Lectures on Representation Theory

Contents
1 Representations of compact groups 2
1.1 Basic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3 The Peter–Weyl theorem for finite groups . . . . . . . . . . . . . . . . . . . . 17
1.4 Compact groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.5 Characters for compact groups . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.6 The Peter–Weyl theorem for compact groups . . . . . . . . . . . . . . . . . . 29
1.7 Exercise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2 Lie groups and Lie algebras 41

2.1 Lie groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.2 Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3 Lie subgroups and subalgebras . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.4 Solvable Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.5 Nilpotent Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
2.6 Semisimple Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

3 Representions of semisimple Lie algebras 75

3.1 Basic definitions and examples . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.2 Decomposition of representations . . . . . . . . . . . . . . . . . . . . . . . . 78
3.3 Root space decompositions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.4 Root systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.5 Simple Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.6 Highest weight representations . . . . . . . . . . . . . . . . . . . . . . . . . . 98
3.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

4 Representations of compact Lie groups 112

4.1 Lie algebra of compact Lie groups . . . . . . . . . . . . . . . . . . . . . . . . 112
4.2 Maximal tori of compact Lie groups . . . . . . . . . . . . . . . . . . . . . . . 118
4.3 Roots and weights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
4.4 Weyl integration formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

1
 4.5 Weyl character formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.6 Representions of complex reductive groups . . . . . . . . . . . . . . . . . . . 141
4.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

A Appendix 151
A.1 Linear algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
A.2 Topological spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
A.3 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
A.4 Measures and Hilbert spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 165

1 Representations of compact groups

In this section, for a compact group G, we prove the Peter–Weyl theorem about the de-
composition of L2 (G) into a topological sum of irreducible finite-dimensional representations
of G × G. We start with basic definitions and examples, followed by the character theory,
the Peter–Weyl theorem for finite groups. Then we introduce the Haar measures, matrix
coefficients, and Hecke operators. Finally, we prove the Peter–Weyl theorem for compact
groups and conclude that the characters of the irreducible representations are indeed dense
in the space of central functions.

1.1 Basic definitions

Definition 1.1.1. Let G be a group.

1. By a finite-dimensional representation on G we mean a group homomorphism

ρ : G → GL(V ), dimC V < ∞.

Equivalently, we view V as a G-module with the action given by gv = ρ(g)(v). We

define the degree of ρ as deg ρ = dim V .

2. Let V and W be two representations of G. A G-homomorphism φ : V −→ W is

defined to be a linear map φ : V −→ W such φ(gx) = gφ(x). We also call such a φ a
G-equivariant map or intertwining map. Let HomG (V, W ) denote the vector space of
the G-homomorphisms of G-modules V and W

3. Let Rep(G) denote the category of representations of G with morphisms HomG (V, W )
as above with obvious composition law. A morphism φ : V −→ W is called an isomor-
phism if there is another morphism ψ : W −→ V such that

φ ◦ ψ = 1W , ψ ◦ φ = 1V .

Definition 1.1.2. Let G be a group, X a set, and Fun(X, C) the space of complex functions
on X.

2
 1. By an action of G on a set X we mean a group homomorphism

α : G −→ Aut(X),

where Aut(X) is the group of bijective self-maps on X with composition law.

2. For an action α of G on X, the associated permutation representation is defined by

α∗ : G −→ GL(Fun(X, C)) by

(α∗ (g)f )(x) = f (α(g)−1 x), g ∈ G, x ∈ X.

Example 1.1.1. Here are some basic examples.

1. Let G be a finite group of order n, and V = Fun(G, C) the vector space of complex
functions on G. Then V is a finite dimensional vector space of dimension n with a
base given by the functions δg for g ∈ G: δg (x) = δg,x . We define two representations
L and R of G on V by

L(g)(f )(x) = f (g −1 x), R(g)f (x) = f (xg).

It is easy to verified that

L(h)δg = δhg , R(h)δg = δgh−1 .

These representations are called left and right regular representations. These two
representations are isomorphic to each other by the map φ : V −→ V : f (x) 7→ f (x−1 ).

2. One can show that these two representations are commutative to each other. Thus we
get a regular representation of G × G on V :

LR : G × G −→ GL(V ), LR(g, h)f (x) = f (g −1 xh).

3. Let G = Sn be the permutation of n-letters. Then Sn acts on [n] := {1, · · · , n}. We

may identity Fun([n], C) with Cn by

f 7→ (f (1), f (2), · · · , f (n)).

Then the induces representation of Sn on Cn is given by

g(x1 , · · · , xn ) = (xg−1 x , · · · xg−1 (n) ).

4. Let G = GLn (V ) then G acts on V thus induces a representation on Fun(V, C) as well

as the subspace Symd [V ∗ ] of homogenous functions on V of degree d.

3
Example 1.1.2. When G = Z, each finite degree representation of G is isomorphic to the
unique representation given by the Jordan form. More precisely, for each positive integer n
and each A ∈ GLn (C), we define a representation ρA : Z −→ GLn (C) by ρA (g) = Ag . Then,
∼
for a finite degree representation ρ : Z −→ GL(V ), by choosing an isomorphism V −→ Cn , ρ
is isomorphic to the representation ρA with A := ρ(1) ∈ GLn (C). Moreover for A ∈ GLn (C)
∼ ∼
and B ∈ GLm (C), any isomorphism φ : ρA −→ ρB is given by an isomorphism φ : Cn −→ Cm
as C-vector spaces such that

φρA (g) = ρB (g)φ, g ∈ Z.

∼
This is equivalent to φA = Bφ. Thus we have shown that ρA −→ ρB is equivalent to A ∼ B,
or that A and B have the same Jordan form. More precisely,
Pit is given by set of pairs (ni , λi )
of positive integers ni and complex numbers λi such that ni = dim V .

Definition 1.1.3. Let G be a group.

1. Let (ρ, V ) be a represenation of G. A subrepresentation of (ρ, V ) on group G is a

subspace W ⊂ V such that for any w ∈ W and g ∈ G, ρ(g)w ∈ W . We also call such
W a G-invariant subspace.

2. Let V be a represenation of G and W a subrepresenation, then define the quotient

representation V /W with the action given by g(x + W ) = gx + W for x ∈ V .

3. Let V be a representation of G. Then define G-invariant subspace of fixed vectors by

V G := {v ∈ V | gv = v, ∀g ∈ G}.

Example 1.1.3. Let G be a finite group with regular representations L and R on the space
V of functions on G. Then V G is the space of constant functions on G.
Remark 1.1.1. For a group G, the category Rep(G) is in fact an abelian category in the sense
that for any G-homomorphism φ : V −→ W , we can define new representations:

Kerφ := {x ∈ V : φ(x) = 0}, Imφ = φ(V ), Cokerφ = W/Imφ.

ψ φ
A short exact sequence 0 −→ U −→ V −→ W −→ 0 has the same meaning as in the
category of linear spaces.

Tensor algebras of representatons

We can also talk about more operations on Rep(G):

Definition 1.1.4. Let V and W are representations of G. Then we can define the following
representations of G:

1. V ⊕ W by g · (v ⊕ w) = (g · v) ⊕ (g · w).

2. V ⊗ W by g · (v ⊗ w) = (g · v) ⊗ (g · w).

4
 3. The tensor algebra ⊗n V by g(v1 ⊗ · · · ⊗ vn ) = (gv1 ) ⊗ · · · ⊗ (gvn ).

4. The symmetric algebra and the exterior algebra S n V, n V by taking the quotient rep-
V
resentations (so g(v1 · · · vn ) = (gv1 ) · · · (gvn ) and g(v1 ∧ · · · ∧ vn ) = (gv1 ) ∧ · · · ∧ (gvn ).)

5. Hom(V, W ) by
(g · f )(v) = g · f (g −1 · v)

Example 1.1.4. The standard representation std : GLn (C) → GLn (C) is irreducible since
GLn (C) acts transitively on GLn (C). Let m be a positive integer. We can also define the
representations
Symm (std) : GLn (C) → GL(Symm (Cn )).
m
^ m
^
(std) : GLn (C) → GL( (Cn )).
Vn
So when m = n we get (std) is just the determinant map det : GLn (C) → C× .

Definition 1.1.5 (Duals and complex conjugations). Let V be a representation of G. We

want to define two representations.

1. We define the dual space of V by V ∗ = Hom(V, C). The element in V ∗ is called linear
forms. This representation is defined by

(g · f )(v) = f (g −1 · v), ∀f ∈ V ∗ .

Thus if V = Cn , and a g ∈ G is represented by a matrix A, then the g acts on V ∗ = Cn

−1
is represented by t A .

2. Let c : C −→ C be the conjugation map c(z) := z̄. For a complex space V define its
complex conjugation by
V̄ = C ⊗C,c V.
Thus we have a R-linear conjugation map c : V −→ V̄ by c(v) = v̄ := 1 ⊗ v with action
by C is given by conjugation:

av = 1 ⊗ av = ā ⊗ v = ā(1 ⊗ v) = āv̄.

The action of G on V̄ is given by

gv̄ = gv.
If V = Cn , and g is represented by a matrix A, then the action of g on V̄ = Cn is
represented by Ā.

Remark 1.1.2. Let U, V ∈ Rep(G).

1. There is a canonical G-isomorphism V ∗ ⊗ W 7→ Hom(V, W ) given by

ℓ ⊗ w 7→ (v 7→ ℓ(v)w).

5
 2. The G-module structure on Hom(V, W ) can be characterized by the followin gcommu-
tative diagram
φ
V /W .
g g
 g·φ 
V / W

3. The invariants under the action of G in Hom(V, W ) is identified with HomG (V, W ):

HomG (V, W ) = Hom(V, W )G .

Remark 1.1.3. Let V be a finite dimensional represenation of a group G. Then we have an

isomorphism of G-modules:
∼
φ : V ∗ ⊗ V −→ End(V ), φ(ℓ ⊗ v)(w) = ℓ(w)v.

We want to translate some structures of End(V ) to V ⊗ V .

1. The composition map on End(V ) is translated to the following composition V ∗ ⊗ V :

(ℓ ⊗ v) ◦ (ℓ′ ⊗ v ′ ) := ⟨ℓ, v ′ ⟩ · ℓ′ ⊗ v.

with identity element given by

X
eV = e∗i ⊗ ei ,
i

where {e1 , · · · , en } is a base of V and {e∗1 , · · · , e∗n } is its dual base of V ∗ in the sense
e∗i (ej ) = δij .

2. That the trace map on tr : End(V ) −→ C is given by the pairing

⟨·, ·⟩ : V ∗ ⊗ V −→ C.

⟨·, ·⟩ : V ∗ ⊗ V → C, (ℓ, v) 7→ ⟨ℓ, v⟩ = ℓ(v).

This pairing is G-equivariant in the sense that ⟨gℓ, gv⟩ = g⟨ℓ, v⟩ = ℓ(v).

Restriction and induced representations

Given a homomorphism ψ : G1 −→ G2 , and a representation ρ : G2 −→ GL(V ), then we
can define the restriction representation of G2 φG1 := ρ ◦ ψ : G1 −→ GL(V ). Conversely,
give a representation η : G1 −→ GL(W ), then we define the induced representation of η by
C[G2 ] ⊗C[G1 ] W . We write V |G1 when V is considered as a G1 -module. These two process
are adjoint to each other by the following identity:
∼
HomG1 (W, V |G1 ) −→ HomG2 (C[G2 ] ⊗C[G1 ] W, V ), φ 7→ (a ⊗ v 7→ aφ(v)).

6
 One particular case is when ψ is an embedding G1 = H −→ G2 = G of a subgroup in a
group G. Then we can define another induced representation by
IndG
H (W ) := {f : G −→ V, f (hg) = hf (g).}
The action of G on IndG
H (W ) is given by right translation:

(gf )(x) = f (xg).

The induced representations and restriction representations are also related to by following
adjoint properties:
∼
HomG (V, IndG
H (W )) −→ HomH (V |H , W ), f 7→ (v 7→ f (v)(e)).
If W is the trivial representation of H, then IndG
H (W ) is simply the permutation represen-
tation of G on the space X = H\G with the right action.
There is an embedding
C[G] ⊗C[H] W −→ IndG H (W )

by sending g ⊗ v to a unique fg⊗v function on G supported on the coset Hg −1 on which it

is given by f (hg −1 ) = hv. The image of the above embedding is the subspace of V -valued
functions on G with support on finitely many cosets H\G. This implies that when H has
a finite index in G, this embedding is an isomorphism. If H does not have a finite index in
G, both representations are usually infinite-dimensional. In practice, we need to put more
restrictions on functions on f depending on the structure of G. For example, we can assume
f to be smooth, holomorphic, algebraic, and of a certain degree.

Irreducible representation
Definition 1.1.6. A representation V of a group G is irreducible if it has no non-trivial
proper subrepresentation W : 0 ̸= W ̸= V .
Lemma 1.1.7. If G is commutative, then its irreducible representations are all 1 dimen-
sional.
Proof. Let V be an irreducible representation of G. For any g ∈ G acting on V , there exists
λ ∈ C such that
0 ̸= Vλ := {v ∈ V | gv = λv}.
Since G is commutative, for any h ∈ G and any v ∈ Vλ , we have
gh · v = hg · v = λhv.
Then Vλ is an G-invariant space. Thus V = Vλ . That implies that every vector is an
eigenvector for any g ∈ G. The action of G on V is given by eigenvalues:
χ : G −→ C× .
So for any v ∈ V , the set C · v is invariant under the action of G. This means V = C · v.

