be a square complex matrix. Then t 7→ etX is a smooth curve in
Mn (C) and
d tX
e = XetX = etX X.
dt
In particular,
d tX
e = X.
dt t=0
Proof. Since for each i, j, the entry (etX )ij is given by a convergent power series in t, we can
differentiate the series for etX term by term. The result follows easily.
At some points, we may wish to explicitly compute the exponential of a matrix X. There
is a general method for doing this.
5.6. THE TANGENT SPACE 55
Case 1: X is diagonalizable. If X is diagonalizable, then there exists an invertible

matrix C such that X = CDC −1 with
 
λ1 0
D=
 ... .

0 λn
It is easy to check that  

eλ1 0
eD = 
 ... ,

0 eλn
and so, by Proposition 5.5.6, we have
 
eλ1 0
eX = C  0
 ...  −1
C ,
0 eλn
Case 2: X is nilpotent. A matrix X is nilpotent if X m = 0 for some m > 0 (hence

X ` = 0 for all ` > m). In this case the series for eX terminates after the first m terms and
so can be computed explicitly. For example (an example related to the Heisenberg group),
if  
0 a b
X = 0 0 c  ,
0 0 0
then  
0 0 ac
X 2 = 0 0 0  ,
0 0 0
and X 3 = 0. Thus  
1 a b + ac/2
eX = 0 1 c .
0 0 1
General case: X arbitrary. By a theorem in linear algebra, every matrix X can be

written uniquely in the form X = S +N , with S diagonalizable, N nilpotent, and SN = N S.
Therefore,
eX = eS+N = eS eN ,
and we can compute eS and eN as in the previous cases.
5.6 The tangent space

Definition 5.6.1 (Smooth paths and their derivatives). Let S be a space of matrices. We
say that a path t 7→ A(t) is smooth, or differentiable, if the coordinate functions aij (t) are
differentiable. If A(t) is smooth, its derivative A0 (t) is defined in the usual way by
A(t + ∆t) − A(t)
lim .
∆t→0 ∆t
Remark 5.6.2. It follows from the definition that A0 (t) is just the matrix with entries a0ij (t)
where aij (t) are the entries of A(t).
Definition 5.6.3 (Tangent vectors and the tangent space). The tangent vectors at the
identity 1 of a matrix Lie group G are the matrices X of the form
X = A0 (0)
where A(t) is a smooth path in G with A(0) = 1. The tangent space to G at 1 is the set of
all tangent vectors at 1.
Remark 5.6.4. We will see later (Proposition 5.10.1) that if X is a tangent vector at the
identity to a matrix Lie group G, then eX ∈ G. It follows that an equivalent definition of
the tangent space is the set of all X such that etX ∈ G for all t ∈ R. (See Corollary 5.11.10
and Remark 5.11.11.) In fact, this is the definition used in [Hal03] and it will be useful to
sometimes use this characterization when computing tangent spaces.
Proposition 5.6.5 (Tangent spaces of O(n), U(n) and Sp(n)). The tangent spaces at the
identity of the groups O(n), U(n) and Sp(n) are given as follows.
(a) O(n): {X ∈ Mn (R) | X + X T = 0}.
(b) U(n): {X ∈ Mn (C) | X + X ∗ = 0}.
(c) Sp(n): {X ∈ Mn (H) | X + X ∗ = 0}.
Proof. Let A(t) be a smooth path in O(n), U(n) or Sp(n) with A(0) = 1. Then
A(t)A(t)∗ = 1 for all t
(note that M ∗ = M T for M ∈ Mn (R)). Taking the derivative with respect to t, we obtain
A0 (t)A(t)∗ + A(t)A0 (t)∗ = 0.
(There are a few technicalities here,

which are left as an exercise. For instance, one should
d T d T d
check that dt (M ) = ( dt M ) , dt A(t) = A0 (t) – see Exercise 5.6.1.) Since A(0) = 1, we
have
A0 (0) + A0 (0)∗ = 0.
It follows that the tangent spaces are contained in the sets in the statement of the proposition.
Now suppose that a matrix X satisfies X + X ∗ = 0. Then etX is a path in the corre-
sponding group since
∗
etX (etX )∗ = etX etX = etX e−tX = et(X−X) = e0 = I.
d
Since dt t=0
etX = X, we see that X is in the tangent space.
5.6. THE TANGENT SPACE 57
Definition 5.6.6 (Skew-symmetric and skew-Hermitian). If X ∈ Mn (R), we say X is skew-

symmetric if X + X T = 0. If X ∈ Mn (C) or X ∈ Mn (H), we say X is skew-Hermitian if
X + X ∗ = 0.
Remark 5.6.7. From now on, when we refer to a tangent vector to a (matrix) Lie group G, we
mean a tangent vector at the identity and by the tangent space of G, we mean the tangent
space at the identity.
Proposition 5.6.8 (Tangent space of SO(n)). The tangent space of SO(n) is the same as
the tangent space of O(n), namely the space of real skew-symmetric matrices:
{X ∈ Mn (R) | X + X T = 0}.
Proof. Since all paths in SO(n) are also paths in O(n), all tangent vectors to SO(n) are
also tangent vectors to O(n) and hence are skew-symmetric. Conversely, let X be a skew-
symmetric matrix. Then t 7→ etX is a path in O(n) with tangent vector X. Now, when
t = 0, we have etX = e0 = I, which has determinant one. Since all matrices in O(n)
have determinant ±1 and etX is a continuous path, we must have det etX = 1 for all real t.
Therefore, t 7→ etX is actually a path in SO(n) and hence X is a tangent vector to SO(n).
We cannot use the same type of argument we used for SO(n) to find the tangent space to
SU(n) because elements of U(n) can have determinant equal to any unit complex number.
To determine the tangent space of SU(n), we need the following result.
Proposition 5.6.9 (Determinant of the exponential). For any square complex matrix A,
det eA = etr A .
Proof. Suppose the result is true for upper triangular matrices. Then for an arbitrary com-
plex matrix A, we can find an invertible complex matrix B and an upper triangular complex
matrix T such that A = BT B −1 . Then
−1 −1 )
det eA = det(eBT B ) = det(BeT B −1 ) = det eT = etr T = etr(BT B = etr A .
Thus, it suffices to show the result for an upper triangular matrix T . Suppose
 
t11 ∗ ∗ ··· ∗
 0 t22 ∗ · · · ∗ 
 
T =  0 0 t33 · · · ∗  .
 
 .. .. 
 . . 
0 0 · · · 0 tnn
Then for k > 0, T k is upper triangular with ith diagonal entry equal to tkii . Thus eT is upper
triangular with ith diagonal entry equal to etii . Therefore,
det eT = et11 et22 · · · etnn = et11 +t22 +···+tnn = etr T .
Proposition 5.6.10 (Tangent space of SU(n)). The tangent space of SU(n) is
{X ∈ Mn (C) | X + X ∗ = 0, tr X = 0}.
Proof. Since any path in SU(n) is also a path in U(n), the tangent space to SU(n) is included
in the tangent space to U(n). By Remark 5.6.4, it suffices to consider paths of the form
t → etX in order to compute the tangent space of SU(n). Note that t 7→ etX is a path in
SU(n) if and only if
det etX = 1 ∀ t ⇐⇒ et tr X = 1 ∀ t ⇐⇒ t tr X = 0 ∀ t ⇐⇒ tr X = 0.
The result follows.
Proposition 5.6.11 (Tangent space of GL(n, C) and SL(n, C)). The tangent space of GL(n, C)
is Mn (C) and the tangent space of SL(n, C) is
{X ∈ Mn (C) | tr X = 0}.
Proof. The tangent space of GL(n, C) is contained in Mn (C) by definition. Furthermore, for
X ∈ Mn (C), t 7→ etX is a path in GL(n, C) (recall that etX is always invertible) with tangent
vector X. Hence the tangent space to GL(n, C) is all of Mn (C).
The statement for SL(n, C) follows from Proposition 5.6.9. The details are left as an
exercise (Exercise 5.6.3).
Proposition 5.6.12 (Some other tangent spaces). We have the following tangent spaces
(where J = J2n is the matrix used to define the symplectic groups):
(a) O(n, C), SO(n, C): {X ∈ Mn (C) | X + X T = 0}.
(b) Sp(n, R): {X ∈ Mn (R) | JX T J = X}.
(c) Sp(n, C): {X ∈ Mn (C) | JX T J = X}.
(d) Heisenberg group H: {X ∈ M3 (R) | X is strictly upper triangular}.
Exercises.

d
5.6.1. Suppose A(t) is a smooth path in Mn (C). Prove that dt
(AT ) = ( dtd A)T and d
dt
A(t) =
A0 (t).
5.6.2. Show that the exponential map exp : u(n) → U (n) is surjective. Hint: Recall that if
A is a unitary matrix, then there is a unitary matrix U and a diagonal matrix D such that
A = U DU ∗ . (The proof of this fact is similar to the proof of the analogous statement for
symmetric matrices given in Section 3.2.)
5.6.3. Complete the proof of Proposition 5.6.11.

5.7. THE TANGENT SPACE AS A LIE ALGEBRA 59
5.7 The tangent space as a Lie algebra

The tangent space has interesting structure – it is a Lie algebra – reflecting the structure of
the Lie group itself. To prove this, we need to define a Lie bracket and then show that the
tangent space satisfies the axioms of a Lie algebra.
The first property we need to verify is that the tangent space is actually a vector space.
We write T1 (G) for the tangent space of G at the identity.
Proposition 5.7.1 (Tangent spaces are vector spaces). For any Lie group G, T1 (G) is a
real vector space.
Proof. Suppose X, Y ∈ T1 (G). Then there exists paths A(t) and B(t) in G such that
A(0) = B(0) = 1, A0 (0) = X and B 0 (0) = Y . Then C(t) = A(t)B(t) is also a smooth path
in G (see Exercise 5.7.1) with C(0) = 1. Thus C 0 (0) ∈ T1 (G). Now
d
C 0 (0) = A(t)B(t) = A0 (0)B(0) + A(0)B 0 (0) = X + Y.
dt t=0
Therefore X + Y ∈ T1 (G) and the tangent space is closed under vector addition.
Now let r ∈ R. Since D(t) = A(rt) is a smooth path in G and D(0) = A(0) = 1, we have
0
D (0) ∈ T1 (G). Since
D0 (0) = rA0 (0) = rX,
we see that rX ∈ T1 (G). Therefore T1 (G) is also closed under scalar multiplication and
hence is a vector space.
It is straightforward to verify that Mn (C) is a Lie algebra under the Lie bracket given
by [A, B] = AB − BA (see Exercise 5.4.1). Therefore, in order to show that T1 (G) is a Lie
algebra, it now suffices to show that it is closed under this Lie bracket (since then it is a Lie
subalgebra of Mn (C)).
Proposition 5.7.2 (Tangent spaces are closed under the Lie bracket). For any Lie group
G, the tangent space T1 (G) is closed under the Lie bracket.
Proof. Suppose X, Y ∈ T1 (G). Then there exists paths A(t) and B(t) in G with A(0) =
B(0) = 1, A0 (0) = X, and B 0 (0) = Y .
For a fixed s ∈ R, consider the path
Cs (t) = A(s)B(t)A(s)−1 .
Then Cs (t) is smooth and Cs (0) = 1 and so Cs0 (0) ∈ T1 (G). Now,
Cs0 (0) = A(s)B 0 (0)A(s)−1 = A(s)T A(s)−1
is a smooth function of s since A(s) is. Therefore D(s) = A(s)Y A(s)−1 defines a smooth
path in T1 (G) and so its tangent vector at s = 0 must also lie in T1 (G) (the tangent of a
path in some vector space is an element of that vector space). Now,
D0 (0) = A0 (0)Y A(0)−1 + A(0)Y (−A0 (0)) = XY − Y X = [X, Y ],
where we have used A0 (0) = X, A(0) = 1 and ds d
s=0
A(s) = −A0 (0) when A(0) = 1 (see
[Sti08, Exercise 4.3.2]). Hence [X, Y ] ∈ T1 (G) as desired.
Definition 5.7.3 (Lie algebra of a Lie group). The Lie algebra g of a Lie group G is the
tangent space T1 (G) with the Lie bracket [X, Y ] = XY − Y X. We denote the Lie algebra of
a particular Lie group by using lowercase gothic (Fraktur) letters. For instance, gl(n, C) is
the Lie algebra of GL(n, C), so(n) is the Lie algebra of SO(n), etc. (One exception to this
rule is that we will not assume that h is the Lie algebra of the Heisenberg group H, since h
is often used for a Cartan subalgebra of a give Lie algebra.)
Example 5.7.4. By the results of Section 5.6, we have

gl(n, R) = Mn (R), gl(n, C) = Mn (C),
sl(n, R) = {X ∈ Mn (R) | tr X = 0}, sl(n, C) = {X ∈ Mn (C) | tr X = 0},
o(n) = so(n) = {X ∈ Mn (R) | X + X T = 0},
o(n, C) = so(n, C) = {X ∈ Mn (C) | X + X T = 0},
u(n) = {X ∈ Mn (C) | X + X ∗ = 0},
su(n) = {X ∈ Mn (C) | X + X ∗ = 0, tr X = 0},
sp(n) = {X ∈ Mn (H) | X + X ∗ = 0},
sp(n, R) = {X ∈ Mn (R) | JX T J = X}, sp(n, C) = {X ∈ Mn (C) | JX T J = X}.
Remark 5.7.5. (a) One of the most important ideas in Lie theory is to study Lie groups by
looking at their Lie algebras. Lie algebras are “flat” (i.e. they are vectors spaces) and
are thus easier to study than “curved” Lie groups.
(b) While the Lie algebra captures a lot of information about the Lie group, it does not
always capture G entirely. For instance, we have seen that O(n) and SO(n) have the
same Lie algebra.
(c) We usually consider the tangent spaces at the identity but in fact all of the tangent spaces
are isomorphic. This is because of the symmetry of a Lie group G given by multiplying
by an element g ∈ G (which maps the identity to g). See [Sti08, Exercises 5.4.4 and
5.4.5].
Since the Lie algebra of a Lie group is a real vector space, it has a dimension. In our
particular examples, we can compute this dimension explicitly.
Proposition 5.7.6 (Dimension of so(n), u(n), su(n), and sp(n)). As vector spaces over R,
we have
(a) dim so(n) = n(n − 1)/2.
(b) dim u(n) = n2 .
(c) dim su(n) = n2 − 1.
(d) dim sp(n) = n(2n + 1).
Proof. (a) We know that

so(n) = {X ∈ Mn (R) | X = −X T }
5.8. COMPLEX LIE GROUPS AND COMPLEXIFICATION 61
is the space of skew-symmetric matrices. For a skew-symmetric matrix X, its diagonal entries
are zero and its entries below the diagonal are determined by its entries above the diagonal
(which can be arbitrary). Therefore, dim so(n) is the number of entries above the diagonal,
which is
(n − 1) + (n − 2) + · · · + 1 = n(n − 1)/2.
(b) We know that u(n) is the space of skew-Hermitian matrices. Such matrices have pure
imaginary entries on the diagonal and n(n − 1)/2 arbitrary entries above the diagonal (which
determine the entries below the diagonal). So the number of independent real parameters is
n(n − 1)
2· + n = n2 .
2
(c) We know that su(n) is the space of skew-Hermitian matrices with trace zero. The
condition of the trace being zero is the same as saying that the nth diagonal entry must be
the negative of the sum of the other diagonal entries. Therefore, the number of independent
real parameters is n2 − 1.
(d) We know that sp(n) consists of the skew-Hermitian quaternion matrices. Thus, as
above, such a matrix has n(n − 1)/2 quaternion entries above the diagonal (which determine
the entries below the diagonal), and n pure imaginary quaternion entries on the diagonal.
So the number of independent real parameters is
n(n − 1)
4· + 3n = n(2n + 1).
2
Exercises.
5.7.1. Show that if A(t) and B(t) are smooth paths in a Lie group G, then A(t)B(t) is also
a smooth path in G.
5.8 Complex Lie groups and complexification