7
Example 1.1.5. Let G = S3 . Then we have three obvious representations: the trivial rep-
resentation triv, the sign representation sgn,Pand the “standard representation” std on the
3
subspace V of C of points (x1 , x2 , x3 ) with xi = 0. It is easy to see that all of them are
irreducible.
Theorem 1.1.6 (Schur’s lemma). Let V and W be irreducible finite dimensional represen-
tations of a G, then
• If V is not isomorphic to W then HomG (V, W ) = 0.
• If V ∼
= W , then
dimC HomG (V, W ) = 1.
In particular EndG (V ) = C.
Proof. If HomG (V, W ) ̸= 0 for any φ ∈ HomG (V, W ), we have that Im(φ) ̸= 0. Consider the
exact sequence
0 → Kerφ → V → φ(V ) → 0.
Since V is irreducible and Ker(φ) ̸= V , we must have Kerφ = 0. Since W is irreducible
and φ(V ) ,→ W is a non-zero subrepresentation, we must have φ(V ) = W . Thus φ is an
isomorphism. This proves the first part.
For the second part, we fix one isomorphism α : V −→ W . Then we get an bijection:
∼
EndG (V ) −→ HomG (V, W ) : φ 7→ α ◦ φ.
Thus it suffices to prove the lemma in the case V = W . Now let φ ∈ EndG (V ), there is a
λ ∈ C such that
0 ̸= Vλ = {v ∈ V | φ(v) = λv}.
This is a subspace of V that is invariant under action of G. Since V is irreducible we have
Vλ = V .
Corollary 1.1.8. Let V be a finite dimensional representation of a finite group G. Then V
has a decomposition
V = V1⊕n1 ⊕ V2n2 ⊕ · ⊕ Vr⊕nr
where V1 , · · · , Vr are irreducible representations of G, and ni are positive integers. Moreover,
the above decomposition is unique up to isomorphisms.

Canonical embeddings
Theorem 1.1.9 (Canonical embeddings). Let V and W be two finite-dimensional represen-
tations of a group G. Assume that W is irreducible and let V ′ be the sum of all irreducible
subrepresentations W ′ isomorphic to W . Then V ′ is the direct sum of n-copies of W with
n = dim HomG (W, V ). More precisely, the natural map
Φ : HomG (W, V ) ⊗C W −→ V
is injective with image V ′ .

8
Proof. It is clear that V ′ is the image of the above map. Thus, we can replace V with V ′
and assume that V is the sum of subrepresentations φ(W ) for all φ ∈ HomG (W, V ). Since
V is finite-dimensional, there are finitely many φi ∈ HomG (W, V ) such that
n
X
V = φi (W ).
i=1

We may choose such an expression with n minimal. We claim that in thisP case, the above
expression is a direct sum: this means that for any vi ∈ φi (W ), the equality
P vi = 0 implies
all vi = 0. Otherwise,
P there is an i such that vi ̸= 0. Then vi = − j̸=i vj . This implies
that φi (W ) ∩ ( j̸=i Pφj (W ) ̸= 0. Since W isPirreducible, we must have that φi (W ) is a
subrepresentation of j̸=i φj (W ). Thus V = j̸=i φj (W ). This contradicts the minimality
of n. Thus, we have shown V = W ⊕n . Now, by Schur’s lemma, we have
∼
HomG (W, V ) = HomG (W, W )⊕n −→ Cn .

Thus, Φ is a surjective map of two vector spaces with the same dimension, which must be
isomorphic.
Definition 1.1.10. Let V and W be two finite-dimensional representations of a group G.
Assume that W is irreducible. The multiplicity of W in V is defined to be dim HomG (W, V )
which is also the maximal number n so that W ⊕n can be embedded into V .

Complete reducibility
Definition 1.1.11. Let V be a representation of a group G. Let W, W ′ are two subrepre-
sentations of V . We say that V is the direct sum of W, W ′ if V is a direct sum of W, W ′ as
a linear space:
W + W ′ = V, W ∩ W ′ = 0.
In this case, we say that W, W ′ are complements to each other and write V = W ⊕ W ′ .
For finite group, we have the following result:
Theorem 1.1.12. Let G be a finite group and V a representation of G of finite degree. Then
every subrepresentation of G has a complement representation.
The key to proving this theorem is to construct an invariant hermitian form.
Definition 1.1.13. Let V be a finite dimensional complex space. By a hermitian form we
mean a pairing:
⟨·, ·⟩ : V × V −→ C
such that the following conditions hold:
1. the pairing is lear in the first variable:

⟨ax + by, z⟩ = a⟨x, z⟩ + b⟨y, z⟩, a, b, ∈ C, x, y, z ∈ V.

9
 2. the pairing is conjugate symmetric:

⟨x, y⟩ = ⟨y, x⟩.

3. the pairing is positive definite

⟨x, x⟩ > 0, ∀0 ̸= x ∈ V.

Lemma 1.1.14. If V be a finite dimensional representation of a finite group G. Then there

is a hermitian form ⟨·, ·⟩ on V such that

⟨gx, gx⟩ = ⟨x, y⟩, ∀g ∈ G, x, y ∈ V.

Proof. Let
⟨·, ·⟩0 : V × V → C
be a Hermitian form on V which may not be invariant under G. Since G is finite, we can
define a new Hermitian form by averaging:
1 X
⟨x, y⟩ := ⟨gx, gy⟩0
|G| g∈G

which is invariant under G.

Proof of Theorem 1.1.12. By the above Lemma, there is an invariant form ⟨·, ·⟩ on V . Let
W ⊥ be the orthogonal complement of W :

W ⊥ = {v ∈ V, ⟨v, w⟩ = 0 ∀w ∈ W } .

Then it is easy to see that W ⊥ is a complement representation of W .

Semi-simple and unitary representations

Definition 1.1.15. A representation V of G is semisimple (or completely reducible) if for
every G-invariant subspace W ⊂ V there is a G-invariant subspace W ′ ⊂ V such that
V = W′ ⊕ W.

Theorem 1.1.16 (Canonical decomposition theorem). Let V be a semisimple representation

of G. Then the following holds:

1. Any subrepresentation U of V is also semisimple.

2. There are irreducible representations W1 , · · · , Wr such that

∼
M
V −→ Wi⊕ni .
i

10
 3. The following canonical map is bijective:
∼
M
HomG (Wi , V ) ⊗C Wi −→ V.
i

Proof. For the first part, let W be a subrepresentation of U and W ′ be the complement of
U in V : V = W ⊕ W ′ . Then U = W ⊕ (U ∩ W ′ ). Thus U ∩ W ′ is a complement of W in U .
For the second part, we use induction on dim V . There is nothing to prove if V = 0.
Otherwise, V has an irreducible subrepresentation of W . Let U the complement of W in
V . Since U is semisimple by part 1 and dim U < dim V , by induction, U is the sum of all
irreducible representations of U . So is V .
For the last part, we use Theorem 1.1.9. Let Vi be the preimage of Wini in part 2. Then
by Schur’s lemma,
M
HomG (Wi , V ) = HomG (Wi , Wj )⊕nj = HomG (Wi , Wi )⊕ni = Hom(Wi , Vi ).
j

Thus the map

Φi : HomG (Wi , V ) ⊗C Wi −→ V
has the image Vi . By Theorem 1.1.9, Φi is injective. Thus their sum gives the decomposition
in part 3.
Example 1.1.7. If G = Z with a representation ρA for A a Jordan matrix with invariants
(ni , λi ). Then ρA is semisimple if and only only if A is semisimple in the sense that all ni = 1

Definition 1.1.17. Let (V, ⟨·, ·⟩) a hermtian space. The unitary group for (V, ⟨·, ·⟩) is define
associated unitary group as follows:

U(V, ⟨·, ·⟩) = {g ∈ GL(V ) : ⟨gx, gy⟩ = ⟨x, y⟩, ∀x, y ∈ V } .

∼
Example 1.1.8. If V −→ Cn with standard hermitian form
X X X
⟨ xi e i , yj ej ⟩ = xi ȳj .

Then U (V ) is the usual unitary group of matrices A such that AtĀ = In .

Definition 1.1.18. Let G be a group.

1. A unitary representation of G is on a hermitian space V is a homomorphism of Lie

group:
ρ : G → U (V, ⟨·, ·⟩).
Equivalently, a unitary representation of on a hermitian space V is an action of G on
V preserving the hermitian structure on V .

11
 2. A representation ρ : G −→ GL(V ) on a complex vector space V is unitarizable if there
is a hermitian form ⟨·, ·⟩ such that ρ factors through the unitary representon:

ρ : G −→ U (V, ⟨·, ·⟩) −→ GL(V ).

Equivalently, a unitarizable representation of on a space V is an action of G on V

preserving one hermitian structure on V .

Notice that in general, a unitarizable representation V of G may have more than one
hermitian form ⟨·, ·⟩ invariant under G.

Lemma 1.1.19. Let G be a group. Then every unitary representation is semi-simple.

Proof. If W is a subrepresentation of a unitary representation V of G, then we can define

the orthogonal complement

W ⊥ = {v ∈ V | ⟨v, w⟩ = 0, ∀w ∈ W }.

This is a G-invariant space since ⟨gv, w⟩ = g⟨v, w⟩ = 0, for any v ∈ W ⊥ . And we have
V = W ⊕ W ⊥.
Example 1.1.9. 1. If G is finite, then all the representations of G are unitarizable, thus
semi-simple.

2. If G = Z with a representation ρA for A a Jordan matrix with invariants (ni , λi ). Then

ρA is unitarizable if and only if all ni = 1 and all |λi | = 1.

1.2 Characters
Definition 1.2.1. Let V be a finite dimensional vector space over a field k, and let ρ : G →
GL(V ) be a representation of G. The character of ρ is the function

χρ : G → k, g 7→ tr(ρ(g)),

where tr is the trace of ρ(g) as a linear transform on V . We also use χV to denote χρ if we

consider V as a G-module.

There are some fundamental properties of characters:

Lemma 1.2.2. Let ρ : G → GL(V ) be a representation on G.

1. χV (hgh−1 ) = χV (g), for any g, h ∈ G;

2. χV (1) = tr(I) = dim V ;

3. χV = ρ if ρ is 1-dimensional.
∼
4. If representations V −→ W , then they have the same character.

12
Lemma 1.2.3. Let V, W be representations of G. Then

1. For any short exact sequence E : 0 → V → U → W → 0 in Rep(G),

χU = χV + χW .

2. χV ⊗W = χV χW .

3. χV ∗ (g) = χV (g −1 ) for all g ∈ G.

4. χHomG (V,W ) (g) = χV (g −1 )χW (g) (deduced from 2 and 3)

Proof. 1. Let (ei )1≤i≤m (resp. (fj )1≤j≤n ) be a base for V (resp. W ) and write
X X
gei = aik ek , gfj = bjℓ fℓ .
k ℓ

P P
Thus tr(g|V ) = aii , tr(g|W ) = j bjj . Lift fj ’s to some vectors fej ∈ U . Then (ei , fej )
is a base for U . For each g ∈ G, we want to find the matrix to represent g under the
above base: X X X
gei = aik ek , g fej = bjℓ feℓ + cjm em .
k ℓ m

Thus X X
tr(g|U ) = aii + bjj = tr(g|V ) + tr(g|W ).
i j
P P
2. Suppose V = Cei , W = Cfj . Then we can write
X X
ρV (g)(ei ) = aik ek , ρW (g)(fj ) = bjℓ fℓ .

So X
(ρV ⊗ ρW )(g)(ei ⊗ fj ) = aik bjℓ ek ⊗ fℓ .
k,ℓ

We get
X X X
tr((ρV ⊗ ρW )(g)) = aii bjj = ( aii )( bjj ) = tr(ρV (g)) × tr(ρW (g)).
i,j i j

Cei , then we can write V ∗ =

P P ∗
3. Suppose V = Cei . If
X X
g · ei = aik ek , g · e∗j = bjℓ e∗ℓ ,
k ℓ

we have
⟨gei , ge∗j ⟩ = ⟨ei , e∗j ⟩ = e∗j (ei ) = δij .

13
 Write X X X
δij = ⟨ aij ej , bjℓ e∗ℓ ⟩ = aik bjk .
j ℓ

If the matrix ρV (g) = A = (aij ) and ρV ∗ (g) = B = (bij ), we have AT B = I, so

tr(B) = tr((AT )−1 ) = tr(A−1 ).

Example 1.2.1. Let G = Sn . Then each element has a permutation decomposition and is
conjugate to

σn1 ,n2 ,··· ,nr := (1, 2, · · · , n1 )(n1 + 1, n1 + 2, · · · n1 + n2 ) · · · (n1 + · · · + nr−1 + 1, · · · , n)

P
with n1 ≥ n2 ≥ · · · ≥ nr such that ni = n. Thus for any representation ρ of Sn , χρ is
determined by its values at σn1 ,n2 ,··· ,nr . For example for G = S3 , we have three representations
triv, sgn, and std. Here is the character table

triv sgn std

1 1 1 2
(12) 1 −1 0
(123) 1 1 −1
Remark 1.2.1. We may interpret the above theorem in Grothendieck ring K 0 (G) as follows.
We define the Grothendieck K-group of Rep(G). Let V ∈ Rep(G), denote the isomorphic
classes of V by [V ]. Let F (G) be the free abelian group generated by [V ]. For any short
exact sequence
E : 0 → V1 → V → V2 → 0
in Rep(G), we associate an element of

[E] = [V ] − [V1 ] − [V2 ].

Let E(G) denote the subgroup of F (G) generated by [E] for all short exact sequences in
Rep(G). Then we define the Grothendieck K-group of Rep(G) by

K 0 (G) := F (G)/E(G).

So K 0 (G) is a ring on Z because we define the tensor product and direct sum.
Denote the set of central functions on G by Z(G, C). So actually the morphism

χ : K 0 (G) → Z(G, C), [V ] 7→ χV

∼
After we studied the Peter-Weyl’s theorem, we will prove that K0 (G) ⊗Z C −→ Z(G, C).