Recall that the Lie algebra of a Lie group is a real vector space (and the Lie bracket is bilinear
for real scalars).
Definition 5.8.1 (Complex matrix Lie group). A matrix Lie group G ⊆ GL(n, C) is complex
if its Lie algebra is a complex subspace of gl(n, C).
Definition 5.8.2 (Complexification of a vector space). If V is a finite-dimensional real vector

space, then the complexification of V , denoted VC , is the space of formal linear combinations
v1 + iv2
with v1 , v2 ∈ V . This is a real vector space with “componentwise” addition and scalar
multiplication given by
a(v1 + iv2 ) = av1 + i(av2 ), a ∈ R.
It is a complex vector space if we define
i(v1 + iv2 ) = −v2 + iv1 .
We can identify V with the real subspace {v + i0 ∈ VC } and so view V as a real subspace of
VC .
Remark 5.8.3. (a) In terms of the tensor product, we have VC = V ⊗R C.
(b) Suppose V is already a complex vector space. Then it is also a real vector space and we
can form VC . Note that VC is not equal to V . The dimension of VC is always twice the
dimension of V . In the language of tensor products, this non-equality comes from the
fact that we are tensoring over R and not over C. For example, CC is isomorphic (as a
complex vector space) to C2 .
Proposition/Definition 5.8.4 (Complexification of a Lie algebra). If g is a finite-dimensional

real Lie algebra, then the Lie bracket on g has a unique extension to gC which makes gC into
a complex Lie algebra, called the complexification of g.
Proof. If such a Lie bracket on gC exists, it must be given by
[X1 + iX2 , Y1 + iY2 ] = ([X1 , Y1 ] − [X2 , Y2 ]) + i([X1 , Y2 ] + [X2 , Y1 ]), X1 , X2 , Y1 , Y2 ∈ g, (5.2)
(by C-bilinearity and the fact that the bracket on gC must be an extension of the Lie bracket
on g – that is, the restriction of the Lie bracket on gC to g must agree with the original Lie
bracket on g). Thus, the uniqueness result follows. For existence, we must show that the
Lie bracket defined by (5.2) is C-bilinear, skew-symmetric, and satisfies the Jacobi identity.
It is easy to see that the bracket defined by (5.2) is skew-symmetric, thus to show complex
bilinearity, it suffices to show bilinearity in the first argument. Also, (5.2) is clearly real
bilinear (since the bracket on g is). So to show complex bilinearity, we only need to show
that
[i(X1 + iX2 ), Y1 + iY2 ] = i[X1 + iX2 , Y2 + iY2 ], for X1 , X2 , Y1 , Y2 ∈ g.
Now,

i[X1 + iX2 , Y2 + iY2 ] = i ([X1 , Y1 ] − [X2 , Y2 ]) + i([X2 , Y1 ] + [X1 , Y2 ])
= (−[X2 , Y1 ] − [X1 , Y2 ]) + i([X1 , Y1 ] − [X2 , Y2 ])
= [−X2 + iX1 , Y1 + iY2 ]
= [i(X1 + iX2 ), Y1 + Y2 ],
as desired. We know that the Jacobi identity
[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0

5.8. COMPLEX LIE GROUPS AND COMPLEXIFICATION 63
holds for X, Y, Z ∈ g. Now, the left hand side is complex linear in X. So for X1 , X2 ∈ g, we
have
[X1 + iX2 , [Y, Z]] + [Y, [Z, X1 + iX2 ]] + [Z, [X1 + iX2 , Y ]]

= [X1 , [Y, Z]] + [Y, [Z, X1 ]] + [Z, [X1 , Y ]] + i [X2 , [Y, Z]] + [Y, [Z, X2 ]] + [Z, [X2 , Y ]] = 0
Thus the Jacobi identity holds for X ∈ gC and Y, Z ∈ g. Since it is also complex linear in Y
and Z, repeating the above argument then shows it holds for X, Y, Z ∈ gC .
Remark 5.8.5. Note that the definition of complexification given in [Sti08, §5.6] is slightly
different. Instead of considering formal expressions X1 + iX2 as above, [Sti08, §5.6] uses
the fact that one considers spaces inside gl(n, C), where complex multiplication is de-
fined and X1 + iX2 is to be viewed as an expression is that space. Thus, for instance,
gl(n, C)C = gl(n, C) in [Sti08, §5.6] (whereas this is not the case in the above definition of
complexification, as noted in Remark 5.8.3).
Proposition 5.8.6. The Lie algebras gl(n, C), sl(n, C), so(n, C), and sp(n, C) are complex
Lie algebras (hence GL(n, C), SL(n, C), SO(n, C), and Sp(n, C) are complex Lie groups).
Furthermore, we have the following isomorphisms of complex Lie algebras.
gl(n, R)C ∼
= u(n)C ∼
= gl(n, C),
su(n)C ∼= sl(n, R)C ∼
= sl(n, C),
so(n)C ∼
= so(n, C),
sp(n, R)C = sp(n)C ∼
∼ = sp(n, C).
Proof. The first statement is left as an easy exercise. Every X ∈ gl(n, C) can be written
uniquely in the form X1 + iX2 where X1 , X2 ∈ gl(n, R). This gives an isomorphism of
complex vector spaces between gl(n, R)C and gl(n, C). It is easy to check (exercise) that
this isomorphism commutes with the Lie bracket and is thus an isomorphism of complex Lie
algebras. Similarly, sl(n, R)C ∼
= sl(n, C).
Also, for X ∈ gl(n, C), we have
X − X∗ X + X∗
X= +i ,
2 2i
and (X −X ∗ )/2 and (X +X ∗ )/2i are both skew-symmetric. To show that this decomposition
is unique, suppose
X = X1 + iX2 , X1 , X2 ∈ u(n).
Then
X + X ∗ = (X1 + iX2 ) + (X1 + iX2 )∗ = (X1 + iX2 ) + (X1∗ − iX2∗ )

= X1 + iX2 − X1 + iX2 = 2iX2 .
and
X − X ∗ = (X1 + iX2 ) − (X1 + iX2 )∗ = X1 + iX2 − X1∗ + iX2∗

= X1 + iX2 + X1 − iX2 = 2X1 .
Thus
X1 = (X − X ∗ )/2, X2 = (X + X ∗ )/2i,
and so the decomposition is unique. Therefore u(n)C ∼= gl(n, C). Since X has trace zero if
∼
and only if X1 and X2 do, we see that su(n)C = sl(n, C).
The other proofs are analogous.
Definition 5.8.7 (Real forms). Suppose g is a complex Lie algebra. A real Lie algebra g̃ is
called a real form of g if g̃C ∼
= g.
So, for example, gl(n, R) and u(n) are real forms of gl(n, C). Note however that gl(n, R)
and u(n) are not isomorphic (real) Lie algebras, except when n = 1.
5.9 The matrix logarithm

Our goal is to give a (more) precise relationship between a Lie group and its Lie algebra. We
have the exponential map which maps the Lie algebra into the Lie group. We now consider
the inverse map.
In this course, the notation log will mean loge . Recall the series for log(1 + x):
x2 x3 x4
log(1 + x) = x − + − + ··· , x ∈ R, |x| < 1.
2 3 4
Definition 5.9.1 (Matrix logarithm). For a square matrix I + A with kAk < 1, we define
its logarithm by
A2 A3
log(I + A) = A − + − ··· .
2 3
By the submultiplicative property, we know this series is absolutely convergent for kAk <
1 and so log(I + A) is a well-defined continuous function in this neighbourhood of I.
Proposition 5.9.2 (Inverse property of the matrix logarithm). (a) For any matrix X such
that keX − Ik < 1, we have
log(eX ) = X.
(b) For any matrix X such that kX − Ik < 1, we have
elog X = X.
Proof. By the definition of the matrix logarithm, we have
log(eX ) = log(I + (ex − I))

X X2

= log I + + + ...
1! 2!
2 3
X X2 1 X X2 1 X X2

= + + ... − + + ... + + + ... − ....
1! 2! 2 1! 2! 3 1! 2!
5.10. THE EXPONENTIAL MAPPING 65
Since the series are absolutely convergent, we can rearrange terms. Collecting powers of X m
gives
X 1 1 2 1 1 1
log(e ) = X + − X + − + X3 + . . . .
2! 2 3! 2 3
The coefficient in front of X m for m ≥ 2 is exactly the same as the coefficient that appears
in the expansion of log(ex ) for a real number x with |ex − 1| < 1. Since we know that
log(ex ) = x for such x, these coefficients are zero. Therefore log(eX ) = X as desired.
Similarly, if kX − Ik < 1, we have
exp(log X) = exp(log(I + (X − I)))

(X − I)2 (X − I)3

= exp (X − I) − + − ...
2 3
2
(X − I)2 (X − I)2

1
= I + (X − I) − + ... + (X − I) − + ... + ··· .
2 2! 2
Again, since the series are absolutely convergent, we can rearrange terms. We collect powers
of X m and use the fact that elog x = x for x ∈ R, |x − 1| < 1, to conclude that elog X = X.
Proposition 5.9.3 (Multiplicative property of the matrix logarithm). If AB = BA and

log A, log B and log(AB) are all defined, then
log(AB) = log A + log B.
Proof. Let X = log A and Y = log B. Then

(A − I)2 (A − I)3
X = log(I + (A − I)) = (A − I) − + − ...,
2 3
(B − I)2 (B − I)3
Y = log(I + (B − I)) = (B − I) − + − ....
2 3
Since A and B commute, these series commute, hence X and Y commute. Thus
AB = eX eY = eX+Y .
Taking log of both sides, we get
log(AB) = log(eX+Y ) = X + Y = log A + log B,
by the inverse property of the matrix logarithm.
5.10 The exponential mapping

When we computed the tangent spaces of the classical groups, we used the explicit defining
equations (such at AA∗ = I) to find an equation defining the Lie algebra and then used
the exponential map to show that all matrices satisfying this defining equation are in the
tangent space. However, in an arbitrary (matrix) Lie group, we may not have these explicit
defining equations at our disposal. So we need a more general (abstract) approach.
Proposition 5.10.1 (Exponentiation of tangent vectors). If X is a tangent vector at the

identity to a matrix Lie group G, then eX ∈ G. In other words, exp maps the tangent space
T1 (G) into G.
Proof. Suppose A(t) is a smooth path in G with A(0) = I and A0 (0) = X. Then
A(∆t) − 1 A(1/n) − 1
X = A0 (0) = lim = lim .
∆t→0 ∆t n→∞ 1/n
Now, using the definition of the logarithm of a square matrix, we have
log A(1/n) = log(I + (A(1/n) − I))

(A(1/n) − I)2 (A(1/n) − I)3
= (A(1/n) − I) − + − ··· .
2 3
Multiplying both sides by n (or dividing by 1/n) gives
(A(1/n) − I)2

A(1/n) − I A(1/n) − I A(1/n) − I
n log A(1/n) = − − + ··· . (5.3)
1/n 1/n 2 3
Now, since we are interested in the limit as n → ∞, we can restrict our attention to n > N
for some N > 0. Also, since A(0) = I and A(t) is continuous, for all 0 < < 1/2, there
exists an N > 0 such that kA(1/n) − Ik < < 1/2 for n > N . Then the series in parentheses
has a sum of absolute value less than

+ 2 + 3 + · · · = < 2,
1−
and so this sum tends to zero as n → ∞. Therefore, by (5.3), we have
lim n log A(1/n) = A0 (0) − A0 (0)0 = A0 (0) = X.

n→∞
Exponentiating both side of this equation, we obtain
eX = elimn→∞ n log A(1/n)

= lim en log A(1/n) (since exp is continuous)
n→∞
n
= lim elog A(1/n) (since eA+B = eA eB when AB = BA)
n→∞
= lim A(1/n)n (since exp is the inverse of log).
n→∞
Now, A(1/n) ∈ G since A(t) is a path in G. Thus A(1/n)n ∈ G since G is a group (i.e.
closed under products). Thus we have a convergent sequence of elements of G. The limit
is nonsingular since the limit, eX , has inverse e−X . Therefore, by the closure property of
(matrix) Lie groups, eX ∈ G as desired.
Remark 5.10.2. This is the first time we have really made use of the closure (under limits)
property of a Lie group.
5.11. THE LOGARITHM INTO THE TANGENT SPACE 67
We have seen examples where every element of a Lie group G is the exponential of an
element of its Lie algebra (for instance, SO(2) and SU(2)). This also turns out to be true
for GL(n, C) (i.e. every invertible complex matrix is the exponential of some n × n matrix).
However, this is not true for all matrix Lie groups.
Consider, for example
−1 1
A= ∈ SL(2, C).
0 −1
We claim that there is no X ∈ sl(2, C) such that eX = A. Let X be an arbitrary matrix in
sl(2, C). Then tr X = 0 and so the two eigenvalues of X are negatives of each other. If both
eigenvalues are zero, then eX has 1 as an eigenvalue since if v is an eigenvector of X with
eigenvalue 0, then
eX v = (I + X + · · · ) v = v.
But A does not have 1 as an eigenvalue and so eX 6= A. Therefore, the eigenvalues of X
must be distinct. Thus X is diagonalizable and hence eX is diagonalizable. But A is not
diagonalizable. Therefore eX 6= A.
Therefore, we see that the exponential mapping is not always onto. It is also not always
one-to-one (remember the example of S1 ∼ = SO(2)). However, we will see that it is locally
one-to-one and onto.
Remark 5.10.3. A natural question one might ask is “what is the image of the exponential
map?” This is actually a highly nontrivial question and the answer is not known in general,
although there has been some progress towards answering it.
5.11 The logarithm into the tangent space

We have shown that exp maps the Lie algebra into the Lie group. The next natural question
is “Does the logarithm map the Lie group into the Lie algebra?” We first consider a slightly
modified notion of a tangent vector which will be useful.
Definition 5.11.1. X is a sequential tangent vector to G at 1 if there is a sequence (Am ) of

elements of G, and a sequence (αm ) of real numbers, such that
Am − 1
lim Am = 1, lim = X.
m→∞ m→∞ αm
Note that the difference between this definition and the usual definition of a tangent
vector is that we take a sequence and not a continuous limit.
Proposition 5.11.2. We have that X is a sequential tangent vector to a Lie group G if and
only if it is a tangent vector to G.
Proof. Suppose X is a tangent vector to G. Then there exists a path A(t) in G with A(0) = I
and A0 (0) = X. Then the sequence (Am = A(1/m)) tends to I and
Am − I
lim = A0 (0) = X
m→∞ 1/m
and so X is a sequential tangent vector.

Now suppose that X is a sequential tangent vector. Then there is a sequence (Am ) in G,
and a sequence (αm ) of real numbers, such that
Am − I
lim Am = I, lim = X. (5.4)
m→∞ m→∞ αm
Our goal is to show that etX ∈ G for all t ∈ R. Then X is the tangent vector to this path.
We first show that eX ∈ G. It follows from (5.4) that αm → 0 as m → ∞. Thus
1/αm → ∞. Let
am = nearest integer to 1/αm .
Then we also have
lim am (Am − I) = X,
m→∞
since

1 1
(Am − I) − am (Am − I) = − am (Am − I) < kAm − Ik → 0 as m → ∞.
αm αm
Since am is an integer, by the multiplicativity property of the logarithm, we have
(Am − I)2

am Am − I
log(Am ) = am log(Am ) = am (Am − I) − am (Am − I) − + ··· .
2 3
As before, we know that the terms in parenthesis tends to zero as m → ∞. Thus
X = lim am (Am − I) = lim log(Aamm ).

m→∞ m→∞
Therefore am am
eX = elimm→∞ log(Am )
= lim elog(Am )
= lim Aamm .
m→∞ m→∞
Since am is an integer, Aamm ∈ G. Then by the closure of G under limits, we have eX ∈ G.

tX
Now, to show that e ∈ G for all t ∈ R, replace 1/αm in the above argument by t/αm
and let
bm = nearest integer to t/αm .
Then we have
lim bm (Am − I) = tX.
m→∞
If we consider the series for

log(Abmm ) = bm log(Am )
and argue as above, we find that
etX = lim Abmm ∈ G.

m→∞
Remark 5.11.3. The argument used in the above proof (namely, the passage from a sequence
to a smooth path) is an essential ingredient in proving that matrix Lie groups are smooth
manifolds.
5.11. THE LOGARITHM INTO THE TANGENT SPACE 69
Definition 5.11.4. A neighbourhood of 1 in G is a set of the form
Nδ (1) = {A ∈ G | kA − 1k < δ},
for some δ > 0. The above set is called a δ-neighbourhood of 1.
Proposition 5.11.5 (The logarithm of a neighbourhood of 1). For any matrix Lie group G
there is a neighbourhood Nδ (1) mapped into T1 (G) by the logarithm.
Proof. We prove the claim by contradiction. Suppose that no neighbourhood of 1 is mapped

into T1 (G) by the logarithm. Then there exists a sequence (Am ) in G with limm→∞ Am = 1
and log Am 6∈ T1 (G) for all m.
Since, by the definition of a matrix Lie group, G is contained in some Mn (C), we have
log Am ∈ Mn (C) and we can write (uniquely)
log Am = Xm + Ym ,
where Xm ∈ T1 (G) and 0 6= Ym ∈ T1 (G)⊥ where T1 (G) is the orthogonal complement to

T1 (G) in Mn (C). Since log is continuous and Am → 1, we have Xm , Ym → 0.
Consider the matrices
Ym
∈ T1 (G)⊥ .
kYm k
These all have absolute value 1 and so lie on the sphere of radius 1 and centre 0 in T1 (G)⊥ .
Since this sphere is bounded, there is a convergent subsequence, whose limit Y is a vector
of absolute value 1 in T1 (G)⊥ . We replace our original sequence with this subsequence and
thus have
Ym
lim = Y.
m→∞ kYm k
Now consider the sequence of terms
Tm = e−Xm Am .
Since e−Xm ∈ G (we know that exp maps tangent vectors into G) and Am ∈ G, we have
Tm ∈ G for all m. Also, by the inverse property of log, we have Am = eXm +Ym . Therefore
Tm = e−Xm eXm +Ym

2
(Xm + Ym )2

Xm
= I − Xm + + ... I + X m + Ym + + ...
2! 2!
= I + Ym + higher order terms.
Now, the higher order terms in Xm that do not involve any positive power of Ym are exactly
the terms that appear in the expansion of e−Xm eXm = I and so they sum to zero. Therefore
Tm − I Ym
lim = lim = Y.
m→∞ kYm k m→∞ kYm k
Since each kYm k → 0, we have Tm → I and thus Y is a sequential tangent vector. Therefore,
by Proposition 5.11.2, it is a tangent vector and so lies in T1 (G). But we saw above that
0 6= Y ∈ T1 (G)⊥ and so Y 6∈ T1 (G). This contradiction demonstrates that our initial

assumption was false. Therefore there is a neighbourhood of the identity mapped into T1 (G)
by log.
Corollary 5.11.6. There exists δ > 0 such that the log and exp functions give a bijection,
continuous in both directions, between Nδ (1) in G and log Nδ (1) in T1 (G).
Remark 5.11.7. A continuous bijection with continuous inverse is called a homeomorphism

(it is an isomorphism in the category of topological spaces). It follows from Corollary 5.11.6
that the topological dimension (a concept we have not defined) of a Lie group G is equal to
the dimension of its Lie algebra.
Corollary 5.11.8. If G is a connected matrix Lie group, then every element A ∈ G can be
written in the form
A = eX1 eX2 . . . eXm ,
for some X1 , . . . , Xm ∈ g.
For the following proof, we need to use a well-known property of S compact sets: If X is a
compact set and (Uα )α∈A is a collection of open sets in X such that α∈A Uα = X, then there
0
S
exists a finite subset A ⊆ A such that α∈A0 Uα = X. In fact, this is the defining property
of compact sets in general – our characterization of compact sets as closed and bounded sets
in Cn or Rn is equivalent (for subsets of Cn and Rn ).
Proof. Let V be a neighbourhood of I in G as in Corollary 5.11.6. Since G is connected,
we can find a continuous path γ : [0, 1] → G with γ(0) = I and γ(1) = A. Since γ is
continuous and [0, 1] is compact, the image γ([0, 1]) is also compact. Now, consider the
def
collection {g · V | g ∈ γ([0, 1])}. Since g ∈ g · V = {g · h | h ∈ V }, we have
[
γ([0, 1]) ⊆ g · V.
g∈γ([0,1])
S
Thus, since γ([0, 1]) is compact, there is a finite set S ⊆ γ([0, 1]) with γ([0, 1]) ⊆ g∈S g · V .
Thus we can choose a sequence 0 = t0 < t1 < t2 < · · · < tm = 1 and I = A0 = γ(t0 ), A1 =
γ(t1 ), . . . , Am = γ(tm ) = A with
A−1
i−1 Ai ∈ V for all 1 ≤ i ≤ m.
By the definition of V , we can find Xi ∈ g with eXi = A−1
i−1 Ai for i = 1, . . . , m. Then
A = (A−1 −1 −1
0 A1 )(A1 A2 ) · · · (Am−1 Am ) = e
X1
· · · eXm .
Remark 5.11.9. The key idea in the above proof is that connected Lie groups are generated
by a neighbourhood of the identity.
The proof of the following corollary is left as an exercise (Exercise 5.11.1).
Corollary 5.11.10 (An alternate description of the Lie algebra). Suppose G ⊆ GL(n, C) is
a matrix Lie group with Lie algebra g. Then X ∈ Mn (C) is in g if and only if etX ∈ G for
all t ∈ R.
5.12. SOME MORE PROPERTIES OF THE EXPONENTIAL 71
Remark 5.11.11. By Corollary 5.11.10, we could define the Lie algebra of G to be the set of
all matrices X such that etX ∈ G for all t ∈ R. This is the approach adopted in [Hal03].
Exercises.
5.11.1. Prove Corollary 5.11.10.
5.12 Some more properties of the exponential

Lemma 5.12.1. There exists a constant c such that for all B ∈ Mn (C) with kBk < 12 , we
have
k(log(I + B)) − Bk ≤ ckBk2 .
Proof. Since
∞ m ∞
X
m+1 B 2
X B m−2
log(I + B) − B = (−1) =B (−1)m+1 ,
m=2
m m=2
m
we have
∞ 1 m−2

X
k log(I + B) − Bk ≤ kBk2 2
.
m=2
m
Since the sum in the last expression is convergent, we are done.
Proposition 5.12.2 (Lie product formula). If X, Y ∈ Mn (C), then

m
eX+Y = lim eX/m eY /m .
m→∞
Proof. Multiplying the series for eX/m and eY /m we get

X/m Y /m X Y 1
e e =I+ + +O .
m m m2
For sufficiently large m, X/m and Y /m are in the domain of the logarithm, and thus
eX/m eY /m → I as m → ∞. Now, by Lemma 5.12.1, we have

X/m Y /m
X Y 1
log e e = log I + + +O
m m m2
2!
X Y X Y 1
= + +O + +O
m m m m m2

X Y 1
= + +O .
m m m2
Exponentiating gives
X/m Y /m X Y 1
e e = exp + +O
m m m2
and so
X/m Y /m m
1
e e = exp X + Y + O .
m
Since exp is continuous, we have
m
lim eX/m eY /m = exp(X + Y ),
m→∞
as desired.
Definition 5.12.3 (One-parameter subgroups). A one-parameter subgroup of GL(n, C) is a

function A : R → GL(n, C) such that
(a) A is continuous,
(b) A(0) = I,
(c) A(t + s) = A(t)A(s) for all t, s ∈ R.
Note that the last condition in the above definition is equivalent to saying that A is a
group homomorphism from (R, +) to G (the second condition is actually redundant). Recall
that R is a Lie group under addition. A one-parameter subgroup is a homomorphism R → G
of Lie groups. We wish to give a nice characterization of these one-parameter subgroups.
We first need a technical lemma.
Let
Bδ = {X ∈ Mn (C) | kXk < δ}
be the open ball of radius δ about zero in Mn (C). Choose δ sufficiently small so that Bδ is
mapped homeomorphically onto its image under the exponential map (with inverse given by
log). Then set U = exp(Bδ/2 ).
Lemma 5.12.4. Every g ∈ U has a unique square root h ∈ U , given by h = exp( 21 log g).
Proof. Let X = log g. Then, if h = exp(X/2), we have
h2 = eX = g,
and so h is a square root of g. Now suppose h0 ∈ U satisfies (h0 )2 = g. Let Y = log h0 . Then
eY = h0 , e2Y = (h0 )2 = g = eX .
We have Y ∈ Bδ/2 and so 2Y ∈ Bδ . We also have X ∈ Bδ/2 ⊆ Bδ . Since exp is injective on

Bδ and exp(2Y ) = exp(X), we must have 2Y = X. Therefore
h0 = eY = eX/2 = h,
which shows uniqueness.

5.13. THE FUNCTOR FROM LIE GROUPS TO LIE ALGEBRAS 73
Theorem 5.12.5 (One-parameter subgroups). If A is a one-parameter subgroup of GL(n, C),

then there exists a unique X ∈ Mn (C) such that
A(t) = etX for all t ∈ R.
Proof. Uniqueness follows from the fact that if A(t) = etX , then X = A0 (0). It remains to
show existence.
We fix δ and U as above. Since A is continuous, there exists a t0 > 0 such that A(t) ∈ U
for all t ∈ [−t0 , t0 ]. Let X = t10 log(A(t0 )). Then
t0 X = log(A(t0 )) ∈ Bδ/2 and A(t0 ) = exp(t0 X).
Then A(t0 /2) ∈ U and A(t0 /2)2 = A(t0 ). By the above lemma, we have
A(t0 /2) = exp(t0 X/2).
Repeating this argument, we have
A(t0 /2k ) = exp(t0 X/2k ) ∀k ∈ Z>0 .
Furthermore, for any integer m, we have
A(mt0 /2k ) = A(t0 /2k )m = exp(mt0 X/2k ).
Thus
A(t) = exp(tX)
for all real numbers t of the form mt0 /2k . Since this set of numbers is dense in R and both
exp(tX) and A(t) are continuous, it follows that A(t) = exp(tX) for t ∈ R.
5.13 The functor from Lie groups to Lie algebras

Lie groups and Lie algebras both form categories – we have homomorphisms of Lie groups
and Lie algebras and these satisfy certain natural axioms (associative composition, identity
maps, etc.). To every Lie group, we can associate its Lie algebra. We would now like to
extend this identification to a functor from the category of Lie groups to the category of
Lie algebras. So for every Lie group homomorphism Φ : G → H, we need to associate a Lie
algebra homomorphism φ : g → h and we want this association to commute with composition.
Theorem 5.13.1. Let G and H be matrix Lie groups, with Lie algebras g and h, respectively.
Suppose that Φ : G → H is a Lie group homomorphism. Then there exists a unique real linear
map φ : g → h such that
Φ eX = eφ(X) for all X ∈ g.

The map φ has the following additional properties.

(a) φ is a homomorphism of Lie algebras: φ([X, Y ]) = [φ(X), φ(Y )] for all X, Y ∈ g.
(b) φ(AXA−1 ) = Φ(A)φ(X)Φ(A)−1 , for all X ∈ g, A ∈ G.

d

(c) φ(X) = dt
Φ etX t=0
for all X ∈ g.
Furthermore, this association defines a functor from the category of Lie groups to the category
of Lie algebras. Namely, suppose G, H, and K are matrix Lie groups and Φ : H → K and
Ψ : G → H are Lie group homomorphisms. Let φ, ψ and λ be the Lie algebra homomorphisms
corresponding to Φ, Ψ and Φ ◦ Ψ, respectively. Then
λ = φ ◦ ψ.
Remark 5.13.2. The property φ(X) = dtd Φ etX t=0 says that φ is the derivative of Φ at the

identity (every smooth map of smooth manifolds induces a map on the tangent spaces). In
practice, this is how one computes φ. Since it is real linear, it suffices to compute it on a
basis of g.
Proof. The composition of Lie group homomorphisms
R → G → H, t 7→ etX 7→ Φ(etX ),
is again a Lie group homomorphism and thus t 7→ Φ(etX ) is a one-parameter subgroup of

H. Therefore, by Theorem 5.12.5, there is a unique matrix Z such that
Φ(etX ) = etZ for all t ∈ R. (5.5)
Since etZ = Φ(etX ) ∈ H, we know that Z ∈ h by our alternative description of the Lie
algebra of H (Corollary 5.11.10). We define φ(X) = Z and check that the map φ has the
desired properties.
Step 1. Φ(eX ) = eφ(X) for all X ∈ g.
Setting t = 1 in (5.5), we have
Φ(eX ) = eZ = eφ(X) .
Step 2. The map φ is (real) linear.