14
Characters of representations of finite groups
We shall assume that G is finite if not specified in the following. We will show that Rep(G)
has a one-to-one correspondence to the set of characters of representations on G. Let C(G, C)
denote Fun(G, C) with a hermitian product:
1 X
⟨f, g⟩ = f (x) · g(x).
|G| x∈G

Theorem 1.2.2 (Main theorem). Let V, W be representations of a finite group G. Then We

have Schur orthogonality
1 X
⟨χV , χW ⟩ = χV (g)χW (g) = dimC HomG (W, V ).
|G| g∈G

Proof of 1.2.2. Step 1: Reduce to the case where W = C is trivial. Since G is finite, we can
put a hermitian form on W which is invariant under G. By Definition 1.1.17, we have an
isomorphism
∼
W ∗ −→ W̄ .
Thus we have
χV χW = χV χW̄ = χV ⊗W̄ = χV ⊗W ∗
and
HomG (W, V ) = (W ∗ ⊗ V )G = HomG (C, W ∗ ⊗ V ).
Thus the identity in the theorem for the pair (V, W ) is equivalent to the identity for the case
(W ∗ ⊗ V, C).
Step 2: Reduce to the case W = C is trivial, and V is irreducible. Since G is finite, then
all the representations are semi-simple. So assume that
X
V = ni Vi = ⊕i Vi⊕ni , ni ∈ N

Where Vi are irreducible and [Vi ] ̸= [Vj ] for i ̸= j. Then we have

X
χV = ni χVi .

Step 3: Now assume W trivially, and V is irreducible. The theorem is clear if V is trivial.
Now suppose that V is not not trivial. Let (ei )1≤i≤d be an orthogonal base for Vi under some
hermitian form. Then
1 X 1 XX 1 X
χV (g) = ⟨gei , ei ⟩ = ⟨ gei , ei ⟩.
|G| g∈G |G| g∈G i |G| g∈G

1
gei is invariant under G. Since V is non-trivial and irreducible, V G =
P
Notice that |G| g∈G
0.

15
Corollary 1.2.4. Let V, W be irreducible finite dimensional representations of finite group
G. Then (
1 X 0 V ̸= W
⟨χV , χW ⟩ = χV (g)χW (g) =
|G| g∈G 1 V =W

Combining with Theorem 1.1.16, we have:

Corollary 1.2.5. Two representations of a finite group G are isomorphic to each other if
and only if they have the same character.
Corollary 1.2.6. A representation V of a finite group is irreducible if and only if ⟨χV , χV ⟩ =
1.
Example 1.2.3. Let G = Z/N Z for some N ∈ N so the irraducible representations are
ρn : Z/nZ −→ C× , x 7→ exp(2πinx/N ).
So χρn (x) = exp(2πinx/N ), and
(
1 X 0 m ̸= n
exp(2πinx/N )exp(2πimx/N ) = .
N 1 m=n
g∈Z/N Z
P P
If ρ is any representation, then ρ = ai ρni , χρ (x) = ai exp(2πinx/N ).
Let G is a finite group of order n. Consider the right regular representation on C(G, C) :=
Fun(G, C). By Exercise 1.7.6, (
n g = e,
χR (g) =
0 g ̸= e.
Consequently, we have for any irreducible representation V ,
⟨χR , χV ⟩ = dim V.
This shows the following
Theorem 1.2.7 (Peter–Weyl theorem for right regular representions). Let G be a finite group
of order n and {[V1 ], · · · , [Vr ]} be the set of all equivalent classes of irreducible representations
with dimension ni . Then
∼
X
C(G, C) −→ ni Vi .
i
Consequently, we have X
n= n2i .
Example 1.2.4. Let G = S3 . Then we have listed three representations triv, sgn, and std of
dimensions 1, 1, 2. It is clear that
|S3 | = 6 = 12 + 12 + 22 .
So triv, sgn, and std are all irreducible representations of S3 .

16
1.3 The Peter–Weyl theorem for finite groups
Let G be a finite group and C(G, C) the space of functions on G. Then we have some regular
representations of G on GL(C(G, C)):

L : G −→ GL(C(G, C)), L(g)f (x) := f (g −1 x),

R : G −→ GL(C(G, C)), R(g)f (x) := f (xg),

LR : G × G −→ GL(C(G, C)), LR(g1 , g2 )f (x) = f (g1−1 xg2 ).
For any V ∈ Rep(G), we have a representation V ∗ ⊠ V ∈ Rep(G × G) as the linear space
V1 ⊗ V2 as with G1 × G2 given by

(g1 , g2 )(ℓ ⊗ v) = g1 ℓ ⊗ g2 v.

Then define a matrix coefficient map AV : V ∗ ⊠ V −→ C(G, C) by

AV (ℓ ⊗ v)(g) := ⟨ℓ, gv⟩, ∀ℓ ⊗ v ∈ V ∗ ⊠ V.

It is clear that this is a homomorphism of G × G-equivariant map for the representation LR

on C(G, C).
Example 1.3.1. Let G be a group and V ∈ Rep(G) a representation. Take a basis {e1 , · · · , en }
for V and a bssis {e∗1 , · · · , e∗n } for the dual V ∗ . Let eV := i e∗i ⊗ ei ∈ V ∗ ⊠ V . Then
P

X
AV (eV )(g) = ⟨e∗i , gei ⟩ = χV (g).
i

Theorem 1.3.1 (Peter-Weyl Theorem for finite groups). Let G is a finite group and {[V1 ], · · · , [Vr ]}
be the set of all equivalent classes of irreducible representations with dimension. Then the
sum of the matrix coefficient maps for all Vi induce an isomorphism
r
∼
M M
A := AVi : Vi∗ ⊠ Vi −→ C(G, C).
i=1

Proof. By Theorem 1.2.7, two sides have the same dimension. Thus it suffices to prove that
the map A is injective. This follows from Corollary 1.2.4 and 1.2.6 that each Vi ⊠ Vi∗ is
irreducible and non-isomorphic to each other as representations of G × G. More precisely,
the character for Vi∗ ⊠ Vi is χ̄Vi (g1 )χVi (g2 ). It is easy to see that this function has the norm
1 on G × G, and orthogonal to the character of different Vj∗ ⊠ Vj for j ̸= i.
Let Z(G, C) denote the space of central functions of G:

Z(G, C) = f ∈ C(G, C) : f (gxg −1 ) = f (x) :


x, g ∈ G .

17
Theorem 1.3.2. Let G is a finite group and {[V1 ], · · · , [Vr ]} be the set of all equivalent
classes of irreducible representations. Then χVi form an orthonormal base for Z(G, C).
More precisely, the map A induces a bijection:
∼
M M
A: CeVi −→ CχVi .
i i

Proof. Let ∆(G) ⊂ G × G be the diagonal subgroup of elements (g, g). Thenthe map A in
Theorem 1.3.1 induces a bijection of ∆(G) invariants:
r
∼
M M
A := AVi : (Vi∗ ⊠ Vi )∆G −→ C(G, C)∆G .
i=1

It is easy to see that the right-hand side is exactly Z(G, C). Each term on the left-hand side
is given by (V ∗ ⊗ V )G = EndG (V ). By Schur’s lemma, each term on the left hand side is
isomorphic to C. The map AVi sends eVi to χVi . Now the theorem follows from Corollary
1.2.4.
Notice that the space of central functions can be identified with the space of functions
on the conjugacy classes of G. Thus we have
Corollary 1.3.3. The number of irreducible representations of G is equal to the number of
conjugacy classes of G.
Example 1.3.2. Let G = S4 . Then S4 has 5 conjugacy classes represented by
1, (12), (123), (1234), (12)(34).
So it has five irreducible representations. We can list four easy representations: triv, sgn, std, sgn⊗
std where std is the subspace C4 with sum 0. From the character table, they are all irre-
ducible. So there is a missing one called W whose dimension can be calculated by the
formula:
24 = |G| = 12 + 12 + 32 + 32 + dim W 2 .
Thus dim W = 2. To understanding this representation, let X be the conjugacy class of
(12)(34):
X = {(12)(34), (13)(24), (14)(23)}
with an action of G by conjugation. Then G acts on Fun(X, C). Let W be the subspace of
Fun(X, C) with sum of values 0. Then W is the missing 2-dimensional representations of G.
The following is the complete character table:

triv sgn std sgn ⊗ std W

1 1 1 3 3 2
(12) 1 −1 1 −1 0
(123) 1 1 0 0 −1
(1234) 1 −1 −1 1 0
(12)(34) 1 1 −1 −1 2

18
1.4 Compact groups
Topological groups
Definition 1.4.1. A topological group is a topological space G equipped with a group structure
such that the multiplication and the inversion maps

m : G × G → G, (g, h) 7→ gh
i : G → G, g 7→ g −1
are both continuous.
Definition 1.4.2. Let G, H be two topological groups. A morphism (resp. homomorphism)
from G to H is a morphism f : G → H which is both continuous and group morphism(resp.
homomorphism).
Example 1.4.1 (Matrix groups). By a topological ring, we mean a ring R with a topological
structure such that operations +, × are continuous. For example:

R, C, Qp , Zp , H,

where H is the ring of Hamiltonians.

1. For a topological ring R , we have a topological ring Mn (R) of n × n matrices and
general linear group GLn (R) and linear subgroups defined by equations:

(a) when R is commutative, the special subgroup SLn (R) of elements with determi-
nant equal to 1;
(b) the Borel subgroup Bn (R) of upper triangular matrices;
(c) the unipotent rational of Bn (R) consisting of elements with 1 at the diagonal.
(d) the orthogonal subgroup On (R) defined by equation A · t A = 1, and its special
subgroup SOn (R) = On (R) ∩ SLn (R); we denote this as O(n) if R = R.
t
(e) the symplectic  Spn (R) of GL2n (R) defined by equation A · J · A = J
 subgroup
0 1n
where J = .
−1n 0

2. for R = C or H, we have the unitary group U (n) and UH (n) by the equation A · t Ā = 1.
In these examples, O(n), SO(n), U(n), SU(n), UH (n) are compact groups which we will study
closely in this course.
Example 1.4.2 (Projective limits). Let Gn be a sequence of topological groups with surjective
continuous maps Gn → Gn−1 . Then we can form a projective limit G = lim Gn with topology
←−
generated by preimages of opens in Gn ’s via surjections G → Gn . If all Gn compact, then G
is also compact. This is the case when all Gn is finite. In this case we say that G is profinite.
The following are some examples of compact groups coming from number theory:

19
 b := lim Z/nZ = Q Zp ;
1. Z
←−n p

b = lim GLn (Z/n) = Q GLn (Zp );

2. GLn (Z)
←−n p

3. Gal(Q̄/Q) = limK Gal(K/Q) where K runs through the set of finite Galois extensions
←−
of K of Q;
4. lim R/nZ = (R × Z)/Z
b where Z is diagonally embedded into R and Z.
b
←−
Example 1.4.3. We have the following relations of compact groups:
∼ ∼
1. S 1 = R/Z −→ U(1) −→ SO(2) given by
 
2πit cos 2πt sin 2πt
t 7→ e 7→ .
− sin 2πt cos 2πt
The second map is induced by canonical identification R2 = C as real vector spaces.
∼
2. S 3 = UH (1) −→ SU(2) given by
 
a + bi c + di
a + bi + cj + dk 7→ .
−c + di a − bi
∼
This map is induced by identification C2 −→ H as (right) complex vector spaces by
sending (x, y) to x + jy.
∼
3. UH (1)/{±1} −→ SO(3) given by identification that the trace free part H0 of H is
identified with R3 :
Ri + Rj + Rk −→ R3 .
Recall that UH (1) is the subgroup of H× with norm 1. Thus UH (1) acts on H0 by
conjugation:
g(x) = gxḡ, g ∈ UH (1), x ∈ H0 .
This action preserves the norm on R3 , and thus induces a map UH (1) −→ O(3) with
the kernel {±1}. Since UH (1) is connected, this induces an embedding H× /{±1} −→
SO(3). These two groups have the same dimension 3. Thus this is bijective.
∼ ∼
4. (UH (1) × UH (1))/{±(1, 1)} −→ SO(4) by the canonical identification H −→ R4 . We
take an action of UH (1) × UH (1) on H by
(g1 , g2 )(x) = g1 xḡ2 , ∀(g1 , g2 ) ∈ UH (1) × UH (1), ∀x ∈ H.
It is clear that this action keep the norm on H this induces a homomorphism
UH (1) × UH (1) −→ O(4).
Since the image is connected, and the kernel is ±(1, 1), we have an embedding
(UH (1) × UH (1))/{±(1, 1)} ,→ SO(4).
By counting dimensions, we see that both sides are equal to 6. Thus this bijective.

20
Representations of topological groups
Definition 1.4.3. A Representation of a topological group G on a finite dimensional vector
space V over C is topological homomorphism
∼
ρ : G −→ GL(V ) −→ GLd (C).

This can also be interpreted as G-module structure on V :

G × V −→ V : (g, v) 7→ ρ(g)v.

We will also consider infinite-dimensional representations. Here are some examples:

Definition 1.4.4. Let G be a topological group, X a topological set, and C(X, C) the space
of complex valued continuous functions on X.

1. By an action of G on a set X we mean a continuous map

α : G × X −→ X

such that α(e, x) = x and α(g1 g2 , x) = α(g1 , α(g2 , x)).

2. For an action α of G on X, the associated permutation representation is defined by

α∗ : G −→ GL(C(X, C)) by

(α∗ (g)f )(x) = f (α(g)−1 x), g ∈ G, x ∈ X.

Example 1.4.4. Here are some basis examples.

1. The regular actions L, R, LR of G on C(G, C) which are all infinite dimensional.

2. Let G = GLn (C) and X = Cn . For any m ∈ N, let Vm (Cn ) denote the space of
homogenous functions on Cn of degree m. Then Vm (Cn ) is a representation of G. This
also induces representations for any closed subgroup H of G.

3. From given representations, one can always construct new representations by direct
sums, duals, tensors, wedge products, etc.