Since Φ(etX ) = etZ for all t, we have Φ(etsX ) = et(sZ) for all s ∈ R. Thus, φ(sX) = sZ =
sφ(X) for all s ∈ R.
Now, by the above, we have
etφ(X+Y ) = eφ(t(X+Y )) = Φ(et(X+Y ) ).
By the Lie product formula (Proposition 5.12.2) and the fact that Φ is a continuous homo-
morphism, we have
m
etφ(X+Y ) = Φ lim etX/m etY /m
m→∞
m
= lim Φ etX/m Φ etY /m

m→∞
m
= lim etφ(X)/m etφ(Y )/m = et(φ(X)+φ(Y )) .
m→∞
Differentiating at t = 0 gives φ(X + Y ) = φ(X) + φ(Y ). Thus φ is a linear map.

Step 3. φ(AXA−1 ) = Φ(A)φ(X)Φ(A)−1 for all X ∈ g and A ∈ G.

−1
First note that for X ∈ g, A ∈ G, and t ∈ R, we have etAXA = AetX A−1 ∈ G, since
A, etX ∈ G. Thus, by Corollary 5.11.10, we have AXA−1 ∈ g. Then we have
exp tφ(AXA−1 ) = exp φ(tAXA−1 ) = Φ(exp(tAXA−1 )) (by the above)
tX −1
= Φ(Ae A )
= Φ(A)Φ(etX )Φ(A)−1 (Φ is a group homomorphism)
= Φ(A)etφ(X) Φ(A)−1 (by the above).
Differentiating gives the desired result.
Step 4. φ is a homomorphism of Lie algebras.
We have from the above that φ is a linear map. It thus suffices to check that φ([X, Y ]) =
[φ(X), φ(Y )]. Now,
d
[X, Y ] = etX Y e−tX .
dt t=0
(We saw this in the proof of Proposition 5.7.2—it is the standard way of recovering the
bracket in g from the group operation in G.) Therefore,

d tX −tX d
= φ etX Y e−tX

φ([X, Y ]) = φ e Ye .
dt t=0 dt t=0
(Here we have used the fact that derivatives commute with linear transformations.) Then,
by the above, we have
d
φ([X, Y ]) = Φ(etX )φ(Y )Φ(e−tX )
dt t=0
d tφ(X)
= e φ(Y )e−tφ(X)
dt t=0
= [φ(X), φ(Y )]
as desired.
d

Step 5. φ(X) = dt
Φ etX t=0
for all X ∈ g.
By the definition of φ(X), we have
d d tφ(X)
Φ etX

= e = φ(X).
dt t=0 dt t=0

Step 6. Uniqueness: φ is the unique real linear map such that Φ eX = eφ(X) for all
X ∈ g.
Let ψ be such a map. Then, for all X ∈ g and t ∈ R,
etψ(X) = eψ(tX) = Φ etX ,

and so
d
Φ etX

ψ(X) = = φ(X).
dt t=0
Step 7. λ = φ ◦ ψ.
For X ∈ g, we have
(Φ ◦ Ψ) etX = Φ Ψ etX = Φ etψ(X) = etφ(ψ(X)) .

Thus, by uniqueness,
λ(X) = φ(ψ(X)).
Definition 5.13.3 (The adjoint mapping). Let G be a matrix Lie group with Lie algebra
g. Then, for each A ∈ G, we have the adjoint mapping
AdA : g → g, AdA (X) = AXA−1 .
(Note that AXA−1 ∈ g as in the proof of Step 3 of Theorem 5.13.1.)
Let G be a matrix Lie group with Lie algebra g. Let Aut g denote the group of Lie
algebra automorphisms of g (i.e. Lie algebra isomorphisms from g to itself). Note that Aut g
is a group under composition.
Proposition 5.13.4. The map
G → Aut g, A 7→ AdA ,
is a group homomorphism.
Remark 5.13.5. Note that the above proposition implies that, for each A ∈ G, AdA is an
invertible linear transformation of g with inverse AdA−1 , and
AdA ([X, Y ]) = [AdA (X), AdA (Y )], for all X, Y ∈ g.
Note that Aut g is a subgroup of GL(g), the group of invertible linear transformation of g.
Since g is a finite dimensional real vector space of some dimension k, we can identity GL(g)
with GL(k, R) and regard it as a matrix Lie group. So Ad gives rise to a map G → GL(g)
which is continuous (Exercise 5.13.2) and is thus a homomorphism of Lie groups. Therefore,
by Theorem 5.13.1, there is an associated homomorphism of Lie algebras
ad : g → gl(g), X 7→ adX
such that
eadX = Ad(eX ).
Here gl(g) is the Lie algebra of GL(g), namely the space of all linear maps from g to itself.
Proposition 5.13.6. For all X, Y ∈ g, we have
adX (Y ) = [X, Y ].
Proof. By Theorem 5.13.1, we have

d
Ad etX

adX = t=0
.
dt
Therefore
d d tX −tX
Ad etX (Y )

adX (Y ) = t=0
= e Ye t=0
= [X, Y ].
dt dt
Proposition 5.13.7. Suppose G is a connected matrix Lie group, H is a matrix Lie group,
and Φ1 , Φ2 : G → H are Lie group homomorphisms with associated Lie algebra homomor-
phisms φ1 and φ2 . If φ1 = φ2 , then Φ1 = Φ2 .
Proof. Let g ∈ G. Since G is connected, we know from Corollary 5.11.8 that

g = eX1 eX2 . . . eXn
for some Xi ∈ g. Then
Φ1 (g) = Φ1 (eX1 ) · · · Φ1 (eXn )
= eφ1 (X1 ) · · · eφ1 (Xn )
= eφ2 (X1 ) · · · eφ2 (Xn )
= Φ2 (eX1 ) · · · Φ2 (eXn )
= Φ2 (g).
Corollary 5.13.8. Every continuous homomorphism between two matrix Lie groups is smooth.
Proof. Suppose Φ : G → H is a continuous homomorphism of Lie groups. If suffices to

prove that Φ is smooth in a neighbourhood of every point A ∈ G. Now, there exists a
neighbourhood U of A such that B = A exp X, X ∈ g, for all B ∈ U (we translate the
neighbourhood of the identity in the image of the exponential map by multiplying on the
left by A). Then
Φ(B) = Φ(A)Φ(exp X) = Φ(A) exp(φ(X)), for all B ∈ U.
Thus, in the exponential coordinates near A, Φ is a composition of the linear map φ, the
exponential mapping, and left multiplication by Φ(A). Since all of these maps are smooth,
we see that Φ is smooth near any point A of G.
Remark 5.13.9. We have seen most of the ingredients of the proof of the fact that every
matrix Lie group is a Lie group (see [Hal03, Corollary 2.33] for details).
Exercises.
5.13.2. Show that the map Ad : G → GL(g) is continuous.

5.14 Lie algebras and normal subgroups

Recall that any discrete normal subgroup of a path-connected group lies in its centre (see
Theorem 4.3.2). Since we know that SO(n), SU(n), and Sp(n) are path-connected and we
have described their centres, we know their discrete normal subgroups. However, we do not
know (yet) if they have any nondiscrete normal subgroups. It turns out that our knowledge
of the Lie algebra will now allow us to answer this question.
Proposition 5.14.1. If H is a normal subgroup of a matrix Lie group G, then h is an ideal

of g.
Proof. Since any tangent to H at 1 is a tangent to G at 1, h is a subspace of g. Suppose

X ∈ g and Y ∈ h. Let A(s) be a smooth path in G and let B(t) be a smooth path in H
with A(0) = B(0) = 1, A0 (0) = X, and B 0 (0) = Y . Since H is a normal subgroup of G, we
have that
Cs (t) = A(s)B(t)A(s)−1
is a smooth path in H. Then
def
D(s) = Cs0 (0) = A(s)Y A(s)−1
is a smooth path in h. Therefore
D0 (0) = XY − Y X ∈ h.
Thus h is an ideal of g as claimed.
Proposition 5.14.2 (Tangent space visibility). Suppose G is a path-connected matrix Lie

group with discrete centre and H is a nondiscrete normal subgroup. Then h 6= 0.
Proof. Since Z(G) is discrete while H is not, there exists a neighbourhood Nδ (1) in G that
includes an least one element B 6= 1 in H but no element of Z(G) other than 1. Then
B 6∈ Z(G) and so there exists some A ∈ Nδ (1) that does not commute with B (since Nδ (1)
generates G, if B commuted with all elements of Nδ (1), it would lie in Z(G)).
By taking δ sufficiently small, we can assume A = eX and etX ∈ Nδ (1), 0 ≤ t ≤ 1, for
some X ∈ T1 (G). Now let
C(t) = etX Be−tX B −1 .
Thus C(t) is a path in G with C(0) = 1 and C(1) = ABA−1 B −1 . Using the product rule for
differentiation, we obtain
C 0 (0) = X − BXB −1 .
Since H is a normal subgroup of G and B ∈ H, we have etX Be−tX ∈ H. Thus etX Be−tX B −1 ∈
H and so C(t) is actually a path in H. Thus
C 0 (0) = X − BXB −1 ∈ h.
To show that h 6= 0 is thus suffices to show that X − BXB −1 6= 0. We prove this by
contradiction. We have
X − BXB −1 = 0 =⇒ BXB −1 = X
5.15. THE CAMPBELL–BAKER–HAUSDORFF THEOREM 79
−1
=⇒ eBXB = eX
=⇒ BeX B −1 = eX
=⇒ BeX = eX B
=⇒ BA = AB.
But this contradicts the fact that A and B do not commute.
Corollary 5.14.3. Under the assumptions of Proposition 5.14.2, if H 6= G, then h is a

nontrivial ideal of g.
Proof. We know that h is an ideal of g by Proposition 5.14.1 and h 6= 0 by Proposition 5.14.2.

Suppose h = g. Then, by the log-exp bijection between neighbourhoods of the identity in G
and g, we have that Nδ (1) ⊆ H for some neighbourhood Nδ (1) of the identity in G. Since
Nδ (1) generates G, we have H = G. Thus H 6= G =⇒ h 6= g.
A proof of the following result can be found in [Sti08, §§6.4–6.6].
Lemma 5.14.4. The Lie algebras so(n) (n 6= 4), su(n), and sp(n) have no nontrivial ideals.
Corollary 5.14.5 (Normal subgroups of SO(n), SU(n), and Sp(n)). The only nontrivial
normal subgroups of SO(n) (n 6= 4), SU(n) and Sp(n) are the (cyclic) subgroups of their
centres.
Proof. This follows immediately from Theorem 4.3.2, Corollary 5.14.3, and Lemma 5.14.4.
Remark 5.14.6. Recall the discussion of SO(4) in Section 1.8. In particular, SO(4) is not
simple.
We can now prove something that we claimed earlier (see Proposition 1.5.10).
Corollary 5.14.7 (Simplicity of SO(2m + 1)). The matrix Lie group SO(2m + 1) is simple
for m a nonnegative integer.
Proof. This follows from Corollary 5.14.5 and the fact that the centre of SO(2m + 1) is
trivial.
Remark 5.14.8. Note that SO(2m) is not simple since it has centre {±1}.
5.15 The Campbell–Baker–Hausdorff Theorem

Our goal in this section is to show that, near the identity, the multiplication in a Lie group is
determined by the Lie bracket in the corresponding Lie algebra. Remember that we already
know the reverse is true (we can recover the Lie bracket from the group multiplication by
differentiation).
Recall that in some neighbourhood of the identity, any two elements of G have the form
X
e and eY for some X, Y ∈ g and their product is of the form eZ for some Z ∈ g:
eX eY = eZ .
We would like to find a formula for Z in terms of X and Y that only involves Lie algebra
operations (i.e. vector space operations and the Lie bracket). In other words, we want such
a formula for
Z = log eX eY .

Since
X X2 Y Y2
eX = I + + + ··· , eY = I + + + ··· ,
1! 2! 1! 2!
we have
X X mY n X2 Y 2 X mY n
X Y
e e = = I + X + Y + XY + + + ··· + + ··· .
m,n≥0
m!n! 2! 2! m!n!
Since
W2 W3 W4
log(I + W ) = W − + − + ··· ,
2 3 4
we have
Z = log eX eY

2
X2 Y 2 X2 Y 2

1
= X + Y + XY + + + ··· − X + Y + XY + + + ···
2! 2! 2 2! 2!
3
X2 Y 2

1
+ X + Y + XY + + + ··· − ···
3 2! 2!
1 1
= X + Y + XY − Y X + higher-order terms
2 2
1
= X + Y + [X, Y ] + higher-order terms.
2
The Campbell–Baker–Hausdorff Theorem asserts that all of the higher-order terms are ob-
tained from X and Y by Lie brackets.
There are several proofs of the Campbell–Baker–Hausdorff Theorem. Following [Sti08],
we will present a proof by Eichler (1968). This proof is elementary in the sense that it does
not use any sophisticated machinery. One drawback is that it is not extremely intuitive.
Students who wish to see an alternate proof can look at [Hal03, §3.4].
Our proof will be by induction. Let
eA eB = eZ , Z = F1 (A, B) + F2 (A, B) + F3 (A, B) + · · · , (5.6)
where Fn (A, B) is the sum of all the terms of degree n appearing in the formula for Z in
terms of A and B. So Fn (A, B) is a homogeneous polynomial of degree n in A and B. Since
the variables A and B stand for matrices, the variables do not commute in general, but their
multiplication is associative.
Definition 5.15.1 (Lie polynomial). A polynomial p(A, B, C, . . . ) is called Lie if is a linear

combination of A, B, C, . . . and (possibly nested) Lie bracket terms in A, B, C, . . . .
From our computations above, we see that

1
F1 (A, B) = A + B, F2 (A, B) = [A, B].
2
Thus F1 and F2 are Lie polynomials.
Theorem 5.15.2 (Campbell–Baker–Hausdorff Theorem). The polynomial Fn (A, B) is Lie

for all n ≥ 1.
Proof. Our induction hypothesis is that Fm is a Lie polynomial for m < n, and we want to
prove that Fn is Lie. Since we know F1 and F2 are Lie polynomials, we may assume that
n ≥ 3.
Since the multiplication of variables A, B, C, . . . is associative, so is the multiplication of
products of power series in A, B, C, . . . . Therefore, for any A, B, C, we have
eA eB eC = eA eB eC .

Thus, if we set eA eB eC = eW , we have

∞ ∞
! ∞ ∞
!
X X X X
W = Fi Fj (A, B), C = Fi A, Fj (B, C) . (5.7)
i=1 j=1 i=1 j=1
By the induction hypothesis, we have:
• All the homogeneous terms of degree < n in both expressions for W in (5.7) are Lie.
• All the homogeneous terms of degree n in both expressions for W in (5.7) coming from
i > 1 and j > 1 are Lie (since they are Lie polynomials of Lie polynomials).
The only terms which could possibly not be Lie are the polyonimals
• Fn (A, B) + Fn (A + B, C) on the left (from i = 1, j = n and i = n, j = 1), and
• Fn (A, B + C) + Fn (B, C) on the right (from i = n, j = 1 and i = 1, j = n).
Therefore, the difference between these exceptional terms is a difference of Lie polynomials
and hence is a Lie polynomial. Note that the relation
F ≡Lie G ⇐⇒ F − G is a Lie polynomial (5.8)
is an equivalence relation (Exercise 5.15.1). We want to prove that Fn (A, B) ≡Lie 0.

We have just argued that
Fn (A, B) + Fn (A + B, C) ≡Lie Fn (A, B + C) + Fn (B, C). (5.9)
We will manipulate (5.9) to obtain our desired result. We will use the following facts:
Fact 1. Fn (rA, sA) = 0 for all scalars r, s and n > 1. This is true since rA and sA commute
and so erA esA = erA+sA . Thus Z = F1 (rA, sA) = (r + sA and Fn (rA, sA) = 0 for all
n > 1.
Fact 2. In particular, Fn (A, 0) = 0 for n > 1 (take r = 1, s = 0).
Fact 3. Fn (rA, rB) = rn Fn (A, B) for all r ∈ R and n ≥ 1, since Fn (A, B) is homogeneous of
degree n.
We first replace C by −B in (5.9) to obtain
Fn (A, B) + Fn (A + B, −B) ≡Lie Fn (A, 0) + Fn (B, −B)

≡Lie 0 by Facts 2 and 1.
Thus we have
Fn (A, B) ≡Lie −Fn (A + B, −B). (5.10)
Next, we replace A by −B in (5.9) to obtain
Fn (−B, B) + Fn (0, C) ≡Lie Fn (−B, B + C) + Fn (B, C).
Using Facts 1 and 2 again, we get
0 ≡Lie Fn (−B, B + C) + Fn (B, C).
Replacing B by A and C by B yields
0 ≡Lie Fn (−A, A + B) + Fn (A, B),
and so
Fn (A, B) ≡Lie −Fn (−A, A + B). (5.11)
We now have
Fn (A, B) ≡Lie −Fn (−A, A + B) by (5.11)

≡Lie −(−Fn (−A + A + B, −A − B)) by (5.10)
≡Lie Fn (B, −A − B)
≡Lie −Fn (−B, −A) by (5.11)
n
≡ −(−1) Fn (B, A) by Fact 3.
Thus we have
Fn (A, B) ≡Lie −(−1)n Fn (B, A). (5.12)
Now we replace C by −B/2 in (5.9), giving
Fn (A, B) + Fn (A + B, −B/2) ≡Lie Fn (A, B/2) + Fn (B, −B/2)

≡Lie Fn (A, B/2) by Fact 1.
Thus
Fn (A, B) ≡Lie Fn (A, B/2) − Fn (A + B, −B/2). (5.13)
Next, replacing A by −B/2 in (5.9) yields

Fn (−B/2, B) + Fn (B/2, C) ≡Lie Fn (−B/2, B + C) + Fn (B, C).
Thus, by Fact 1, we have
Fn (B/2, C) ≡Lie Fn (−B/2, B + C) + Fn (B, C).
Replacing B by A and C by B gives
Fn (A/2, B) ≡Lie Fn (−A/2, A + B) + Fn (A, B),
or
Fn (A, B) ≡Lie Fn (A/2, B) − Fn (−A/2, A + B). (5.14)
We now manipulate the two terms on the right side of (5.14).
Fn (A/2, B) ≡Lie Fn (A/2, B/2) − Fn (A/2 + B, −B/2) by (5.13)
≡Lie Fn (A/2, B/2) + Fn (A/2 + B/2, B/2) by (5.10)
−n −n
≡Lie 2 Fn (A, B) + 2 Fn (A + B, B) by Fact 3.
Fn (−A/2, A + B) ≡Lie Fn (−A/2, A/2 + B/2) − Fn (A/2 + B, −A/2 − B/2) by (5.13)

≡Lie −Fn (A/2, B/2) + Fn (B/2, A/2 + B/2) by (5.11) and (5.10)
−n −n
≡Lie −2 Fn (A, B) + 2 Fn (B, A + B) by Fact 3.
Therefore, (5.14) becomes
Fn (A, B) ≡Lie 21−n Fn (A, B) + 2−n Fn (A + B, B) − 2−n Fn (B, A + B).
Using (5.12), this becomes
(1 − 21−n )Fn (A, B) ≡Lie 2−n (1 + (−1)n )Fn (A + B, B). (5.15)
Now, if n is odd, (5.15) tells us that Fn (A, B) ≡Lie 0 as desired. So assume n is even.
Replace A by A − B in (5.15) to get
(1 − 21−n )Fn (A − B, B) ≡Lie 21−n Fn (A, B).
The left side of the above equation becomes
(1 − 21−n )Fn (A − B, B) ≡Lie −(1 − 21−n )Fn (A, −B) by (5.10),
and so
21−n
−Fn (A, −B) ≡Lie Fn (A, B). (5.16)
1 − 21−n
Finally, replacing B by −B in (5.16), we get
21−n
−Fn (A, B) ≡Lie Fn (A, −B)
1 − 21−n
1−n 2
2
≡Lie − Fn (A, B) by (5.16).
1 − 21−n
This implies Fn (A, B) ≡Lie 0 as desired.
Exercises.
5.15.1. Show that (5.8) is an equivalence relation.
Chapter 6
Covering groups
In this chapter we examine in further detail the correspondence between Lie groups and
Lie algebras. We already know examples of nonisomorphic Lie groups with isomorphic Lie
algebras. In order to have a one-to-one correspondence between Lie groups and Lie algebras,
we need to restrict our attention to certain types of Lie groups.
6.1 Simple connectedness

We have see that for a matrix Lie group homomorphism Φ : G → H, there exists a unique
Lie algebra homomorphism φ : g → h such that
exp(φ(X)) = Φ(exp(X)), for all X ∈ g.
We would like to turn our attention to the converse. Namely, given a Lie algebra homomor-
phism φ ∼= g → h, is there a Lie group homomorphism Φ ∼ = G → H satisfying the above
equation? We will see that the answer is yes, under certain additional assumptions. These
assumptions are related to the notion of simple connectedness.
Definition 6.1.1 (Simply connected). A subset S of Cn is simply connected if it is path-

connected and every loop in S can be shrunk continuously to a point in S.
More precisely, assume that S is connected. Then S is simply connected if, given any
continuous path A(t), 0 ≤ t ≤ 1, in S with A(0) = A(1), there exists a continuous function
A(s, t) from [0, 1]2 to S, such that
(a) A(s, 0) = A(s, 1) for all 0 ≤ s ≤ 1 (i.e. for any fixed s, A(s, t) is a loop),
(b) A(0, t) = A(t) for all 0 ≤ t ≤ 1 (i.e. fixing s = 0 gives our original path), and
(c) A(1, t) = A(1, 0) for all 0 ≤ t ≤ 1 (i.e. fixing s = 1 gives a constant path).
There are many equivalent definitions of simply connected.
Proposition 6.1.2 (Simply connected alternate definitions). A path-connected subset S of

Cn is simply connected (in the sense of Definition 6.1.1) if and only if it satisfies any one of
the following equivalent conditions.
85
86 CHAPTER 6. COVERING GROUPS
(a) For any two paths p and q in S from point A ∈ S to point B ∈ S, there is a deformation
of p to q with endpoints fixed. A deformation (or homotopy) of p to q is a continuous
function d : [0, 1]2 → S such that
(a) d(0, t) = p(t),

(b) d(1, t) = q(t), and
(c) d(s, 0) = p(0) = q(0) = A and d(s, 1) = p(1) = q(1) = B for all s.
(b) Any continuous map f : S1 → S (recall that S1 is the unit circle) can be contracted to a
point. That is, there exists a continuous function F : D → S, where D is the unit disk,
such that F |S1 = f .
(c) (For those with some background in topology.) The fundamental group of S is trivial.
Proof. It is clear that (a) implies that S is simply connected (set A(t) = p(t) and let q(t) be
the constant path at the point A(0)). Since we will not use these other definitions in this
class, we omit the proofs of the other implications.
It is rather hard to prove that a given space is simply connected, except in simple cases.
Proposition 6.1.3. The space Rn is simply connected.
Proof. Let A(t) be a loop in Rn . Define
A(s, t) = (1 − s)A(t).
Then A(s, t) is continuous and
(a) A(s, 0) = (1 − s)A(0) = (1 − s)A(1) = A(s, 1) for all 0 ≤ s ≤ 1,
(b) A(0, t) = A(t) for all 0 ≤ t ≤ 1, and
(c) A(1, t) = 0 = A(1, 0) for all 0 ≤ t ≤ 1.
Thus A(s, t) satisfies the conditions of Definition 6.1.1.
Proposition 6.1.4. The k-sphere Sk , k > 1, is simply connected.
Proof. This can be proven using the compactness of Sk and stereographic projection. See
the exercises of [Sti08, §8.7].
We will prove that the unit circle S1 is not simply connected. In particular, the path
(cos 2πt, sin 2πt), 0 ≤ t ≤ 1, cannot be contracted to a point.
Recall that the function f (θ) = (cos θ, sin θ) maps R onto S1 . This is called a covering
of S1 and the points f −1 (z), z ∈ S1 , are said to lie over z. In other words, the points
θ + 2nπ ∈ R lie over the point (cos θ, sin θ) ∈ S1 . This map is certainly not 1-1 and therefore
cannot be a homeomorphism.
However, if we restrict f to any interval of R with length < 2π, the result is a 1-1 map
that is continuous in both directions. Thus we say that f is a local homeomorphism.
6.1. SIMPLE CONNECTEDNESS 87
Definition 6.1.5 (Local homeomorphism). If U and V are subsets of Cn (or Rn , or, more
generally, topological spaces), then a function f : U → V is a local homeomorphism if, for
every x ∈ U , there is an open set W ⊆ U containing x such that f (W ) is open in V and
f |W : W → f (W ) is a homeomorphism.
Definition 6.1.6 (Covering space). Let V be a subset of Cn (or Rn , or, more generally, a
topological space). A covering space of V is a space Ṽ together with a continuous surjective
map f : Ṽ → V such that, for every x ∈ V , there exists an open neighbourhood U of x such
that f −1 (U ) is a disjoint union of open sets in Ṽ , each of which is mapped homeomorphically
onto U by f .
Example 6.1.7. The map f : R → S1 defined above is a covering space. For any x ∈ S1 , we
can take U to be any arc (not equal to the entire circle) containing x.
Proposition 6.1.8 (Unique path lifting). Suppose that f : Ṽ → V is a covering space, p is

a path in V with initial point P , and P̃ is a point in Ṽ over P . Then there is a unique path
p̃ in Ṽ such that
p̃(0) = P̃ , and f ◦ p̃ = p.
We call p̃ the lift of p with initial point P̃ .
Proof. The path p is a continuous function from [0, 1] to V . For every point x in the image
of p, choose a neighbourhood Ux as in Definition 6.1.6. Then the collection {p−1 (Ux )} is an
open cover of [0, 1]. Since [0, 1] is compact, there is a finite subcover. That is, the image
of p is contained in the union of a finite number of the Ux . Let us relabel this finite set by
U1 , . . . , Un in such a way that P ∈ U1 , p−1 (Ui ) ∩ p−1 (Ui+1 ) 6= ∅ for 1 ≤ i ≤ n − 1, and
p(1) ∈ Un .
Now, since U1 ∩ U2 6= ∅, we can choose an a1 ∈ [0, 1] such that p([0, a1 ]) ⊆ U1 and
p(a1 ) ∈ U2 . Let p1 = p|[0,a1 ] . Since the image of p1 is contained in U1 , there is a unique
path p̃1 : [0, a1 ] → Ṽ , with initial point P̃ , such that f ◦ p̃1 = p1 . Namely, we take p̃1 (t) =
f −1 (p1 (t)), where f −1 is the continuous inverse of f in the neighbourhood of P̃ . Let P̃1 =
p(a1 ) be the final point of p̃1 .
Similarly, there is an a2 ∈ [a1 , 1] such that p([a1 , a2 ]) ⊆ U2 and p(a2 ) ∈ U3 . Then there
is a unique path p̃2 : [a1 , a2 ] → Ṽ , with initial point P̃1 , such that f ◦ p̃2 = p2 . Let P̃2 be its
final point.
We continue in this manner and concatenate the paths p̃j in Ṽ to obtain the lift p̃ of p
with initial point P̃ .
Lemma 6.1.9. Suppose that f : Ṽ → V is a covering space and that p and q are paths from
A to B in V and p is deformable to q with endpoints fixed. Then the lift p̃ of p with initial
point Ã is deformable to the lift q̃ of q with initial point Ã with endpoints fixed.
Proof. The proof is similar to that of Proposition 6.1.8 and will be omitted.
Corollary 6.1.10. The unit circle S1 is not simply connected.