Example 1.4.5. When G = R, each finite degree represenation of G is isomorphic to the

unique representation given by Jordan form. More precisely, for each positive integer n and
each A ∈ Mn (C), we define a representation ρA : R −→ GLn (C) by
∞
X
ρA (x) = exA = xm Am /m!.
m=0

With some extra work in calculus, we can show that any a finite degree representation is
isomorphic to some representation ρA . Moreover for A ∈ GLn (C) and B ∈ GLm (C), any

21
 ∼ ∼
isomorphism φ : ρA −→ ρB is given by an isomorphism φ : Cn −→ Cm as C-vector spaces
such that
φρA (g) = ρB (g)φ, g ∈ R.
∼
This is equivalent to φA = Bφ. Thus we have shown that ρA −→ ρB is equivalent to A ∼ B,
or that A and B have the same Jordan form. More precisely, Pit is given by set of pairs (ni , λi )
of positive integers ni and complex numbers λi such that ni = dim V . See Exercise 2.7.8
for characterization for Homcont (R, G) for a general Lie group G.
Example 1.4.6. If G = U(1) ∼ = S1 ∼
= SO(2) ∼ = R/Z, they are all one dimensional commutative
compact topological groups. We have the isomorphism
Z → Ĝ, n 7→ (ρn : z 7→ z n ).
Since every ρn is 1-dimensional hence irreducible and ρn ̸= ρm for n ̸= m. We have the map
is injective.
Let ρ : G → GL(V ) be irreducible. By 1.1.7 dim V = 1. So we can write ρ : U (1) → C× .
We need to prove that ρ = ρn for some n. By universal lifting theorem, ρ lifts to a universal
]
covering map ρe : U f× so that the following diagram is commutative:
(1) −→ C
ρe
]
U (1) / f×
C
πU (1) πC×
 ρ

U (1) / C× .
∼
Recall that U (1) has a universal cover πU (1) : R −→ U (1) : x 7→ e2πix and C∗ has a universal
∼ f× −→ C∗ given by z 7→ e2πiz . Thus ρe is a homomorphism ρe : R −→ C
cover πC× : C −→ C
This must be given by ρe(x) = ax for some a ∈ C. The above commutativity says that for
ρ(e2πix ) = e2πax .
Taking x ∈ Z, this shows that e2πiax = ρ(1) = 1. This shows that a ∈ Z. So ρ is given by
z 7→ z a for a ∈ Z.
Example 1.4.7. Let G = SO(n) which acts on Rn and thus on the spaces Vm (Rn ) homogenous
polynomial functions of degree m. This space is irreducible when m = 1 but not for m = 2
as it contains a fixed function
|x|2 = x21 + x22 + x23 + · · · + x2n .
In fact for any m ≥ 2, there is an embedding Vm−2 (Rn ) into Vm (Rn ) by
f (x) 7→ |x|2 f (x).
We let Hm (Rn ) denote the quotient of these representations. We will show that Hm (Rn )
together form a set of all irreducible representations.
In fact, Hm can be realized as subrepresentations of Vm of harmonic polynomials:
Hm (Rn ) = {f ∈ Vm (Rn ), ∆f = 0}
where ∆ = ∂x21 + ∂x22 + · · · ∂x2n is the standard Laplacian operator.

22
Haar measures
For a locally compact space X, by a measure µ (or dµ) on X we mean a positive distribution.
Thus for any compactly supported continuous function f ∈ Cc (X, R), we can define an
integral: Z
µ(f ) = f (x)dµ(x).
X

Definition 1.4.5 (Haar Measure). Let G be a topological group. A left invariant (or Haar)
measure on G is a nonzero measure µ such that for any f ∈ Cc (G),
Z Z  Z Z 
f (gx)dµ(x) = f (x), resp. f (xg)dµ(x) = f (x)dµ(x). .
G G G G

Example 1.4.8. The Lebesgue measure on Rn is a left and right Haar measure.
Example 1.4.9. If G is discrete, then the counting measure dx is a left and right Haar measure:
Z X
f (x)dµ := f (x).
G x∈G

∼ 2
Example 1.4.10. If G = GLn , and let dL g = dL gij be the Lebesgue measure on Mn −→ Rn
Q
which gives the usual integral:
Z
2
f (g)dL (g), f ∈ Cc (Rn )

Then dg = | det g|−n dL g is a left and right Haar measure on G. In other words, for any
f ∈ Cc (G), and h ∈ G,
Z Z Z
f (g)dg = f (hg)dg = f (gh)dg.
G G G
n n n
This can be seen by view M n = R × R × · · · × R via their column vectors g = (g1 , · · · , gn )
for gi ∈ Rn . Then dL g = dgi where dgi is the Lebesgue measure on Rn . For any h ∈ G,
Q
we have hg = (hg1 , · · · , hgn ). The h as a linear transform on Rn will change measure by a
factor | det g|: d(hgi ) = | det h|dgi . It follows that

dL (hg) = | det h|n dL g.

It follows that for the measure dg = | det g|−n dL g, we have

d(hg) = | det hg|−n dL (hg) = | det g|−n dL (g) = d(g).

This shows that dg is left-invariant. Similarly, we can show that dg is the right invariant.
Example 1.4.11. The Haar measure µ on the circle group U (1) can be defined by paratriza-
tion: for any f ∈ C(S 1 ), Z Z 1
f (g)dg = f (e2πit )dt.
S1 0

23
Proposition 1.4.6. Let µ be a left Haar measure on G
1. Let c ∈ R>0 , cµ is a left Haar measure.

2. The µ(g −1 ) defined as follows is a right Haar measure:

Z Z
−1
f (g)dµ(g ) := f (g −1 )dµ(g)
G G

Theorem 1.4.12. Every locally compact topological group G has a left Haar measure. If
µ1 and µ2 are two left Haar measures on G, then there exists a unique c ∈ R>0 such that
µ2 = cµ1 .
Example 1.4.13.
S ∗Let G = ← limn Gn is a profinite group with surjections πn : G −→ Gn . Then
−
Cc (G, R) = n πn C(Gn , R) where πn∗ is the pull-back morphism on functions:

πn∗ (f )(x) = f (πn x), ∀x ∈ G.

Then we define a Haar measure on G by

Z
1 X
φ(x)dx = f (x)
G |Gn | x∈G
n

if φ = πn∗ f . It is easy to see that this definition does not depend on the choice of n and f .
Remark 1.4.1. Let G be a matrix group or, more generally, a Lie group, then the left Haar
measure can be constructed through left-invariant differential forms. More precisely, for any
given form ωe ∈ ΩnG,e at the unit element which gives the orientation of G, then we define
its translation to any point g ∈ G by

ωg = ℓ∗g−1 ωe ∈ ΩG,g ,

where ℓg is the map G −→ G : x 7→ gx. All

R ωg together form a left-invariant form ω on X.
This form defines a left Haar measure by G f ω.

Modular function
Let µ be a left Haar measure on G, since for g ∈ G, then for any g ∈ G,
Z
f 7→ f (xg)dµ
G

is still a left Haar measure, so there exists a unique c(g) ∈ R>0 such that
Z Z
f (xg)dµ = c(g) f (xg)dµ.
G G

The function c : G → R>0 is called the modular function of G.

24
Definition 1.4.7. G is unimodular if for any g ∈ G, c(g) ≡ 1. Then any left Haar measure
is right haar measure.

Example 1.4.14. A commutative group is unimodular.

Theorem 1.4.8. Any compact topological group is unimodular.

Proof. Let G be a compact group and dg a left-invariant measure. Then g 7→ c(g) is a

continuous homomorphism c : G −→ R>0 . Since G is compact, c(G) is a compact subgroup
of R>0 . However R>0 does not have non-trivial compact subgroup. There c(g) ≡ 1. This
shows that dg is also right invariant.

Lemma 1.4.9. If G is compact, then all representations of G are unitarizable, thus semi-
simple.

Proof. Let ρ : G → GL(V ) be a representation. Let

⟨·, ·⟩0 : V × V → C

be a Hermitian form on V . Note V may not be unitary under this Hermitian form(i.e. there
is a g ∈ G such that ⟨gx, gy⟩0 ̸= ⟨x, y⟩0 , for x, y ∈ V .) Since G is compact. We can define a
new Hermitian form by integration using Haar measure
Z
⟨x, y⟩ := ⟨gx, gy⟩0 dg
G

which is invariant under G.

Representation Hm (Rn )
In the following, we will use Haar measure to study the representation Hm (V n ) of SO(n)
constructed in Example 1.4.7. Let’s embed SO(n − 1) into SO(n) as the subgroup stablising
the point e1 = (1, 0, 0, · · · ).

Theorem 1.4.10. Assume that n ≥ 3. Then every representation Hm (Rn ) of SO(n) is

irreducible. Moreover m
M
Hm (Rn )|SO(n−1) = Hi (Rn−1 ).
i=0

Proof. Since all representations of SO(n) are semisimple, we may safely consider Hm (Rn ) as
a subrepresentation complement to Vm−2 (Rn ) and write:

(1.4.1) Hm (Rn ) = Vm (Rn ) − Vm−2 (Rn ).

The irreducibility comes from the following two estimates of SO(n − 1) invariants:

1. dim Hm (Rn )SO(n−1) = 1;

25
 2. for any nonzero subrepresentation V of Vm (Rn ), dimSO(n−1) ≥ 1.

In this way, Hm (Rn ) can’t be written as the direct sum of two nonzero subrepresentations.
Thus must be irreducible.
For the first estimate, notice that every polynomial P in Vm (Rn ) can be written as
X
P (x1 , x2 , · · · , xn ) = xi1 Pi (x2 , x3 , · · · , xn ), Pi ∈ Vm−i (Rn−1 )

Thus we have m
M
n
Vm (R )|SO(n−1) = Vi (Rn−1 ).
i=0

Bring this to above expression 1.4.1 gives:

(1.4.2) Hm (Rn )|SO(n−1) = Vm (Rn−1 ) ⊕ Vm−1 (Rn−1 ).

Thus
dim Hm (Rn )SO(n−1) = dim Vm (Rn−1 )SO(n−1) + Vm−1 (Rn−1 )SO(n−1) .
It is easy that since n − 1 ≥ 2, dim Vm (Rn−1 )SO(n−1) is zero unless m is even in which case,
it is generated by |x|m . This proves the first estimate.
For the second estimate, we let f ∈ V be a nonzero function. So pick p ∈ Rn such that
f (p) ̸= 0. Since f is homogeneus, we may assume that |p| = 1. Then there is a h ∈ SO(n)
such that p = he. Now we define a new element in V SO(n−1) using Haar measure on SO(n−1):
Z
f (x) :=
e f (hgx)dg.
SO(n−1)

From definition, it is easy to see that fe(e) = f (p) ̸= 0. Thus dim V SO(n−1) ̸= 0. This proves
the first part of the Theorem.
For the second of Theorem 1.4.10, we use quality 1.4.2 and the fact that

Vm (Rn−1 ) = Hm (Rn−1 ) ⊕ Hm−2 (Rn−1 ) ⊕ · · · .

1.5 Characters for compact groups

In the following, we shall assume that G is compact if not specified, and dg is the Haar
measure on G such that vol(G) = 1.

Theorem 1.5.1 (Main theorem). Let V, W be representations of compact topological group

G. Then Z
⟨χV , χW ⟩ = χV (g)χW (g)dg = dimC HomG (W, V ).
G

26
Proof of 1.5.1. Step 1: Reduce to the case where W = C is trivial. Since G is compact, we
can put a hermitian form on W which is invariant under G. By Definition 1.1.17, we have
an isomorphism
∼
W ∗ −→ W̄ .
Thus we have
χV χW = χV χW̄ = χV ⊗W̄ = χV ⊗W ∗
and
HomG (W, V ) = (W ∗ ⊗ V )G = HomG (C, W ∗ ⊗ V ).
Thus the identity in the theorem for the pair (V, W ) is equivalent to the identity for the case
(W ∗ ⊗ V, C).
Step 2: Reduce to the case W = C is trivial, and V is irreducible. Since G is compact,
then all the representations are semi-simple. So assume that
X
V = ni Vi = ⊕i Vi⊕ni , ni ∈ N

Where Vi are irreducible and [Vi ] ̸= [Vj ] for i ̸= j. Then we have

X
χV = ni χVi .

Step 3: Now assume that W is trivial and V is irreducible. The theorem is straightforward
if V is trivial. Now suppose that V is not not trivial. Let (ei )1≤i≤d be an orthogonal base
for Vi under some hermitian form. Then
Z Z X XZ
χV (g)dg = ⟨gei , ei ⟩dg = ⟨ gei , ei ⟩.
G G i i G

R
Notice that G
gei is invariant under G. Since V is non-trivial and irreducible, V G = 0.

Corollary 1.5.1. Let V, W be irreducible finite dimensional representations of compact topo-

logical group G. Then (
0 V ̸= W
Z
χV (g)χW (g)dg =
G 1 V =W

Combining with Theorem 1.1.16, we have that the representation of a compact group is
uniquely determined by its character:

Corollary 1.5.2. Two representations of a compact group are isomorphic to each other if
and only if they have the same character.

Corollary 1.5.3. Let V be a representation of a compact group. Then V is irreducible if

and only if ⟨χV , χV ⟩ = 1.

27
Example 1.5.2. Let G = R/Z so the irraducible representations are

ρn : R/Z −→ C× , x 7→ exp(2πinx).

So χρn (x) = exp(2πinx). And

(
0 m ̸= n
Z
exp(2πinx)exp(2πimx)dx = .
R/Z 1 m=n
P P
If ρ is any representation, then ρ = ai ρni , χρ (x) = ai exp(2πinx). This is the Fourier
polynomial.
We will show that {χV , V ∈ Rep(G) irreducible} is an orthogonal basis for central func-
tions of G.

Matrix coefficients
Let G be a topological group and C(G, C) the space of continuous functions on G. Then we
have some regular representations of G on GL(C(G, C)):

L : G −→ GL(C(G, C)), L(g)f (x) := f (g −1 x),

R : G −→ GL(C(G, C)), R(g)f (x) := f (xg),

LR : G × G −→ GL(C(G, C)), LG(g1 , g2 )f (x) = f (g1−1 xg2 ).
For any V ∈ Rep(G), we have a representation V ∗ ⊠ V ∈ Rep(G × G) as the linear space
V1 ⊗ V2 as with G1 × G2 given by

(g1 , g2 )(v ⊗ ℓ) = g1 v ⊗ g2 ℓ.