Proof. We prove that the upper semicircle path
p(t) = (cos πt, sin πt), 0 ≤ t ≤ 1,
from (1, 0) to (−1, 0) is not deformable to the lower semicircle path
q(t) = (cos(−πt), sin(−πt)), 0 ≤ t ≤ 1,
from (1, 0) to (−1, 0). The lift p̃ of p with initial point 0 has final point π. However, the lift
q̃ fo q with initial point 0 has final point −π. Hence, there is no deformation of p̃ to q̃ with
fixed endpoints. Therefore, there is no deformation of p to q with fixed endpoints.
6.2 Simply connected Lie groups

2
Since matrix Lie groups are subsets of Mn (C) and can thus be viewed as subsets of Cn , we
can apply to them the above discussion of simple connectedness. Since SU(2) can be viewed
as the group of unit quaternions, it can be viewed at the 3-sphere S3 inside R4 . Therefore,
SU(2) is simply connected. On the other hand, SO(2) can be identified with the unit circle
and therefore is not simply connected.
What about SO(3)? Each element of SO(3) is a rotation of R3 and can therefore be
described by a vector v in R3 (the axis of rotation) and an angle −π ≤ θ ≤ π (we adopt
the convention that rotation are “right-handed”). Replacing v by −v if necessary, we can
assume 0 ≤ θ ≤ π.
Let B be the closed ball of radius π in R3 , centred at the origin. Then we have a map
ψ : B → SO(3),
where ψ(u) is the rotation about the axis u through the angle |u| (and ψ(0) = I). This map
is continuous, even at the origin. It is injective except for the fact that it maps antipodal
points on the boundary of B to the same rotation.
Therefore, we can identify SO(3) (homeomorphically) with B/ ∼, where ∼ denotes the
identification of antipodal points on the boundary. It is known that B/ ∼ is homeomorphic
to the manifold RP3 (real projective space of dimension 3) and is not simply connected.
Specifically, consider a diameter of B (a line through the origin from one point on the
boundary to the antipodal point). This is a loop in B/ ∼ (since the antipodal points are
identified) that cannot be shrunk continuously to a point in B/ ∼. We will come back to
this issue, with a proof of this last point, in Section 6.4.
While we will not prove it, it turns out that we have the following.
(a) Among the compact classical groups, SU(n) and Sp(n) are simply connected, while
SO(n) (n ≥ 2) and U(n) are not.
(b) Among the other classical groups, SL(n, C) and Sp(n, C) are simply connected, while
GL(n, R)+ (n ≥ 2), GL(n, C), SL(n, R) (n ≥ 2), SO(n, C) and Sp(n, R) are not.
6.3. THREE LIE GROUPS WITH TANGENT SPACE R 89
6.3 Three Lie groups with tangent space R

We have seen that SO(2) and O(2) have the same tangent space (at the identity) and that
this tangent space is R (recall that SO(2) can be identified with the unit circle). The Lie
bracket on this tangent space is trivial since
[X, Y ] = XY − Y X = 0 for X, Y ∈ R.
Thus the Lie algebras of SO(2) and O(2) are isomorphic.
Recall that R, with vector addition, is also a matrix Lie group. Its tangent space is also
R itself (as above, the Lie bracket has to be trivial) and so it has the same tangent space as
SO(2) and O(2).
However, even though SO(2), O(2), and R are Lie groups with the same Lie algebras,
they are not isomorphic Lie groups. For instance, they have different topological properties:
• SO(2) is connected but not simply connected,
• O(2) is not connected,
• R is connected and simply connected.
They can also be distinguished algebraically:
• SO(2) has exactly two elements (±I) that square to the identity. (To see this, identify
SO(2) with the rotations in the plane.)
• O(2) has at least four elements that square to the identity:

±1 0
.
0 ±1
(In fact, O(2) has an infinite number of elements that square to the identity—all
reflections of the plane have this property.)
• R has only one element, namely 0, that ‘squares’ to the identity (remember that 0 is
the identity of R since the group operation is addition, and ‘squaring’ is really adding
a real number to itself).
Other examples of nonisomorphic Lie groups with isomorphic Lie algebras can be found
in [Sti08, Ex. 9.1.2–9.1.4]. To do these exercises, it is useful to know about the Lie algebra
of a product of Lie groups. Recall that if G and H are (matrix) Lie groups, then G × H is
also a (matrix) Lie group (see Section 1.8).
If V1 and V2 are vector spaces, then we define
V1 ⊕ V2 = {(v1 , v2 ) | v1 ∈ V1 , v2 ∈ V2 }
with vector addition being defined component wise and scalar multiplication defined by
r(v1 , v2 ) = (rv1 , rv2 ).
If g and h are Lie algebras, then g ⊕ h becomes a Lie algebra with bracket defined by
[(X1 , Y1 ), (X2 , Y2 )] = ([X1 , X2 ], [Y1 , Y2 ]), X1 , X2 ∈ g, Y1 , Y2 ∈ h.
Lemma 6.3.1. If G and H are Lie groups with Lie algebras g and h. Then the Lie algebra
of G × H is g ⊕ h.
Proof. The proof is left as an exercise (Exercise 6.3.1).
Exercises.
6.4 Three Lie groups with the cross-product Lie alge-

bra
We have seen that su(2) is isomorphic, as a Lie algebra, to R3 with the cross-product (see
Exercise 5.4.3). The Lie algebra so(3) of SO(3) is also isomorphic to R3 with the cross-
product (see [Sti08, Ex. 5.2.4]). Furthermore, we have seen that O(3) has the same Lie
algebra as SO(3).
However, the Lie groups SU(2), SO(3) and O(3) are not isomorphic. For instance, they
have different topological properties:
• O(3) is not connected,
• SO(3) is connected, but not simply connected (as we will see below),
• SU(2) is connected and simply connected (recall that SU(2) can be identified with the
three sphere of unit quaternions).
The relationship between SO(3) and O(3) is rather straightforward: SO(3) is the con-
nected component of O(3) containing the identity. What is the precise relationship between
SO(3) and SU(2)?
Recall that SU(2) can be identified with the set of unit quaternions (when we use the
presentation of quaternions as 2×2 complex matrices) and that SO(3), the group of rotations
of R3 , can be identified with pairs {±q} of unit quaternions—conjugation by a quaternion
yields a rotation of the space of unit quaternions and conjugation by q and −q produces the
same rotation (see Section 1.5). Furthermore we have the map (see Proposition 1.5.7)
SU(2) → SO(3), q 7→ {±q}.
This a 2-to-1 map that is locally 1-to-1. In fact, it is a covering space. We can now complete
the argument started in Section 6.2 that SO(3) is not simply connected. A path in SU(2)
from a unit quaternion q to its antipode −q is sent, under the above map SU(2) → SO(3),
to a loop in SO(3) since q and −q are identified under this map. By Lemma 6.1.9, this loop
cannot be contracted to a point because its lift in SU(2) cannot be contracted (the endpoints
differ). Thus SO(3) is not simply connected.
6.5. THE LIE GROUP-LIE ALGEBRA CORRESPONDENCE 91
We say that SU(2) is a double cover of SO(3). We noted before that SU(n) is simply
connected for all n. It turns out that all the groups SO(n), n ≥ 3, are doubly covered by
simply connected groups. These double covers are called the spin groups Spin(n).
In low dimensions, there are isomorphisms among the classical Lie groups (“accidental
isomorphisms”). In particular, there are isomorphisms between low dimensional spin groups
and certain classical groups. Specifically, we have
• Spin(3) ∼
= Sp(1) = SU(2),
• Spin(4) ∼
= Sp(1) × Sp(1),
• Spin(5) ∼
= Sp(2),
• Spin(6) ∼
= SU(4).
The Spin groups play an important role in physics. In general, the Spin groups can be
constructed as a certain group of invertible elements in the so-called Clifford algebra over
Rn . In particular, Spin(n) is a matrix Lie group.
6.5 The Lie group-Lie algebra correspondence

In this section, we will discuss a (partial) converse to Theorem 5.13.1. It turns out that we
need to add an extra assumption: that the matrix Lie group be simply connected.
Theorem 6.5.1. Suppose G and H are matrix Lie groups with Lie algebras g and h. Let
φ : g → h be a Lie algebra homomorphism. If G is simply connected, then there exists
a unique Lie group homomorphism Φ : G → H such that Φ(exp X) = exp(φ(X)) for all
X ∈ g.
Before proving this theorem, we state one immediate corollary, which serves as partial
motivation.
Corollary 6.5.2. Two simply connected Lie groups with isomorphic Lie algebras are iso-
morphic.
Proof. Suppose G and H are simply connected Lie groups with isomorphic Lie algebras g and
h. Let φ : g → h be a Lie algebra isomorphism. By Theorem 6.5.1, we have a corresponding
Lie group homomorphism Φ : G → H. Since φ−1 : h → g is also a Lie algebra homomorphism,
there exists a corresponding Lie group homomorphism Ψ : H → G. It remains to show that
Ψ ◦ Φ = idG and Φ ◦ Ψ = idH .
Since our association of Lie algebras to Lie groups is a functor, the Lie algebra homo-
morphism associated to Ψ ◦ Φ is φ−1 ◦ φ = idg . Therefore, by Proposition 5.13.7, we have
Ψ ◦ Φ = idG . Similarly, Φ ◦ Ψ = idH .
Proof of Theorem 6.5.1. Our proof will proceed in steps.
1. Define Φ in a neighbourhood of the identity.

2. Define Φ on an arbitrary element via a path and using a partition of the path.
3. Prove independence of the partition.
4. Prove independence of the path
5. Prove Φ is a homomorphism and Φ(exp X) = exp(φ(X)).
Step 1. We know there exists a neighbourhood V of the identity in G such that the
exponential mapping has an local inverse mapping V into the Lie algebra g. We make V
small enough so that for all A, B ∈ V , the Campbell–Baker–Hausdorff Theorem applies to
log A and log B. This neighbourhood V will be fixed for the remainder of the proof.
On V , we define Φ : V → H by
Φ(A) = exp(φ(log A)), A ∈ V.
In other words, on V , we have Φ = exp ◦φ ◦ log. This is the only possible definition if we
want to satisfy Φ(exp(X)) = exp(φ(X)) for all X ∈ g.
Step 2. We now want to extend the definition of Φ to all of G. Let A ∈ G. Since G is

path-connected, there exists a path A : [0, 1] → G with A(0) = I and A(1) = A. Using a
compactness argument similar to the one used in Corollary 5.11.8, there exist numbers
0 = t0 < t1 < t2 < · · · < tm = 1
such that for each i = 0, . . . , m − 1,
ti ≤ s ≤ t ≤ ti+1 =⇒ A(t)A(s)−1 ∈ V. (6.1)
Note that this implies, in particular, that A(t1 ) ∈ V since t0 = 0 and A(0) = I. Now, we
write A = A(1) in the form
A = A(1)A(tm−1 )−1 A(tm−1 )A(tm−2 )−1 · · · A(t2 )A(t1 )−1 A(t1 ).

Since we want Φ to be a homomorphism, we then define
Φ(A) = Φ A(1)A(tm−1 )−1 Φ A(tm−1 )A(tm−2 )−1 · · · Φ A(t2 )A(t1 )−1 Φ(A(t1 )),

(6.2)
where each factor is defined in Step 1, since the argument lies in V .
Step 3. Our definition above involved a partition of the path. We would like to show
that the definition is actually independent of this choice. It is in this step that we use the
Campbell–Baker–Hausdorff Theorem. We will show that “refining” the partition (that is,
adding additional partition points) does not change the definition of Φ(A). Do do this, it
suffices to show that the result is not changed under the addition of a single partition point.
Since any two partitions have a common refinement (for instance, take the union of the two
sets of partition points), the result will follow. Recall the definition of the Lie polynomials
Fn given in (5.6).
6.5. THE LIE GROUP-LIE ALGEBRA CORRESPONDENCE 93
Note that if a given partition satisfies (6.1), then any refinement of that partition also
satisfies this condition. Suppose we insert an extra partition point s between ti and ti+1 .
Then, the factor Φ(A(ti+1 )A(ti )−1 ) in (6.2) will be replaced by
Φ A(ti+1 )A(s) Φ A(s)A(ti )−1 .

Since ti ≤ s ≤ ti+1 , (6.1) implies that
A(ti+1 )A(s)−1 , A(s)A(ti )−1 , A(ti+1 )A(ti )−1 ∈ V.
Thus, the result would follow if we knew that Φ was a homomorphism on V (as defined in
Step 1). Now, since V is in the image of the exponential map, any element of V can be
written in the form eX for some X ∈ g. We have
P
X Y Fn (X,Y )
Φ(e e ) = Φ e n≥1
= eφ( Fn (X,Y ))
P
n≥1
P
Fn (φ(X),(Y ))
=e n≥1
= eφ(X) eφ(Y )
= Φ(eX )Φ(eY ),
where, in the third equality, we used the fact that φ is a Lie algebra homomorphism and the
Fn are Lie polynomials. Thus Φ is a homomorphism on V and we are done this step.
Step 4. We now need to prove that the definition of Φ(A) is independent of the path we
chose. This is where we used the fact that G is simply connected.
Suppose that A0 , A1 : [0, 1] → G are two paths from I to A. Then A0 and A1 are
homotopic with endpoints fixed. Namely, there exists a continuous map
A : [0, 1] × [0, 1] → G
such that
A(0, t) = A0 (t), A(1, t) = A1 (t) for all t ∈ [0, 1],
and
A(s, 0) = I, A(s, 1) = A for all s ∈ [0, 1].
By an compactness argument similar to the one used in Step 1, there exists an integer N
such that for all (s, t), (s0 , t0 ) ∈ [0, 1] × [0, 1] with |s − s0 | ≤ 2/N and |t − t0 | ≤ 2/N , we have
A(s, t)A(s0 , t0 )−1 ∈ V .
We will now gradually deform A0 into A1 by defining a sequence of paths Bk,` (t), k =
0, . . . , N − 1, ` = 0, . . . , N . We define Bk,` (t) as follows. If ` > 0, let

k+1
0 ≤ t ≤ `−1

A N , t
 N
,
`−1 `
Bk,` (t) = “diagonal” (see picture) N ≤ t ≤ N ,

A Nk , t `

≤ t ≤ 1.