Then define a matrix coefficient map AV : V ∗ ⊠ V −→ C(G, C) by

AV (ℓ ⊗ v)(g) := ⟨ℓ, gv⟩, ∀v ⊗ ℓ.

It is clear that this is a homomorphism of G × G-equivariant map for the representation LR

on C(G, C).
Example 1.5.3. Let G be a group and V ∈ Rep(G) a representation. Take a basis {e1 , · · · , en }
∗
for V and a bssis {e1 , · · · , en } for the dual V . Let eV := i ei ⊗ e∗i ∈ V ∗ ⊠ V . Then
∗ ∗
P

X
AV (eV )(g) = ⟨e∗i , gei ⟩ = χV (g)
i

If V is irreducible, then AV is isomorphic to its image ImAV which is generated by χV̄ . In

other words, it is generated over C by functions LR(g1 , g2 )χV for all g1 , g2 ∈ G.

28
1.6 The Peter–Weyl theorem for compact groups
Hecke algebras
Let G be a compact group with a Haar measure dx of total volume 1. Let C(G, C) be
the space of continuous function on G. For any two functions φ1 , φ2 on G, we define their
convolution by
Z Z
−1
(φ1 ∗ φ2 )(x) = φ1 (g)φ2 (g x)dg = φ1 (xg −1 )φ2 (g)dg.
G G

It is easy to prove that this involution satisfies the composition law:

Lemma 1.6.1.

(φ1 ∗ φ2 ) ∗ φ3 = φ1 ∗ (φ2 ∗ φ3 ), φ1 , φ2 , φ3 ∈ C(G, C).

Proof. Let x ∈ G. The left hand side is

Z
(φ1 ∗ φ2 ) ∗ φ3 (x) = (φ1 ∗ φ2 )(xg −1 )φ3 (g)dg
ZG Z
= φ1 (h)φ2 (h−1 xg −1 )φ3 (g)dhdg
Z Z
= φ1 (h)(φ2 ∗ φ3 )(h−1 x)dh

=φ1 ∗ (φ2 ∗ φ3 )(x)

Thus C(G, C) together with involution ∗ forms a new ring structure for C(G, C) which
we called the Hecke algebra of continuous functions on G. Notice that C(G, C) does not
have a unit element unless G is finite. Let Z(G, C) be the subspace of central functions:

Z(G, C) = φ ∈ C(G, C) : φ(gxg −1 ) = φ(x),


g, x ∈ G .

Lemma 1.6.2. The Z(G, C) is the center of C(G, C) under convolution:

Z(G, C) = {φ ∈ C(G, C) : φ ∗ f = f ∗ φ, f ∈ C(G, C)} .

Proof. Let f ∈ C(G, C), φ ∈ Z(G, C). So φ(g −1 x) = φ(xg −1 ), for any x, g ∈ G.
Z Z
−1
f ∗ φ(x) = f (g)φ(g x)dg = f (g)φ(xg −1 )dg = φ ∗ f (x).
G G

Thus f are in the center of (C(G, C), ∗).

Conversely suppose that φ is in the center of (C(G, C), ∗), so for any f ∈ G, we have
Z Z
−1
f (g)φ(g x)dg = (f ∗ φ)(x) = (φ ∗ f )(x) = φ′ (g)φ(xg −1 )dg.
G G

29
In other words, Z
f (g)(φ(xh−1 ) − φ(h−1 x))dg = 0.
G

For a fixed x, take f = φ(xh−1 ) − φ(h−1 x), then we get

Z
|φ(xh−1 ) − φ(h−1 x)|2 dg = 0.
G

So φ(xg −1 ) − φ(g −1 x) = 0, for any x, g ∈ G.

Remark 1.6.1. A more natural construction of Hecke algebra is to use the space of distribu-
tions D(G, R). In this we define the product µ1 , µ2 of two distribution using ”pushforward”
of multiplications:
m : G × G −→ G : µ1 ∗ µ2 = m∗ (µ1 × µ2 ).
More precisely, for a function f ∈ C(G),
Z Z
f (x)d(µ1 ∗ µ2 ) := f (xy)dµ1 (x)dµ2 (y).
G G×G

This large Hecke algebra has a unit element δe . Recall δ : G −→ D(G, R), g 7→ δg is an
injective map defined by Diract distribution:
Z
f (x)dδg (x) := f (g).
G

There is a natural map C(G, R) −→ D(G, R) which brings f to f (x)dx.

Let (ρ, V ) be a finite dimensional representation of G. Then we can define a map
Z
ρ : C(G, C) −→ End(V ) : ρ(φ)v = φ(g)ρ(g)vdg.
G

Lemma 1.6.3. The above map is a ring homomorphism, i.e.,

ρ(φ1 ) ◦ ρ(φ2 ) = ρ(φ1 ∗ φ2 ).

Proof. For any v ∈ V , The left hand side equals

Z  Z Z 
ρ(φ1 ) φ2 (g)ρ(g)vdg = φ1 (h)ρ(h) φ2 (g)ρ(g)vdg dh
G G G
Z Z
= φ1 (h)φ2 (g)ρ(hg)vdhdg
Z Z
= φ1 (h)φ2 (h−1 g)ρ(g)(v)dhdg
Z
= (φ1 ∗ φ2 )(g)ρ(g)vdg.

30
 The importance of introducing the above representation is the following:
Proposition 1.6.4. Let V be a finite dimensional irreducible representation of G. Then the
algebra Z(G, C) acts on V is given by a ring homomorphism
1
λV : Z(G, C) −→ C, φ 7→ ⟨φ, χ̄V ⟩.
dim V
Proof. By definition for any φ ∈ Z(G, C),

ρ(g) ◦ ρ(φ) = ρ(φ) ◦ ρ(g).

Thus ρ(φ) ∈ EndG (V ). By Schur’s lemma, it is given by a multiplication by scale λ(φ). To

compute this scalar, we use trace:
1
λ(φ) = tr(ρ(φ)).
dim V
Take a base ei of V and dual e∗i for V ∗ . Then
Z
1 X ∗ 1 X ∗
λ(φ) = ⟨e , ρ(φ)ei ⟩ = ⟨e , φ(g)gei ⟩dg
dim V i i dim V i i G
Z Z
1 X
∗ 1
= φ(g) ⟨ei , gei ⟩dg = φ(g)χV (g)dg
dim V G i
dim V G
1
= ⟨φ, χ̄V ⟩
dim V

Remark
R 1.6.2. The representation ρ also induces a representation of D(G, R) on V by ρ(µ) =
G
ρ(x)dµ(x). More precisely for any v ∈ V ,
Z
ρ(µ)v := ρ(x)vdµ(x).
G

Conversely, from any representation ρ of D(G, R) we can define a representation of G by

ρ(g) := ρ(δg ).

Representations on Hilbert spaces

Definition 1.6.5. Let G be topological group. By a unitary representation ρ of G on a
Hilbert space H , we mean a homomorphism:

ρ : G −→ U (H )

such that the induced map

G × H −→ H
is continuous.

31
 For a typical example, we take a topological space X with measure dµ and a continuous
action by G:
G × X −→ X
such that dµ is invariant under G:
Z Z
f (gx)dµ(x) = f (x)dµ(x), ∀f ∈ Cc (X), g ∈ G.
X X

Then we define a representation of G on L2 (X) by

ρ(g)f (x) = f (g −1 x).

Example 1.6.1 (Regular represenations). Let X = G with a measure dµ and L2 (G) is the
Hilbert space of L2 -functions.

1. If dµ is left-invariant, then we get an action of G on L2 (G) by left multiplication:

L(g)f (x) := f (g −1 x), g, x ∈ G.

2. If dµ is right invariant, then we get an action of G on L2 (G) by

R(g)f (x) := f (xg), g, x ∈ G.

3. If dµ is in both left and right invariant, then we can define an action of G × G on L2 (G)
by
LR(g1 , g2 )f (x) := f (g1−1 xg2 ), g1 , g2 , x ∈ G.

In the following, we want to define the induced action of Hecke algebras using Riesz’s
theorem A.4.7:

Lemma 1.6.6. Let ρ : G −→ U(H ) be a unitary representation of a compact group G on a

Hilbert space. Then there is a unique ring homomorphism

ρ : C(G) −→ B(H , H ), (f, v) 7→ ρ(f )v, ∀v ∈ H , f ∈ C(G)

from the Hecke algebra of G to the ring of bounded operators on H such that for any w ∈ H ,
Z
⟨ρ(f )v, w⟩ = f (g)⟨ρ(g)v, w⟩dg.
G

Proof. It is clear that the above integration is absolutely convergent as ⟨ρ(g)v, w⟩ is a con-
tinuous function on G. For fixed v, it defines a linear functional on H when w varies. This
functional is in fact bounded:
Z Z Z
f (g)⟨ρ(g)v, w⟩dg ≤ |f (g)| · |⟨ρ(g)v, w⟩|dg ≤ |f (g)| · |ρ(g)v| · |w|dg ≤ |f | · |v| · |w|.
G G G

32
By Riesz’s representation theorem A.4.7, this functional defines an element in H , say ρ(f )v.
From above estimate, we see that

|ρ(f )v| ≤ |f ||v|.

Thus ρ defines a map C(G) −→ B(H ). It is easy to check that this is, in fact, a ring
homomorphism, just as in the case of finite-dimensional representations.
We will formally write this representation as
Z
ρ(f )v = f (g)ρ(g)v.
G

The main result of this section is the following:

Proposition 1.6.7. Let G be a compact group and ρ : G −→ U (H ) be a unitary represen-
tation. Let H0 be the sum of all finite-dimensional representations of G in H . Then H0 is
dense in H .
Proof. Let H1 be the orthogonal complement of H0 . It is a Hilbert subspace of H invariant
under G. We need only prove that H1 = 0. Otherwise, take a nozero v0 ∈ H1 and define an
operator T0 by
T0 : H1 −→ H1 : T0 (v) = ⟨v, v0 ⟩v0 .
Then T0 is self-adjoint:
⟨T0 v, w⟩ = ⟨v, v0 ⟩⟨v0 , w⟩ = ⟨v, T0 w⟩.
It is also compact as it has rank 1, and is semi-positive definite:

⟨T0 v, v⟩ = |⟨v, v0 ⟩|2 .

As in above lemma, using Riesz’ theorem, define new operator T ∈ B(H1 , H1 ) such that for
any v, w ∈ H1 , Z
⟨T v, w⟩ = ⟨g ◦ T0 ◦ g −1 v, w⟩dg.
G
We formally write this operator as
Z
T = g −1 T gdg.
G

Then T is also self-adjoint, compact, and non-zero by the following computation:

Z Z
−1
⟨T v0 , v0 ⟩ = ⟨(g ◦ T0 ◦ g )v0 , v0 ⟩dg = |⟨g −1 v0 , v0 ⟩|2 dg > 0.
G G

By Spectral Theorem A.4.4, T has positive eigenvalue λ with finite dimensional eigen
space Vλ . Since T is G-invariant, Vλ is invariant under G. Thus it is a finite dimensional
representation of G. So it belongs to H0 . This contradicts the orthogonality H0 and H1 .

33
Peter–Weyl Theorem
Let G be a compact group with a Haar measure dx of total volume 1. Let L2 (G) be the
corresponding Hilbert space of L2 -functions. Recall that this completes the space C(G, C)
of continuous functions with L2 -norm.
b be the set of finite dimensional irreducible representations of G. Then in §1.5, for
Let G
each V ∈ G,
b we have constructed matrix coefficient map:

AV : V ∗ ⊠ V −→ L2 (G).
In this way we have defined a map
M
A: V ∗ ⊠ V → L2 (G).
V ∈Ĝ

Hλ
L
Definition 1.6.8. Let {Hλ } be a family of Hilbert spaces. Define the compeletion of
with respect to the direct sum norm to be
( )
M X
Hλ =: (vλ ) | vλ ∈ Hλ , 2
∥vλ ∥ < ∞ .
d
λ

Our main result is as follows:

Theorem 1.6.9 (Peter–Weyl Theorem). The map A is injective with dense images.
Proof. By Exercises 1.7.2, 1.7.3, we now that we ImAV are non-zero irreducible representa-
tions of G × G not isomorphic to each other. Thus A is injective. To show that it is dense,
let H be the orthogonal complement of ImA. We want to show that H = 0. Assume
H ̸= 0; Then it is a unitary subrepresentation of G × G by LR. By Theorem 1.6.7, there
is an irreducible representation W of H of G × G. By Exercise 1.7.3, W is isomorphic to
U ⊠ V for some irreducible representations U and V of G.
Now we want to study the action of Hecke algebra C(G, C) on W defined in §1.6 for the
right regular representation R. By Proposition 1.6.4, for any central function φ of G, the
action of R(φ) on W ( copies of V ) is given by scale
1
⟨φ, χV̄ ⟩.
dim W
In particular take φ = χV̄ and f ∈ W , we have
1
R(φ)f = f.
dim W
Now we can compute the left hand side directly:
Z
R(φ)f (x) = φ(g)f (xg)dg = ⟨L(x)−1 f, φ̄⟩.
G

It is clear that L(x)−1 f ∈ W ⊂ H and φ̄ = χV = A(eV̄ ) ∈ Im(A) (see Example 1.5.3). This
implies that R(φ)f = 0. Thus a contradiction.

34
 The isomorphism in the theorem can be viewed as the Fourier expansion of L2 (G) with
respect to RL action on G × G.
Example 1.6.2. Let G = R/Z, V is irreducible representation thus dimension 1. So Vn : x 7→
exp(2πinx), Vn∗ x 7→ exp(−2πinx). The action of LR : G × G −→ U (L2 (G)) is given by

LR(g1 , g2 )f (x) = f (x − g1 + g2 ).

Corollary 1.6.10. The space Z(G, C) of central functions on G has an orthonormal base
consisting of χV for V ∈ G.
b

Let O(G) denote the image of A. We call it the space of matrix coefficients of G. Then
the Peter–Weyl theorem simply says that O(G) is dense in C(G, C). Notice that we can give
a much more elementary description of O(G): it is the subalgebra of G consisting of so-called
G-finite functions, namely the functions f generating finite-dimensional subrepresentations
of G × G From this description, this is clear that O(G) is a subalgebra of C(G, C). Let
Z (G, C) denote the center of O(G). By Exercise 1.7.18, we have

Theorem 1.6.11. The Hecke action defines a ring isomorphism:

∼
M
O(G) −→ End(V )
V ∈G
b

with inverse given by the direct sum of the matrix coefficients maps:
∼
(dim V ) · AV ∗ : End(V ) −→ V ⊠ V ∗ −→ O(G).

In particular, the center Z (G, C) of O(G) is generated by χV as a linear space.

1.7 Exercise
Exercise 1.7.1. Let G1 and G2 be two groups with product G = G1 × G2 . For any represen-
tations V1 and V2 of G1 and G2 respectively, we define a representation V1 ⊠ V2 as the linear
space V1 ⊗ V2 as with G1 × G2 given by

(g1 , g2 )(v1 ⊗ v2 ) = g1 v1 ⊗ g2 v2 .

Show that the nature map induces an isomorphism

V1G1 ⊗ V2G2 −→ (V1 ⊠ V2 )G1 ×G2 .

Hint: write this map in two steps:

W1G1 ⊗ W2G2 −→ (W1G1 ⊗ W2 )G2 −→ ((W1 ⊗ W2 )G1 )G2 .

35
Exercise 1.7.2. Let G1 and G2 be two groups with product G = G1 × G2 . Let V1 , W1 be
finite dimensional representations of G1 , and V2 , W2 be finite dimensional representations of
G2 . Show that the following map is an isomorphism:
HomG1 (V1 , W1 ) ⊗ HomG2 (V2 , W2 ) −→ HomG1 ×G2 (V1 ⊠ V2 , W1 ⊠ W2 ).
Hint: Use the identity Hom(V, W ) = V ∗ ⊗ W and use the previous exercise.
Exercise 1.7.3. Let G1 , G2 be groups, and G = G1 × G2 their product. Let V be a finite
dimensional representation of G. Then V is irreducible if and only if there are irreducible
representations V1 and V2 of G1 and G2 such that
∼
V −→ V1 ⊠ V2 .
Moreover, the pair V1 , V2 is uniquely determined by V up to isomorphisms.
Hint: To show that every irreducible G module is of form V1 ⊠ V2 , take V1 to be any
irreducible representation of V when restrict to the subgroup G1 of G and V2 an irreducible
G2 -submodule of HomG1 (V1 , V ). By Theorem 1.1.9, this gives an isomorphism of G-modules:
∼
V1 ⊠ V2 −→ V.
Exercise 1.7.4. Find all irreducible representations of Q.
Exercise 1.7.5. Let V be a finite dimensional representation of a group G. Show the following
formulae:
∼
V ⊗ V −→ Sym2 V ⊕ ∧2 V.
1
χV (g)2 + χV (g 2 )


χSym2 V (g) =
2
1
χV (g)2 − χV (g 2 ) .


χ∧2 V (g) =
2
Exercise 1.7.6. Let X be a finite set with a group G action. Let V = Fun(X, C) be the space
of functions on G on which G acts by permutations. Show that for any g ∈ G, χV (g) is the
number of set X G of the fixed points of g on X:
χV (g) = # {x ∈ X : gx = x} .
Exercise 1.7.7. Let X be a finite set with an action by finite group G. Let V = Fun(X, C)
with permutation action by G. Let W be the subspace of V functions with a zero-sum of
values. Then we have a decomposition:
V = triv ⊕ W.
This exercise connects the irreducibility of W and transitivity of X.
1. The set Gx for some x ∈ X is called an orbit of X. Let c be the number of distinct
orbits of X. We say that the action of G on X is transitive if c = 1. Show that
c = ⟨χV , 1⟩ = 1 + ⟨χW , 1⟩.
Thus the action of G on X is transitive if and only if ⟨χW , 1⟩ = 0.

36
 2. Let G acts on the product X × X diagonally: g(x, y) = (gx, gy). Denote the corre-
sponding representation of G by

U := Fun(X × X, C).

Show the following decomposition:

∼
U −→ V ⊗ V = triv ⊕ W ⊕2 ⊕ W ⊗ W.

Thus
χU = χ2V = 1 + 2χW + χ2W .

3. Assume that X has at least two elements and that X is transitive in the sense that X
has only one orbit. We say that the action of G on X is doubly transitive if for any
pairs (x, y), (x′ , y ′ ) of points in X × X with x ̸= y and x′ ̸= y ′ there is a g ∈ G such
that g(x, y) = (x′ , y ′ ). Show that the following properties are equivalent:

(a) the action of G on X is doubly transitive;

(b) the action of G on X × X has two orbits, the diagonal and its complement;
(c) ⟨χ2V , 1⟩ = 2;
(d) χW is irreducible.

Exercise 1.7.8. Find all irreducible representations of the alternative group A4 and write
down their characters table.
Exercise 1.7.9. Let G be a finiteP
group. Let C[G] be the group algebra. Thus each elements
can be written as a formal sum g ag g with addition and multiplication defined as follows:
X X X
ag g + bh h = (ag + bg )g
g g g

X X X XX
( ag g) · ( bh h) = ag bh gh = ( agh−1 bh )g.
g,h g h

We may identify C(G, C) with C[G] by the following map:

∼
X
C(G, C) −→ C[G], f 7→ f (g)g.
g∈G

Show that Z(G, C) is the center of C[G].

Exercise 1.7.10. Let (ρ, V ) be a finite dimensional representation of G. Then we can define
a ring homomorphism C[G] still denoted as ρ:
X X
ρ : C(G, C) −→ End(V ) : ρ( ag g)v = ag ρ(g)v.
g∈G g∈G

37
Assume now V is irreducible. Show that the algebra Z(G, C) acts on V is given by a ring
homomorphism
|G|
λV : Z(G, C) −→ C, φ 7→ ⟨φ, χV ∗ ⟩.
dim V
Hint: first show that ρ(Z(G, C)) ∈ EndG (V ). Then apply Schur’s lemma, it is given by
a multiplication by a scale λ(φ). Express this scale by trace:
1
λ(φ) = tr(ρ(φ)).
dim V
Then compute the trace by taking a base and dual base for V and V ∗ .
Exercise 1.7.11. Let V, W be irreducible representations of G. Consider the maps:

AV : End(V ) −→ C(G, C), ρW : C[G] −→ End(W ).

Then (
|G|
dim V
if V = W
ρW ◦ AV =
0 otherwise
Hint: Use Schur’s lemma to show that ρW ◦ AV is a scale. Compute this constant when
V = W by using the identity eV in End(V ).
Exercise 1.7.12. Let G be denote the following Lie group:
  
y x
G= , x, y ∈ R, y > 0 .
0 1

Show that the left invariant is y −2 dxdy, but the right invariant measure is y −1 dxdy.
Exercise 1.7.13. Let G = G1 × G2 be the product of two locally compact topological groups,
and µ = µ1 × µ2 the product of two nonzero measures on G1 and G2 . Prove that µ is left
(resp. right ) invariant if and only if both µ1 , µ2 are left (resp. right) invariant.
Exercise 1.7.14. Consider the product decomposition H× = UH (1) × R×
+ of the multiplicative
group of Hamiltonians.

1. Show that on R× −1
+ , an invariant measure is given by r dr;

2. Show that on H× , an invariant measure is given by

dx1 dx2 dx3 dx4
dg = , g = x1 + x2 i + x3 j + x4 k ∈ H× .
(x21 + x22 + x23 + x24 )2

3. Using 1.7.13 to find an invariant measure for UH (1) = S 3 in terms of angular coordi-
nates (α, β, γ):

x1 = cos α, x2 = sin α cos β, x3 = sin α sin β cos γ, x4 = sin α sin β sin γ.

38
Exercise 1.7.15. Let Vn (C2 ) (n ∈ N) be the representation spaces of G = SU(2) of homoge-
nous polynomials on C2 degree n.
1. Let g ∈ SU(2) with eigenvalues λ±1 . Show the following formula
n
X
χVn (C2 ) = λn−2k .
k=0

2. Use character computation to establish the Clebsch-Gordan formula:

min(n,m)
∼
M
2 2
Vn (C ) ⊗ Vm (C ) −→ Vn+m−2j (C2 ).
j=0

Exercise 1.7.16. Let V be an irreducible representation of G. Then for any ξ, η ∈ V ∗ ⊠ V ,

1
⟨Aξ, Aη⟩ = ⟨ξ, η⟩.
dim V
Moreover for another irreducible representation W , we have

ImAV ⊥ ImAW .

Hint: for the first part, Exercise 1.7.2using the f that V ∗ ⊠ V is an irreducible represen-
tation of G × G to conclude that there is a constant c such that

⟨Aξ, Aη⟩ = c⟨ξ, η⟩.

Then evaluate this constant by taking ξ = η = eV .

Exercise 1.7.17 (Schur’s orthogonality). Let G be a finite group and V and W two irreducible
representations of G. Let v1 ∈ V, v2 ∈ V ∗ and w1 ∈ W, w2 ∈ W ∗ , then
(
1
1 X ⟨v1 , w2 ⟩⟨v2 , w1 ⟩ if V = W
⟨gv1 , v2 ⟩⟨w1 , gw2 ⟩ = dim V
|G| g∈G 0 if V is not isomorphic to W

Exercise 1.7.18. Let G be a compact group. Let V, W ∈ G

b are irreducible representations of
G. Consider the following maps:

AV : End(V ) −→ C(G, C), ρW : C(G, C) −→ End(W ).

Then (
1
dim V
if V = W ∗
ρW ◦ AV =
0 otherwise
Hint: Verify that ρW ◦ AV : V ∗ ⊠ V −→ W ⊠ W ∗ is a homomorphism of irreducible
G × G- modules. Apply Schur’s lemma to conclude that this map to show that this map and
compute the related constant by evaluating the identity element eV in End(V ).

39
Exercise 1.7.19. We assume that G is compact with Haar measure dx of volume 1. Then
∼
V has a hermitian structure which induces an isomorphism V ∗ −→ V̄ . Thus V ⊠ V̄ has an
associated hermitian structure. On the other hand, we can also put a hermitian structure
on C(G, C) by Z
⟨f, g⟩ := f (x)ḡ(x)dx.
G
This pairing is invariant under action of LR.
Let V be an irreducible representation of G. Then for any ξ, η ∈ V ∗ ⊠ V ,
1
⟨Aξ, Aη⟩ = ⟨ξ, η⟩.
dim V
Moreover for another irreducible representation W , we have

ImAV ⊥ ImAW .

Exercise 1.7.20 (Schur orthogonality). Let G be a compact group and V and W two irre-
ducible representations of G. Let v1 ∈ V, v2 ∈ V ∗ and w1 ∈ W, w2 ∈ W ∗ , then
(
1
⟨v1 , w2 ⟩⟨v2 , w1 ⟩ if V = W
Z
dim V
⟨gv1 , v2 ⟩⟨w1 , gw2 ⟩dg =
G 0 if V is not isomorphic to W

Exercise 1.7.21. Let Z (G, C) (resp. Z (S 1 )) be the space of central functions formed by finite
linear combination of characters of representations of G (resp. S 1 ). Then the restriction of
representation induces a ring homomorphism:

Z (G, C) −→ Z (S 1 )
 
0 1
1. Let w = which acts on S 1 by conjugation. Show that the map h : S 1 → G
−1 0
induces a bijection between the quotient S 1 /Ad(w) to the set of conjugacy classes of
G:
∼
S 1 /Ad(w) −→ G/Ad(G).

2. Show that the map the restriction map Z (G, C) −→ Z (S 1 ) is injective with image
included into Z (S 1 )w .

3. Show that Z (S 1 ) = n∈Z Ce2πinx , Z (S 1 )w = ∞

P P
n=0 C cos 2πnx;

4. Show that χVn (h(x)) = nk=0 e2πi(n−2k)x ;

P

5. Show that X X
CχVn (h(x)) = C cos 2πnx.
n≥0 n≥0

40
 6. Show that Z (G, C) −→ Z (S 1 )w is bijective and that
∞
X
Z (G, C) = CχVn .
n=0

Exercise 1.7.22. Let dg be the Haar measure on SU(2) with total volume 1.

1. Show that, Z Z
χV0 = 1, χVn (g)dg = VnG = 0, ∀n > 0.
G G

2. Show that for any φ ∈ Z (G, C),

Z Z 1
φ(g) = 2 φ(h(x)) sin2 (2πx)dx.
G 0

3. Show that for n ≥ 0,

Z
dim EndG (Vn ) = |χVn (g)|2 dg = 1.
G

Thus each Vn is irreducible.

b = {Vn , n ≥ 0}.
4. Show that G

2 Lie groups and Lie algebras

This section introduces basic notions and preliminary structure results about Lie groups and
Lie algebras. We will start with definitions of Lie groups, their Lie algebras, and exponential
maps, followed by correspondences between Lie subgroups and Lie subalgebras. Then we
study solvable Lie algebras, nilpotent Lie algebras, and finally semisimple Lie algebras.

2.1 Lie groups

Definition 2.1.1. A real (resp. complex) Lie group is a smooth real (resp. complex) manifold
G equipped with a group sturcture such that

• The multiplication map

m : G × G → G, (g, h) 7→ gh

• The inversion map

ι : G → G, g 7→ g −1
are both smooth.

41
More precisely, let g, h ∈ G with local coordinates x = (x1 , · · · , xn ) for g, and y = (y1 , · · · , yn )
for y, and z = (z1 , · · · zn ) for gh, and w = (w1 , · · · , wn ) for g −1 . Then the multiplication
map is given by smooth functions z = m(x, y) and w = ι(x).
A Lie group always means either a real or complex Lie group in the following.
Definition 2.1.2. Let G, H be two real (resp. complex) Lie groups. A homomorphism from
G to H is a map f : G → H, both a smooth (resp. analytic) map of manifolds and the
homomorphism of groups.
Example 2.1.1 (Matrix Lie groups). In Example 1.4.1, when R = R (resp. C), we have real
(resp. complex) matrix Lie groups:

GLn (R), SLn (R), Bn (R), Nn (R), On (R), Spn (R).

and real Lie groups

O(n), SO(n), U(n), SU(n), UH (n).
Example 2.1.2. Let G be a Lie group. The map

Ad(g) : G → G, h 7→ g · h · g −1

is a Lie group homomorphism and also an automorphism which fixes the identity element.
It is called the action of G on itself by conjugation.