N
For ` = 0, we define
k
Bk,0 (t) = A ,t , t ∈ [0, 1].
N
In particular, B0,0 (t) = A0 (t).
`/N
(` − 1)/N
Bk,`
s
k/N (k + 1)/N 1
We will now deform A0 into A1 in steps. First, we deform A0 = B0,0 into B0,1 , then
into B0,2 , B0,3 , etc. until we reach B0,N . Then we deform this into B1,0 , B1,1 , . . . , B1,N . We
continue until we reach BN −1,N , which we finally deform into A1 . We want to show that
Φ(A) as computed on each of these paths is equal to Φ(A) as computed using the next one.
Now, for ` < N , the paths Bk,` (t) and Bk,`+1 (t) are the same except for

`−1 `+1
t∈ , .
N N
Now, choose the partition
1 `−1 `+1 `+2

0, ,..., , , , . . . , 1.
N N N N
Then by our choice of N , this partition satisfies Condition 6.1. Therefore, since our definition
of Φ(A) depends only on the value of the path at the partition points, and Bk,` (t) and
Bk,`+1 (t) agree on the partition points, the value of Φ(A) is the same for these two paths.
A similar argument shows that the value of Φ(A) computed along Bk,N is the same as
along Bk+1,0 . Therefore, the value of Φ(A) is the same for each of our paths Bk,` and then
(by the same argument) the same as A1 . Thus we have shown independence of path.
6.6. COVERING GROUPS 95
Step 5. One can show that Φ is a homomorphism as follows. For A, B ∈ G, choose path
A(t) and B(t) from I to A and B, respectively. Then construct a path C(t) from I to AB
by (
C(t) = A(2t) t ∈ [0, 1/2],
C(t) =
C(t) = AB(2t − 1) t ∈ [1/2, 1].
Then it follows from the definition of Φ above that Φ(AB) = Φ(A)Φ(B) (see Exercise 6.5.1).
Now, since Φ is defined near the identity as Φ = exp ◦φ ◦ log, we have
d d tφ(X)
Φ etX

t=0
= e t=0
= φ(X).
dt dt
Therefore φ is the Lie algebra homomorphism associated to the Lie group homomorphism
Φ, as desired.
Exercises.
6.5.1. Complete Step 5 of the proof of Theorem 6.5.1 by showing that Φ(AB) = Φ(A)Φ(B).
6.6 Covering groups

Definition 6.6.1 (Universal covering group). Suppose G is a connected Lie group. Then
a universal covering group (or universal cover ) of G is a simply connected Lie group G̃
together with a Lie group homomorphism Φ : G̃ → G such that the associated Lie algebra
homomorphism φ : g̃ → g is a Lie algebra isomorphism. The homomorphism Φ is called the
covering homomorphism (or projection map).
Note that this theorem does not specify that the Lie groups be matrix Lie groups. In
fact, the universal cover of matrix Lie group may be a nonmatrix Lie group. For instance, the
^C) of SL(2, C) is a nonmatrix Lie group. However, universal
universal covering group SL(2,
covering groups always exist, as the next theorem tells us.
Theorem 6.6.2 (Existence and uniqueness of universal covering groups). For any connected
Lie group, a universal cover exists. If G is a connected Lie group and (H1 , Φ1 ) and (H2 , Φ2 )
are universal covers of G, then there exists a Lie group isomorphism Ψ : H1 → H2 such that
Φ2 ◦ Ψ = Φ1 . In other words, the diagram
H1 Ψ / H2
Φ1 ~ Φ2
G
commutes.
Proof. We will not prove the existence part of this theorem (see [Hal03, App. C]). Uniqueness
follows from Theorem 6.5.1 as follows. Suppose (H1 , Φ1 ) and (H2 , Φ2 ) are universal covers
of G. Then the corresponding Lie algebra maps φ1 , φ2 are isomorphisms. By Theorem 6.5.1,
there exists a Lie group homomorphism Ψ : H1 → H2 corresponding to the isomorphism
φ−1
2 ◦ φ1 : h1 → h2 of their Lie algebras and a Lie group homomorphism Ψ : H2 → H1
0
−1 0
corresponding to φ1 ◦ φ2 . Then, the composition Ψ ◦ Ψ : H1 → H1 corresponds to the
identity Lie algebra homomorphism. By the uniqueness statement in Theorem 6.5.1, we
have Ψ0 ◦ Ψ = idH1 . Similarly, Ψ ◦ Ψ0 = idH2 . Again, by the uniqueness statement in
Theorem 6.5.1, we have Φ2 ◦ Ψ = Φ1 .
Remark 6.6.3. (a) Since the universal cover of a Lie group G is unique up to isomorphism,
we often speak of the universal cover.
(b) If G̃ is a simply connected Lie group and φ : g̃ → g is a Lie algebra isomorphism, then
by Theorem 6.5.1, there exists a Lie group homomorphism Φ : G̃ → G inducing φ. This
yields a universal cover of G. Using φ, we can identify g̃ and g. Thus, by a slight abuse
of terminology, we sometimes think of the universal cover of G as the unique simply
connected Lie group with the same Lie algebra as G (we use the identity map g → g to
construct the covering homomorphism).
(c) A covering group (not necessarily universal) of a connected Lie group G is a connected
Lie group H (not necessarily simply connected) together with a Lie group homomor-
phism Φ : H → G such that the associated Lie algebra homomorphism φ : h → g is an
isomorphism. In general, a Lie group can have many nonisomorphic covering groups,
with different fundamental groups.
Examples 6.6.4. (a) G = S1 . The universal cover is R and the covering homomorphism is
the map R → S1 given by θ 7→ eiθ .
(b) G = SO(3). The universal cover is SU(2) and we have described the covering homomor-
phism in Section 1.5.
(c) G = U(n). The universal cover is R × SU(n) and the covering homomorphism is the
map
R × SU(n) → U(n), (θ, U ) 7→ eiθ U. (6.3)
Since R and SU(n) are simply connected, R×SU(n) is simply connected. The Lie algebra
homomorphism associated to (6.3) is
R ⊕ su(n) 7→ u(n), (r, A) 7→ irI + A, r ∈ R, A ∈ su(n). (6.4)
It thus suffices to check that (6.4) is a Lie algebra isomorphism and that (6.3) is a Lie
group homomorphism (Exercise 6.6.1).
(d) As discussed before, the universal cover of SO(n), n ≥ 3, is Spin(n) and these are double
covers (i.e. the covering homomorphisms are 2-to-1).
6.7. SUBGROUPS AND SUBALGEBRAS 97
Remark 6.6.5. In the above examples, the universal covers are all matrix Lie groups. In
general, one can show that the universal cover of a compact matrix Lie group is again a
matrix Lie group (not necessarily compact).
Remark 6.6.6. The universal covering construction is inverse to the process of taking the
quotient by a discrete subgroup because the kernel of the covering homomorphism G̃ → G
is a discrete subgroup of G̃, called the fundamental group of G. It can be shown that the
fundamental group of a Lie group is always abelian, even though the fundamental group of an
arbitrary smooth manifold can be any finitely presented group (finite number of generators
and relations). Thus, the topology of Lie groups is rather constrained.
Remark 6.6.7. If a simply connected Lie group with Lie algebra g has a discrete centre, then
the set of all connected Lie groups with Lie algebra g form a lattice, corresponding to the
lattice of subgroups of the centre of the simply connected Lie group. For example, this
happens with g = sl(2, F ), where F = R or C. The Lie group SL(2, F ) has universal cover
^
SL(2, ^
F ) and every Lie group with Lie algebra g has SL(2, F ) as its universal cover and is a
cover of PSL(2, F ) := SL(2, F )/Z(SL(2, F )), where Z(SL(2, F )) = {±1}.
Exercises.
6.6.1. Prove the following:
(a) The map (6.4) is a Lie algebra isomorphism.
(b) The map (6.4) is the Lie algebra homomorphism corresponding to the map (6.3).
(c) The map (6.3) is a Lie group homomorphism.
6.7 Subgroups and subalgebras

Given the close connections between Lie groups and Lie algebras, it seems natural to expect
that there is a relationship between the subgroups of a Lie group and the subalgebras of its
Lie algebra. If G and H are matrix Lie groups with H ⊆ G, then it is clear that h is a Lie
subalgebra of g. In general the converse is false, as the following example illustrates.
Example 6.7.1. Consider the matrix Lie group GL(2, C), with Lie algebra gl(2, C). If we fix
an irrational number a, then

it 0
h= t∈R (6.5)
0 ita
is a one-dimensional real subalgebra of gl(2, C). Any matrix Lie group H with Lie algebra
h would have to contain
it
e 0
H0 = exp(h) = t∈R .
0 eita
Being a matrix Lie group, H would need to be closed in GL(2, C). But the closure of H0 in
GL(2, C) is it
e 0
H1 = s, t ∈ R .
0 eis
Then h would have to contain the Lie algebra of H1 , which is two dimensional. This is a
contradiction. Thus, there is no matrix Lie group corresponding to h.
The problem in Example 6.7.1 is due to the fact that we are restricting our attention to
matrix Lie groups. We can resolve the issue by relaxing this condition.
Definition 6.7.2 (Lie algebra of an arbitrary matrix group). Suppose H is any subgroup
of GL(n, C). We define the Lie algebra of H to be the set
{X ∈ Mn (C) | etX ∈ H for all t ∈ R}. (6.6)
Definition 6.7.3 (Connected Lie subgroup). Suppose G is a matrix Lie group with Lie
algebra g. We say H ⊆ G is a connected Lie subgroup (or analytic subgroup) of G if the
following conditions are satisfied:
(a) H is a subgroup of G.
(b) The Lie algebra h of H is a subspace of g.
(c) Every element of H can be written in the form eX1 eX2 · · · eXm , with X1 , . . . , Xm ∈ h.
Example 6.7.4. The group H0 of Example 6.7.1 is a connected Lie subgroup of GL(2, C) with
Lie algebra h given by (6.5).
The word connected in Definition 6.7.3 is justified by the following result.
Proposition 6.7.5. If H is a connected Lie subgroup of a matrix Lie group G, then H is

path-connected.
Proof. It suffices to show that any element of H can be connected to the identity by a
continuous path taking values in H. We will prove, by induction on m, that any element of
H of the form
h = eX1 eX2 · · · eXm , m ∈ N, X1 , . . . , Xm ∈ h,
can be connected to the identity by a continuous path lying in H. Since H is a connected
Lie subgroup of G, all elements of H are of this form.
For m = 0, we have h = 1 and the result is trivial. Assume m ≥ 1 and that the result is
true for m − 1. For X1 , . . . , Xm ∈ h, define the continuous path
h(t) = eX1 eX2 · · · eXm e−tXm = eX1 eX2 · · · e(1−t)Xm .

6.7. SUBGROUPS AND SUBALGEBRAS 99
By the definition of h, we have e−tXm ∈ H for all t ∈ R. Thus h(t) is contained in H. We

have h(0) = eX1 eX2 · · · eXm and h(1) = eX1 eX2 · · · eXm−1 . By the induction hypothesis, there
is a path, contained in H, connecting h(1) to the identity. This completes the proof of the
inductive step.
Proposition 6.7.6. Suppose H is a connected Lie subgroup of a matrix Lie group G. Then
the Lie algebra h of H is a subalgebra of the Lie algebra g of G.
Proof. If A ∈ H and Y ∈ h, then, for all t ∈ R, we have

−1
etAY A = AetY A−1 ∈ H.
Thus AY A−1 ∈ h. It follows that, for X, Y ∈ h,
etX Y e−tX ∈ h for all t ∈ R.
Since h is a vector space, it is a topologically closed subset of Mn (C). Therefore, for X, Y ∈ h,
d ehX Y e−hX − Y
[X, Y ] = XY − Y X = etX Y e−tX = lim ∈ h.
dt t=0
h→0 h
Therefore h is a Lie subalgebra of g.
Theorem 6.7.7. Suppose G is a matrix Lie group with Lie algebra g, and let h be a Lie
subalgebra of g. Then
H = {eX1 eX2 · · · eXm | X1 , . . . , Xm ∈ h}
is the unique connected Lie subgroup H of G such that the Lie algebra of H is h.
We will not prove Theorem 6.7.7. A proof can be found in [Hal03, §3.8]. The connected
Lie subgroup H corresponding to the subalgebra h in Theorem 6.7.7 may or may not be
matrix Lie group. It is a matrix Lie group precisely when H is a closed subset of G.
However, we have the following result.
Theorem 6.7.8. Suppose that G is a matrix Lie group and that H is a connected Lie
subgroup of G. Then H can be given the structure of a Lie group in such a way that the
inclusion of H into G is a Lie group homomorphism.
Proof. We give only a sketch of the proof. We first note that the topology on H induced from
the topology on G can be very bad. For example, H might not even be locally connected in
this topology. So, in general, we need a different topology on H. For any A ∈ H and ε > 0,
define
UA,ε = {AeX | X ∈ h, kXk < ε}.
We then define a topology on H by using the UA,ε as basic open sets. More precisely, we
define a subset U ⊆ H to be open if, for all A ∈ U , there exists ε > 0 such that UA,ε ⊆ U .
In fact, H can be made into a smooth manifold by using the sets UA,ε as basic coordinate
neighbourhoods, and using the maps X 7→ AeX as local coordinate maps. One can then show
that the product and inverse maps on H are smooth with respect to this smooth manifold
structure.
Finally, one can show that any set of H that is open in the topology induced from the
topology of G is also open in the topology defined above. It follows that the inclusion of G
into H is continuous (the preimage of open sets are open).
Theorem 6.7.9. Every finite-dimensional real Lie algebra is isomorphic to the Lie algebra
of some Lie group.
Proof. We give only a sketch of the proof. Ado’s Theorem (which we have not proven)
states that every finite-dimensional real Lie algebra is isomorphic to a subalgebra of gl(n, R).
Theorem 6.7.8 then implies that there is a Lie group with this Lie algebra. We are implicitly
using here the fact that the more abstract definition of the Lie algebra of a (general) Lie
group coincides with Definition 6.7.2 for connected Lie subgroups.
Proposition 6.7.10. The association of the Lie algebra to a Lie group yields a one-to-
one correspondence between isomorphism classes of simply connected (real) Lie groups and
isomorphism classes of finite-dimensional real Lie algebras.
Proof. Let F be the map from the set of isomorphism classes of simply connected Lie groups
to the set of isomorphism classes of real Lie algebras given by taking the Lie algebra of a
Lie group. Then F is injective by Corollary 6.5.2. Suppose g is a real Lie algebra. Then, by
Theorem 6.7.9, there exists a Lie group G with Lie algebra g. By Theorem 6.6.2, G has a
universal cover G̃ with Lie algebra g. Thus F is surjective.
Another way of summarizing some of the results above is that when we restrict the functor
from Lie groups to Lie algebras given in Theorem 5.13.1 to the (full) subcategory of simply
connected Lie groups, it yields an equivalence of categories between the category of simply
connected Lie groups and the category of finite-dimensional real Lie algebras.
Chapter 7
Further directions of study
In this course, we have only scratched the surface of the topic of Lie groups. There is much
more to learn. In this final chapter we give a brief (and not, by any means, comprehensive)
overview of some further directions of study.
7.1 Lie groups, Lie algebras, and quantum groups

One obvious direction of further study is to expand one’s focus to all Lie groups, instead of
just matrix Lie groups. As we saw in Chapter 6, this more general viewpoint is necessary
if one wants to fully discuss topics just as covering groups, quotients by normal subgroups
(for example, there is a quotient of the Heisenberg group that is not a matrix Lie group—see
[Sti08, p. 72]), etc.
The study of Lie algebras themselves is also a very active area of research. The finite-
dimensional simple complex Lie algebras have been classified. A more general class of Lie
algebras are the so-called Kac–Moody algebras. The affine Lie algebras can, in some sense,
be viewed as the next step beyond finite-dimensional simple Lie algebras. They play an
important role in physics and the theory of vertex algebras. The books [EW06, Hum78] are
good references for the finite-dimensional theory, while [Kac90] is a standard reference for
the infinite-dimensional setting.
To any Lie algebra, one can associated its universal enveloping algebra. It is an associative
algebra that has the same representation theory as the original Lie algebra. These universal
enveloping algebras can be q-deformed to produce quantum groups or quantized enveloping
algebras. Taking the parameter q to 1 recovers the original enveloping algebra. On the other
hand, taking q to zero results in a structure known as a crystal , where algebra is replaced by
combinatorics. This construction plays a central role in combinatorial representation theory.
The books [HK02, Kas95] are good references for these topics.
7.2 Superalgebras and supergroups