Representation of Lie groups

Definition 2.1.3. A Representation of a Lie group G on a finite dimensional vector space
V over C is Lie homomorphism
∼
ρ : G −→ GL(V ) −→ GLd (C).

This can also be interpreted as G-module structure on V :

G × V −→ V : (g, v) 7→ ρ(g)v.

One way to construct representations is using actions defined as follows:

Definition 2.1.4. Let G be a Lie group and X a smooth manifold. By a smooth action of
G on X we mean a map:

m : G × X −→ X : (g, x) 7→ g · x

so that m is a smooth map of manifolds and action of sets:

e · x = x, (g1 g2 ) · x = g1 · (g2 · x).

It G acts on X transitively, we call X a homogenous space of G.

42
 Here are some examples of representations
Example 2.1.3. Example 1.4.4, 1.4.5, 1.4.6, 1.4.7.
Example 2.1.4. We would like to discuss the representations for complex Lie algebra C and
C× .

1. For C, the representations are similar to R in 1.4.5. More every precisely, every repre-
sentation is given by ρA ) := eaA where A is a matrix. Two representations ρA and ρB
are isomorphic to each other if and only if A and B are similar to each other. Thus
the isomorphic classes of representations of C are represented by Jordan forms.

2. For C× , we notice that there is an exponential map C −→ C× with kernel 2πiZ. Thus a
representation ρ of C× has the form ρ(ez ) = ezA for some matrix A such that e2πinA = 1
for all n ∈ Z. Replace A by its Jordan form, then we can see that A is semisimple with
eigenvalues in Z. In other words, all homomorphic representations of C× is isomorphic
to the direct sum of characters of the form z 7→ z n for n ∈ Z.

Lie subgroups
Definition 2.1.5. Let G be a Lie group.

1. A Lie subgroup H ⊂ G is a Lie group such that the inclusion map H ,→ G is an injective
immersion and group homomorphism. (Recall that a smooth morphism between smooth
manifolds is called an immersion if its differential map on every point is injective.)

2. A Lie subgroup H of G is called regular if the topology on G induces the topology on

H.

Example 2.1.5. Let G = R × R. Given α ∈ R, define a Lie subgroup Hα = {(x, αx) | x ∈

∼
R} −→ R. It is clear that Hα is regular in G.
Example 2.1.6. Let G = U(1) × U(1). Given α ∈ R, define a Lie subgroup

Hα = {(e2πix , e2πiαx ) : x ∈ R}.

Then Hα is regular if and only if α ∈ Q.

Example 2.1.7. Regular Lie subgroups of GLn (C):

SLn (C), SLn (C), Un , SUn , On , SOn , Spn (C), Sp2n (R), Bn , Nn .

The following is an important criterion of regular Lie subgroups:

Theorem 2.1.6 (Theorem 2.5.6, Theorem 2.3.6, Corollary 2.3.9 ). Let G be a Lie group and
H be a subgroup. Then the following are equivalent:

1. H is closed;

43
 2. H is regular;

3. H is a stabilizer of an action of G on a manifold X.

Moreover, if H satisfies the above condition, then G/H has a unique smooth manifold struc-
ture such that the projection G −→ G/H is a smooth and submersive map. (Recall that a
smooth morphism between smooth manifolds is called an immersion if its differential map on
every point is surjective.)

Example 2.1.8. Let G = SL2 (R) and H = SO(2). We have two way to construct the manifold
structure on G/H.

1. Let G acts on the Poincaré upper half plane H = {z ∈ C, Imz > 0}:
 
a b az + b
·z = .
c d cz + d
∼
This action is transitive with the stabilizer SO(2) at i ∈ H . Thus we have G/H −→
H.

2. Let G acts on its self by g · A = gA t g. Then the stabilizer of the identity matrix 12 is
SO(2). Thus G/H is identified with the orbit of 12 which is the space X ⊂ SL2 (R) of
symmetric positive definite matrices:
  
a b 2
X= : ac − b = 1, a > 0 .
b c

One special case is when H is a regular normal subgroup.

Corollary 2.1.7. A normal subgroup H of G is a regular normal subgroup if and only if H

is the kernel of a homomorphism of Lie groups G −→ G′ .

Example 2.1.9. For G = GLn (R), SLn (R) is the kernel of det : GLn (R) −→ R× .
Example 2.1.10. If G is a Lie group, the center Z(G) is a closed subgroup. Thus we can form
the quotient G/Z(G). For example G = GLn (R), then Z(G) is the subgroup R× 1n . The
quotient is denoted by PGLn (R). It is called the projective general linear group. However,
it is not proper in a topological sense!

Connected components of Lie groups

Let G be a Lie group. Denote G0 for the connect component of G containing the identity
element e ∈ G. Let π0 (G) denote the set of connected components.
∼
Proposition 2.1.8. Then G0 is a normal subgroup of G with quotient G/G0 −→ π0 (G).

44
Proof. It is clear that G0 has a manifold structure since for any p ∈ G0 every connected
open subset U of in G is automatically in G0 .
For the group structure, we see that the multiplication map G0 × G0 −→ G will have
an image connected and containing e. Thus this image is in G0 . So G0 is closed under
multiplication. Similarly, it is closed under the inverse map. So G0 is a Lie group.
To check if it is normal, we see that for any g ∈ G, gG0 g −1 is connected and containing
e, so it is included in G0 . Now we have coset decomposition
a
G= xG0
x∈G/G0

is simply the decomposition into a disjoint union of connected components. Thus we G/G0 =
π0 (G).
Example 2.1.11. The general linear group GLn (R) consists of two connected components
according to the sign of their determinants; the one with matrices of positive determinants
∼
is the unit component GLn (R)0 . Thus π0 (GL2 (R)) −→ Z/2Z.

Coverings of Lie groups

Proposition 2.1.9. Let G be a connected Lie group and π : (G,e ee) −→ (G, e) a universal
cover. Then there is a unique Lie group structure on G
e with unit ee so that π is a homo-
morphism of Lie groups. Moreover the Kerπ is canonically isomorphic to π1 (G, e) and is
included in the center of G.
e

Proof. As (G,
e ee) is a universal cover of (G, e), (G× e ee×e
e G, e) is a universal cover of (G×G, e×e).
By the Universal Lifting Theorem A.2.5, we can lifting multiplication map m and the inverse
map from G to G: e

m e × G,
e : (G e ee × ee) −→ (G,
e ee), e ee) −→ (G,
ei : (G, e ee).

It is easy to see that me and ei makes G

e a Lie group with unit ee. Alternatively, we can use
the fact that (G, ee) is the homotopy classes of path starting from p. Then can define the
e
multiplication by
m([γ
e 1 ], [γ2 ]) = [m(γ1 , γ2 )]
e −→ G). To show this
where m(γ1 , γ2 )(t) = m(γ1 (t), γ2 (t)). In this way π1 (G, e) = Ker(π : G
is in the center of G,
e notice Ker(π : G e −→ G) is discrete and normal, so it is in the center
by Exercise .
For a general Lie group G, we define Ge as the universal cover of G0 and define π1 (G)
as π1 (G, e). Combining the above two lemmas, we obtain the following fundamental exact
sequence:
0 −→ π1 (G) −→ Ge −→ G −→ π0 (G) −→ 0.

45
Example 2.1.12 (SL2 (R)). Let B2+ (R) be the subgroup of upper triangular matrices with
positive entries at the diagonal. Then we have an Iwasawa decompostion:
SL2 (R) = B2+ (R) × SO2 (R).
∼
Since B2+ (R) −→ R+ × R is simply connected, we have that
π1 (SL2 (R)) = π1 (SO2 (R)) = π1 (S 1 ) = Z.
Example 2.1.13 (Matrix groups).
• For n > 1, SOn (R) is not simply connected and SUn is simply connected.
• The group SU2 is topologically isomorphic to the 3-sphere S 3 , and is isomorphic to the
group of unit quaternions as a group.
• The symplectic group Sp2n (R) is connected Lie group of dimension 2n2 + n.

Local group law

In the following, we want to describe the group law of a Lie group G in a neighborhood of
unity e with coordinate x = (x1 , · · · , xn ). Then the multiplication map and the inverse map
are given by smooth functions z = m(x, y) and w = ι(x). They satisfies the law of groups
which we called a local group law. We take Taylor expansions
∞
X ∞
X
m(x, y) = mi (x, y), ι(x) = ιi (x).
i=0 i=0

where mi , ιi are homogenous functions of degree i. Since 0 is the unit of G, we see that
m0 = 0, ι0 = 0.
Since xe = x, ex = e, we see that m1 (x, y) = x + y, In particular, the m2 (x, y) by linear
in x, y which we denote it as x ∗ y. mi (x, y) has degree ≥ 1 for each variable x, y. Thus we
can write
∞
X
(2.1.1) m(x, y) = x + y + x ∗ y + mi (x, y).
i=3

The function ι(x) is completely determined by the equation m(x, ι(x)) = 0. In particular,
it has the form ∞
X
ι(x) = −x + x ∗ x + ιi (x).
i=3

Example 2.1.14. Let G = GLn (R). At e = 1n , we take a coordinate system by Mn (R) by the
map G 7→ Mn (R) : g 7→ g − 1n . Then the local group law is given by
m(x, y) = (1+x)(1+y)−1 = x+y+xy, ι(x) = (1+x)−1 −1 = x−x2 +x3 +· · · , x, y ∈ Mn (R).
Thus we have that the star product is the usual product on Mn (R): x ∗ y = xy.

46
2.2 Lie algebras
Definition 2.2.1. Let k be a field. An k-algebra g is a vector space over k with a product
[X, Y ] for any X, Y ∈ g that is linear in each variable. The algebra is a Lie algebra if the
product satisfies two rules of Lie bracket:

• The skew symmetry: [X, X] = 0 for all X ∈ g.

• The Jacobi identity:

[X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0.

Definition 2.2.2. A homomorphism from a Lie algebra g to to a Lie algebra h is a linear

map φ : g −→ h such that

φ([X, Y ]) = [φ(X), φ(Y )], ∀X, Y ∈ g.

Here are two basic examples used in this lecture:

Example 2.2.1. Let A be a k-algebra (may not with unit). Then we define a Lie algebra
(A, [·, ·]) by
[X, Y ] = XY − Y X.
One case we use very often is R = Endk (V ) with V a k-vector space. In this case we
denote gl(V ) = (Endk (V ), [·, ·]). When V = k n , then we write gln = gl(k n ).
Example 2.2.2. Let A be a commutative k-algebra. Let Derk (A) denote the space of deriva-
tions on A:

Derk (A) = {D : Endk−linear (A), D(xy) = xD(y) + yD(x)} .

Then D is stable under [·, ·] of Endk−linear (A). Thus we obtain a Lie algebra Derk (A).
One case we use often in this course is for a manifold M . In this case, we take A = C ∞ (M )
and DerR (M ) ⊂ Derk (A) the space of smooth vector fields: DerR (M ) = Γ(T M ).

Lie algebra of a Lie group

Let G be a Lie group, and let Te (G) be the tangent space of G at unit element e. Now for
any representation ρ : G −→ GL(V ), take differential at e, we obtain a map

dρ : Te (G) −→ End(V ).

We will construct a Lie bracket [·, ·] on TG,e so that the above map is a Lie algebra repre-
sentation. More precisely, we will show that there is a g has a Lie algebra structure with
bracket given as follows: for any g ∈ G, we have an automorphism:

AdG (g) : G −→ G, x 7→ gxg −1 .

47
Differentiating this map at the unit element induces an automorphism denoted by Adg :

Adg (g) : g −→ g.

Thus we get a representation

Adg : G −→ GL(g), g 7→ Adg (g).

Differentiating this morphism at origin, we get

ad : g −→ End(g).

Thus we have a bracket

[·, ·] : g × g −→ g : [X, Y ] = ad(X)(Y ).

Proposition 2.2.3. Let G be a Lie group. Then the pairing [X, Y ] on Te (G) is a Lie bracket.
Thus we get a Lie algebra g := (Te (G), [·, , ·]).

The proof of the Proposition takes a few steps.

Lemma 2.2.4. In terms of local coordinate x = (x1 , · · · , xn ) for G at e, and the local group
law (2.1.1),
[X, Y ] = X ∗ Y − Y ∗ X.
∼
Proof. We will fix a local coordinate x at e. Now for any X, Y ∈ Te (G) −→ Rn , then we
have curves g(t) = tX, h(s) = sY on G for small t, s ∈ R so that g ′ (0) = X and h′ (0) = Y .
We use x • y = m(x, y) denote the local group law, and compute the ad(X)(Y ) as follows:

d d d
ad(X)(Y ) = |t=0 Ad(gt )(Y ) = |t=0 |s=0 Ad(gt )(hs )
dt dt ds
d d
= |t=0 |s=0 (gt • hs • ι(gt ))
dt ds
Using local group law, we have

gt • hs = tX + sY + ts(X ∗ Y ) + O(t2 ) + O(s2 ), ι(gt ) = −tX + O(t2 ).

It follows that

gt • hs • ι(gt ) = sY + st(X ∗ Y − Y ∗ X) + O(t2 ) + O(s2 ).

Thus we have
ad(X)(Y ) = X ∗ Y − Y ∗ X.