An interesting and active area of research is the generalization from the “standard” setting to
the “super” setting. A supermanifold is similar to a manifold, except that instead of looking
locally like Rn for some n, it looks like the superspace Rm|n . The variables corresponding to
101
102 CHAPTER 7. FURTHER DIRECTIONS OF STUDY
the n “odd” coordinates supercommute (xy = −yx) instead of commute. A Lie supergroup
is then a group object in the category of supermanifolds (see Definition 1.1.4).
The tangent space of a Lie supergroup is naturally a Lie superalgebra. A Lie superalgebra
is a Z2 -graded vector space g = g0 + g1 with a bilinear operation [·, ·] : g × g → g (the super
Lie bracket) satisfying the following conditions:
(a) It respects the grading: [gi , gj ] ⊆ gi+j for i, j ∈ Z2 ,
(b) It is super skew-symmetric: [x, y] = −(−1)|x||y| [y, x] for all homogeneous x, y ∈ g (here
|x| denotes the parity of x, i.e. x ∈ g|x| ).
(c) It satisfies the super Jacobi identity:
[x, [y, z]] = [[x, y], z] + (−1)|x||y| [y, [x, z]] for all homogeneous x, y, z ∈ g.
A Lie superalgebra g with g1 = 0 is simply a Lie algebra. The study of Lie superalgebras
and Lie supergroups plays an important role in supersymmetry in physics. Many results in
Lie algebras have been generalized to Lie superalgebras (for instance, the classification of
the complex simple Lie superalgebras). However, there are also many examples of theorems
known in the Lie algebra case that have not yet been extended to the Lie superalgebra case.
7.3 Algebraic groups and group schemes

Moving from the setting of smooth manifolds to the setting of algebraic geometry results
in a passage from Lie groups to algebraic groups. More precisely, an algebraic group is a
group object in the category of algebraic varieties (see Section 1.1). Explicitly, an algebraic
group (or group variety) is a group that is an algebraic variety, such that the multiplication
and inversion operations are given by regular functions on the variety. Over a field, affine
algebraic groups (where the algebraic variety is an affine variety) are matrix groups. Thus,
over the real numbers, the definition of an algebraic group is more restrictive than the
definition of a Lie group.
Schemes are generalizations of algebraic varieties. A group object in the category of
schemes is called a group scheme. There is a well-developed theory of group schemes and
this topic continues to be an active area of research.
7.4 Finite simple groups

Killing and Cartan classified the simple Lie groups. This helped pave the way for a classi-
fication of the finite simple groups. Each matrix Lie group G gives rise to infinitely many
finite groups by replacing the entries of G by elements of a finite field (say, the integers
modulo some prime p). Since there is a finite field of size q for each prime power q, there are
infinitely many finite groups corresponding to each infinite matrix Lie group. In fact, each
simple Lie group yields infinitely many finite simple groups in this way. These are called the
simple finite groups of Lie type.
7.5. REPRESENTATION THEORY 103
Theorem 7.4.1 (Classification of finite simple groups). Every finite simple group is
• a cyclic group of prime order,
• an alternating group of degree at least 5,
• a simple group of Lie type, or
• one of 26 sporadic groups.
The largest sporadic group is called the Monster group.
7.5 Representation theory

Once one has learned about the structure of Lie groups themselves, it is natural to investigate
their representations. A (real) representation of a Lie group G is a Lie group homomorphism
G → GL(n, R) or G → GL(V ),
where V is a real vector space. Complex representations are defined similarly.

Representations can be thought of as an action of a group on a vector space. The term
“representation” comes from this interpretation—each element of G is “represented” as a
linear map on a vector space.
We now know that each complex representation of complex Lie group G yields a Lie
algebra homomorphism
g → gl(n, C) or g → gl(V ),
called a representation of g. Often one can answer questions about representations of Lie
groups by studying representations of the corresponding Lie algebra, which are often much
easier to study. The later chapters of [Hal03] concern the representation theory of Lie groups.
The study of representations of finite groups (e.g. permutation groups) is also an impor-
tant and interesting topic and does not require knowledge of Lie groups or Lie algebras.
Index
3-sphere of unit quaternions, 8 covering group, 96

covering homomorphism, 95
absolute value covering space, 87
of a matrix, 52 crystal, 101
absolute value of quaternions, 7
adjoint mapping, 76 δ-neighbourhood, 69
affine Lie algebra, 101 derivative of a path, 56
algebraic group, 102 differentiable path, 55
alternating, 51 Dirac’s belt trick, 14
analytic subgroup, 98 direct product, 17
discrete subgroup, 46
C∗ , 26 distance between matrices, 46
Campbell–Baker–Hausdorff Theorem, 81 dot product, 21
Cartan–Dieudonné Theorem, 14 double cover, 14, 91
category, 1
centres, 44 E(n), 26
circle group, 3 Euclidean group, 26
circle Lie group, 4 Euclidean space, 3
classical groups, 44 exponential map
classical Lie groups, 20 to SO(2), 47
Clifford algebra, 91 to SU(2), 48
compact, 36 exponential of a matrix, 53
compact classical groups, 44 finite group, 21
compact symplectic group, 24 finite simple groups, 102
complex general linear group, 21 functor, 73
complex matrix Lie group, 61 fundamental group, 97
complex orthogonal group, 23
complex special linear group, 21 gauge group, 3
complex symplectic group, 24 general linear group, 3
complexification GL(n, C), 21
of a Lie algebra, 62 gl(n, C), 60
of a vector space, 61 GL(n, R), 21
component, 29 gl(n, R), 60
composition, 1 group, 2
conjugation, 10 circle, 3
connected component, 29 linear, 3
connected Lie subgroup, 98 simple, 14
convergence of matrices, 20 group action, 103
104
INDEX 105
group object, 2 complex, 61

group scheme, 102 matrix logarithm, 64
group variety, 102 maximal torus, 39
groups of Lie type, 102 metaplectic group, 25
Monster group, 103
Heisenberg group, 3, 25 morphism, 1
homeomorphism, 70 Mp(n, R), 25
homomorphism of Lie groups, 27
hyperplane, 14 neighbourhood, 69
nilpotent matrix, 55
ideal of a Lie algebra, 51 normal subgroup, 14, 45, 78
inner product on Rn , 21 normal subgroups
inverse, 2
of SO(n), SU(n), Sp(n), 79
isometry, 14
isomorphism of Lie groups, 27 object, 1
group, 2
J, 24
terminal, 1
J2n , 25
O(n), 12, 22, 26
Jacobi identity, 51
o(n), 60
Kac–Moody algebra, 101 O(n, C), 23
o(n, C), 60
Lie algebra, 51 one-parameter subgroup, 72
alternate description, 70 orientation-preserving, 15
of a Lie group, 60 orientation-reversing, 15
of arbitrary matrix group, 98 orthogonal group, 3, 12, 22, 26
Lie bracket, 50, 51
Lie group, 3, 37 path, 29
Lie group homomorphism, 27 derivative, 56
Lie group isomorphism, 27 differentiable, 55
lie over, 86 smooth, 55
Lie polynomial, 81 path-connected, 29
Lie product formula, 71 Poincaré group, 3
Lie superalgebra, 102 polar decomposition, 33
Lie supergroup, 102 SL(n, C), 35
lift, 87 SL(n, R), 34
linear group, 3, 4, 20 positive definite, 34, 35
linear Lie group, 20 positive matrix, 34, 35
local homeomorphism, 87 positive semidefinite, 35
log, 64 projection map, 95
logarithm, 64 pure imaginary quaternions, 9
Lorentz group, 3
quantized enveloping algebra, 101
matrix absolute value, 52 quantum group, 101
matrix exponential, 53 quaternion conjugate, 8
matrix group, 4, 20 quaternions, 6
matrix Lie group, 3, 20 3-sphere, 8
106 INDEX
absolute value, 7 sp(n, R), 60

pure imaginary, 9 sporadic group, 103
su(2), 51
R∗ , 26 subgroup
real form, 64 discrete, 46
real general linear group, 21 normal, 14, 45, 78
real special linear group, 21 submultiplicative property, 52
real symplectic group, 24 SU(n), 8, 22
reflection in a hyperplane, 14 su(n), 60
representation, 103 supermanifold, 101
Rn , 26 symmetric matrix, 33
rotation, 15 symplectic group, 24
complex, 24
S1 , 4, 26 real, 24
scheme, 102
Schreier’s Theorem, 46 tangent
sequential tangent vector, 67 sequential, 67
simple group, 14, 45 tangent space, 56
simple groups, 102 of SU(2), 49
simple Lie algebra, 51 tangent space visibility, 78
simplicity of SO(2m + 1), 79 tangent vector, 56
simplicity of so(n), su(n), sp(n), 79 terminal object, 1
simply connected, 85 topological dimension, 70
skew-Hermitian, 57 torus, 39
skew-symmetric, 57 translation, 26
SL(n, C), 21
U(n), 8, 22
slnCsl(n, C), 60
u(n), 60
SL(n, R), 21
unitary group, 3, 8, 22
sl(n, R), 60
universal cover, 95
smooth path, 55
universal covering group, 95
SO(2), 4
universal enveloping algebra, 101
SO(n), 12, 22
so(n), 60
SO(n, C), 23
so(n, C), 60
special complex orthogonal group, 23
special linear group, 3
special orthogonal group, 4, 12, 22
special unitary group, 8, 22
spin groups, 91
Sp(n), 24
sp(n), 60
Sp(n, C), 24
sp(n, C), 60
Sp(n, R), 24
Bibliography
[EW06] Karin Erdmann and Mark J. Wildon. Introduction to Lie algebras. Springer Un-
dergraduate Mathematics Series. Springer-Verlag London, Ltd., London, 2006.
[Hal03] Brian C. Hall. Lie groups, Lie algebras, and representations, volume 222 of Grad-
uate Texts in Mathematics. Springer-Verlag, New York, 2003. An elementary
introduction.
[HK02] Jin Hong and Seok-Jin Kang. Introduction to quantum groups and crystal bases,
volume 42 of Graduate Studies in Mathematics. American Mathematical Society,
Providence, RI, 2002.
[Hum78] James E. Humphreys. Introduction to Lie algebras and representation theory, vol-
ume 9 of Graduate Texts in Mathematics. Springer-Verlag, New York-Berlin, 1978.
Second printing, revised.
[Kac90] Victor G. Kac. Infinite-dimensional Lie algebras. Cambridge University Press,

Cambridge, third edition, 1990.
[Kas95] Christian Kassel. Quantum groups, volume 155 of Graduate Texts in Mathematics.
Springer-Verlag, New York, 1995.
[Sti08] John Stillwell. Naive Lie theory. Undergraduate Texts in Mathematics. Springer,
New York, 2008.
107

MAT4144 5158 LieGroups

Uploaded by

Copyright:

Available Formats

MAT4144 5158 LieGroups

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MAT4144 5158 LieGroups

Uploaded by

Copyright:

Available Formats

Introduction to Lie Groups

Department of Mathematics and Statistics

This work is licensed under a

1 Introduction and first examples 1

2 Matrix Lie groups 20

3 Topology of Lie groups 29

4 Maximal tori and centres 39

5 Lie algebras and the exponential map 47

7 Further directions of study 101

Alistair Savage Ottawa, 2015.

Course website: http://alistairsavage.ca/mat4144/

Introduction and first examples

1.1 Category theoretic definitions

Definition 1.1.1 (Category). A category C consists of

• for each two objects A, B ∈ ob C, a class hom(A, B) of morphisms between them,

• for every three objects A, B, C ∈ ob C, a binary operation

hom(B, C) × hom(A, B) → hom(A, C)

called composition and written (f, g) 7→ f ◦ g

such that the following axioms hold:

• Associativity. If f ∈ hom(A, B), g ∈ hom(B, C) and h ∈ hom(C, D), then h ◦ (g ◦ f ) =

Definition 1.1.2 (Terminal object). An object T of a category C is a terminal object if there

Objects Morphisms Terminal object(s) Product

where p1 : G × 1 → G and p2 : 1 × G → G are the canonical projections.

Category Group objects

(c) General linear group GL(n, R) with matrix multiplication.

(d) Special linear group SL(n, R) with matrix multiplication.

(f) Unitary group U(n) and special unitary group SU(n).

1.2 The circle: S1

This rotation corresponds to the map

Rotating first by θ and then by ϕ corresponds to multiplication by Rθ and then by Rϕ or,

zθ (x + iy) = (cos θ + i sin θ)(x + iy)

Composition of rotations corresponds to multiplication of complex numbers since rotating

1.2.2. Prove that the map zθ 7→ Rθ , θ ∈ R, is a group isomorphism from S1 to SO(2).

1.3 Matrix representations of complex numbers

Theorem 1.3.1. The map

it is also a ring homomorphism.

Note that thesquaredabsolute value |a + bi|2 = a2 + b2 is the determinant of the corre-

Note also that if z ∈ C corresponds to the matrix A, then

Remark 1.4.2. Note that C is a subring of H (spanned by 1 and i).

|q1 q2 | = |q1 ||q2 | for all q ∈ H.

If q = a + bi + cj + dk, then q̄ := a − bi − cj − dk is called the quaternion conjugate of q.

Definition 1.4.3 (U(n) and SU(n)). A matrix X ∈ Mn (C) is called unitary if X ∗ X = In .

Remark 1.4.4. Note that X ∗ X = I implies that | det X| = 1.

Proposition 1.4.5. The group S3 of unit quaternions is isomorphic to SU(2).

and the result follows.

1.5 Quaternions and space rotations

where u × v is the vector cross product.

So if u is a unit vector in Ri + Rj + Rk, then u2 = −1. So every unit vector in Ri + Rj + Rk

t = t0 + tI , 1 = |t|2 = t20 + t21 + t22 + t23 = t20 + |tI |2 .

t0 = cos θ, |tI | = sin θ

Proposition 1.5.1. Let t = cos θ + u sin θ, where u ∈ Ri + Rj + Rk is a unit vector. Then

Proof. Note that

twt−1 = (cos θ + u sin θ)w(cos θ − u sin θ)

Therefore, rotation of R3 through an angle α about the axis u is given by conjugation by

and so all rotations of R3 arise as conjugation by a unit quaternion.

which is rotation through an angle of α + 2π about u, which is the same transformation.

Remark 1.5.4. Note that XX T = In implies X T X = In and det X = ±1.

Proposition 1.5.5. The rotations of R3 form a group isomorphic to SO(3).

v T w = (Xv)T (Xw) = v(X T X)w.

Note that thesquaredabsolute value |a + bi|2 = a2 + b2 is the determinant of the corre-