Combining with Example 2.1.14, we obtain:

48
Corollary 2.2.5. Let G be a Lie group. Then the pairing [X, Y ] on Te (G) skew symmetric:
[X, X] = 0. If G = GLn (R) (or GLn (C)), then [X, Y ] is the Lie bracket constructed in
Example 2.2.1.
For Proposition 2.2.3, it remains to prove that [·, ·] satisfies the Jacobi identity for general
Lie group. Before proving this, we notice that the formalism of Lie algebra is functorial in
the following sense: if f : G −→ H is a homomorphism of the Lie group, then dfe on the
tangent spaces at the identity define a homomorphism of Lie algebras:

Lie(f ) : Lie(G) −→ Lie(H).

We apply this to the representation

Ad : G −→ GL(g).

Then we obtain a hommorphism

ad = Lie(Ad) : g −→ gl(g).

Thus we have the identity

ad[X, Y ] = [adX, adY ]
Applying this formula to any element Z ∈ g, we obtain:
Lemma 2.2.6. For any X, Y, Z ∈ g,

[[X, Y ], Z] = [X, [Y, Z]] − [Y, [X, Z]].

Now Proposition 2.2.3 follows from Corollary 2.2.5, Lemma 2.2.6.

Left invariant vector fields

Our first step is to identify g with left invariant vector fields.
Definition 2.2.7. A vector field V : G −→ T (G) on G is left-invariant if ∀g ∈ G, g∗ V (h) =
V (gh) where g∗ = dℓ(g) is the differential for left multiplication map:

ℓ(g) : G → G, h 7→ gh.

If we take h = e, then we get that for any left-invariant V ,

V (g) = g∗ V (e),

thus V is completely determined by V (e) ∈ g. Conversely for any X ∈ g, we define a vector

field r(X) ∈ Γ(T G) by
r(X)(g) := g∗ X.
Then r(X) is an left invariant vector field.

49
Remark 2.2.1. We use the name r(X) because it is the differential of the right regular
representation
R : G −→ GL(C ∞ (G)), R(g)f (h) = f (hg),
r = dR : g −→ gl(C ∞ (G)).
More concretely, let
γ : (−ϵ, ϵ) −→ G
be a curve such that γ(0) = e, γ ′ (0) = X, then we have

d
r(X)f (h) = |t=0 f (hγ(t)).
dt
In fact r is injective and respect to the brackets pairing. This gives another proof that [·, ·]
is a Lie bracket on g.

Exponential maps
In the following, we want to show that there is a unique exponential map:

expG : g −→ G

of smooth manifolds satisfying the following property: for any X ∈ g, the induced morphism

R −→ G : t 7→ etX

is a group homomorphism with

d
(etX ) = X.
dt t=0
Fix X ∈ g. Then the vector field r(X) can be induced to a local flow on G: for any
h ∈ G, there is a unique curve γh : (−ϵ, ϵ) −→ G for some ϵ > 0 such that

γh (0) = h, γh′ (t) = r(X)(γh (t)) = γh (t)∗ X.

Here by convention we write γh′ (t) := d(γh )(d/dt). Now for any g ∈ G, we claim that claim
that g · γh = γX,gh . Indeed, the above identities for gγh can be checked directly:

g · γh (0) = gh,

(g · γh )′ (t) = g∗ γh′ (t) = g∗ γh (t)∗ X = (g · γh (t))∗ X.

We will write γX (t) = γe (t). Then we have γh (t) = hγX (t).

Lemma 2.2.8. For any X ∈ g, γX (t) is defined over all t ∈ R. Moreover γX is a group
homomorphism.

50
Proof. Let (a, b) be the maximal interval containing 0 so that γX is defined. Fix an s ∈ (a, b)
and consider the curve:
ξ(t) = γX (s)−1 γX (s + t), t ∈ (a − s, b − s)
Then it is easy to check that ξ ′ (0) = e, and
ξ ′ (t) = γX (s)−1 ′ −1
∗ γX (s + t) = γX (s)∗ γX (s + t)∗ X = ξ(t)∗ X

Thus ξ is an integral curve for the vector field r(X) on the interval (a − s, b − s), thus
(a − s, b − s) is included into (a, b). This shows that (a, b) = R, and that ξ(t) = γX (t). But
this identity simply means for any s, t ∈ R,
γX (s)γX (t) = γX (s + t).

Definition 2.2.9. Define the exponential map exp : g −→ G by

exp(X) = γX (1).
From the above lemma, we see that d exp = Id and that for any X ∈ g, the map
t 7→ exp(tX)
is a group homomorphism with differential X at 0. This shows that the differential of
exp : g −→ G is the identity map of g. Thus exp is a diffeomorphism near 0. So it has a
local inverse which is called a logarithm. More precisely, there are open neighborhoods V
and U of 0 ∈ g and e ∈ G so that exp and log gives an inverse pair of diffeomorphisms:
∼ ∼
exp : V −→ U, log : U −→ V.
Remark 2.2.2 (Functorility). The formalism of exponential map is functorial in the sense
that for any homomorphism f : G −→ H of Lie groups we have the commutative diagram:
df
g /h

exp exp
 f 
G / H
We give two applications:
1. Let g ∈ G and take f = Ad(g) : G −→ G and write df by Adg (g) on End(g). Then we
obtain for any X ∈ g,
Ad(g)(eX ) = eAdg (g)(X) .
2. Take f = Ad : G −→ GL(g) to obtain that for any X the following identity in GL(g).
∞
X ad(X)
X ad(X)n (Y )
Adg (e )(Y ) = e (Y ) = ,
i=0
n!

where adn (X)(Y ) = [X, [X, · · · [X, Y ]] with X appearing n-times.

51
2.3 Lie subgroups and subalgebras
Lie subalgebras
Definition 2.3.1. Let g be a Lie algebra. A Lie subalgebra of g is a subspace h satisfying

[h, h] ⊂ h.

The main result of this section is the following:

Theorem 2.3.1 (Lie’s second theorem). Let G be a Lie group. If h is a Lie subalgebra of
g = Lie(G), then there exists a unique Lie subgroup H of G such that h = Lie(H). Note that
H may not be regular.

If H is Lie subgroup of Lie group G with Lie algebra h, then for any X ∈ h, exp(X) ∈ H.
Moreover exp(h) contains a neighborhood of H. Thus the uniqueness follows from the
following lemma:

Lemma 2.3.2. Let G be a connected Lie group. Then G is generated by the neighborhood
U of e.

Proof. Let H be the subgroup generated by U . Since ∀h ∈ H the coset hU is a neighborhood

of h in G, H is open. Thus H is further a submanifold of G, also a Lie group. Thus we have
a decomposition of G into a disjoint union
a
G= gH.
g∈G/H

The complement H c of H is the union of gH for g ∈

/ H. Thus H c is open. Thus H is closed.
This shows that H is both open and closed. Since G is connected, we must have G = H.
We will give two approaches to prove its existence. The first one uses the following:

Theorem 2.3.2 (Campbell-Baker-Hausdorff). Let G be a Lie group and g = Lie(G), A, B ∈

g. Then there is a C in the subalgebra of g generated by A and B such that

exp(A) exp(B) = exp(C).

First proof of Theorem 2.3.1. To prove the subgroup theorem, we take H as the set of
exp(X) for X ∈ h. Then the above theorem shows that H is a group. Moreover, in the
neighborhood, we know that exp : h −→ H is a diffeomorphism near a neighborhood of
0 ∈ h. This shows that h = LieH.
For the second approach, we use Frobenius Theorem A.3.31.

52
Second proof of Theorem 2.3.1. For Lie subalgebra h as in the Theorem, we define a vector
subbundle E of T G so that at any g ∈ G, Eg = dL(g)h. Then it is easy to see that Γ(M, E)
is closed under bracket. In fact let Xi be a base of h, then sections X, Y ∈ Γ(G, E) can be
written as X X
X= ϕi (g)R(Xi ), Y = ψj (g)R(Xi )
j

with ϕi , ψj are smooth functions on G. Then

X
[X, Y ] = (ϕi R(Xi )(ψj )R(Xj ) − ψj R(Xj )(ϕi )R(Xi ) + ϕi · ψj R([Xi , Yj ]).
i,j

By Frobenius Theorem A.3.31, for each g ∈ G, there is a submanifold Hg in a neighborhood

of g, such that g ∈ H, and that Th Hg = dL(h)h. We may glueing these Hg ’s to get maximal
connected submanifolds which may not be closed. We still call them Hg . So if h ∈ Hg , then
Hh = Hg .
Write H = He . We claim that H is closed under multiplication. For any h ∈ H, then
h−1 H containing e and such that for any point g ∈ h−1 H, the tangent space of h−1 H is

dL(h−1 )Thg H = dL(h−1 dL(hg)h = dL(g)h.

Thus by the uniqueness of H in Frobenius theorem, h−1 H ⊂ H. This implies that h−1 ∈ H
and H is closed under multiplication; thus H is a subgroup of G with Lie algebra g.

Constructing Lie groups from Lie algebras

In this subsection, we want to prove the following:
Theorem 2.3.3 (Lie–Cartan). Let g be a finite-dimensional Lie algebra and H a Lie group.
∼
1. There is a pair (G, ϕ) where G is a simply connected Lie group and ϕ : h −→ LieG is
isomorphism of Lie algebra. Moreover this pair is unique up to a unique isomorphism
∼
in the sense that for any other pair (G′ , ϕ′ ) there is a unique isomorphism α : G −→ G′
such that ϕ = ϕ′ ◦ dα.

2. For (G, ϕ) as above, we have an bijection:

∼
Hom(G, H) −→ Hom(g, h) : f 7→ df ◦ ϕ.

We use the following theorem without proof for the existence of (G, ϕ).
Theorem 2.3.3 (Ado). Every finite-dimensional Lie algebra is a Lie subalgebra of gln (R)
for n large enough.
This theorem may assume that g is a subalgebra of gln for some n. Then we apply Lie’s
second theorem 2.3.1 to get a connected subgroup G′ of GLn with Lie algebra g. Now take
G=G e′ .

53
Lemma 2.3.4. Let G, H be Lie groups with Lie algebras g and h. Let ρ : g → h be a Lie
algbera homomorphism. Assume that G is simply connected. Then there is a unique Lie
group homomorphism ϕ : G → H such that dϕ = ρ.
Proof. Consider the graph of the mao ρ:

graph(ρ) = {(X, Y ) ∈ g ⊕ h : Y = ρ(X)}.

This is a vector space, and a Lie subalgebra of g ⊕ h. By Lie’s second theorem 2.3.1, there
is a unique connected Lie subgroup K of G × H with graph(ρ) as its Lie algebra. Consider
the composition map
ι p1
ψ : K ,→ G × H → G.
This is a lie group homomorphism, so dψ = dp1 dι is a Lie algebra homomorphism. Since

dp1 : g × h → g

is the projection map, dψ : graph(ρ) → g is bijection. It follows that ψ : K → G is a Lie

group isomorphism by exercise 2.7.9.
Now let ϕ : G → H be the composition
ψ −1 p2
G → K → H.

Then this is a Lie group homomorphism with dϕ = ρ.

Theorem 2.3.5 (Cartan). Every closed subgroup H of G has a group structure that makes
it a regular Lie subgroup of G, and

Lie(H) = {X ∈ Lie(G) | ∀t : exp(tX) ∈ H}.

Proof. Let h denote the right hand side. Then it is a Lie algebra: if X, Y ∈ h, then by
exercise 2.7.7,
2
exp(t(X + Y )) = lim (etX/n etY /n )n , exp([X, Y ]) = lim (etX/n etY /n e−X/n e−Y /n )n .
n−→0 n−→0

The right hand side is in H. Thus X + Y ∈ h and [X, Y ] ∈ h. By Theorem 2.3.1, we have a
connected subgroup H 0 of G with Lie algebra h. Thus H has a Lie group structure so that
H 0 is its connected component. We need to show that the topology using H 0 is the same as
the induced topology from G. This is equivalent to saying that for a small neighborhood U
of G, U ∩ H = U ∩ H 0 . If this is not true, we can find a sequence hn ∈ H \ H 0 convergent
to e ∈ H. We want to show there is contraction.
Let L be a complement of h in Lie(G) so we have direct sum

Lie(G) = h ⊕ L.

Now consider the map

α : Lie(G) −→ G : α(x, y) = exp(x) exp(y), x ∈ h, y ∈ L.

54
Since dα is an identity at 0, we have neighborhhoods V and W of 0 in h and L respectively
so that α is a diffeomorphism on V × W with image U ∈ G. Thus for n large, we may
assume that
hn = exp(xn ) exp(yn ), xn ∈ h, yn ∈ L.
Again xn −→ 0, yn −→ 0. Thus replacing hn by e−xn hn , we may assume that xn = 0. Thus
we may simply assume that hn = eyn with yn ∈ L convergent to 0.
Next we fix a norm | · | on L and consider the sphere SL of vectors of norm 1. Then
yn /|yn | is a sequence in SL . Since SL is compact, it has a limit point. Replace hn by infinite
subsequence, we assume that yn convergent y ∈ L. Now for any t ∈ R, we claim that ty is
in the closure of subgroup of L generated by yn : for any n,

ty = t lim yn /|yn | = lim (t/|yn |)yn .

n→∞ n→∞

Let Nn be the integral part of t/|yn | and ϵn = t/|yn | − Nn . Then |ϵn | ≤ 1. Thus ϵn yn −→ 0.
Thus we may replace t/|yn | by integer Nn . It follows that

ety = lim eNn yn = lim hN

n ∈ H.
n
n→∞ n→∞

This implies that y ∈ h. Thus we have a costradiction.

Homogenous spaces
Let G be a Lie group and H regular subgroups. We want to give a manifold structure on
the quotient G/H.
Theorem 2.3.6. Let H be a regular subgroup of a Lie group G. Then G/H has a unique
smooth manifold structure such that the projection G −→ G/H is a smooth map of manifolds
that is submersive at every point.
Sketch of the proof. Let gH ∈ G/H be any coset. We want to define a smooth neighborhood
gH in G/H. We need only do this when g = e then translate to other points using the action
of G. Let h ⊂ g be the Lie algebras of H and G respectively. Tate a decomposition g = h ⊕ L
for a L a linear subspace of g and take W a small neighborhood of 0 in L Consider the map

α : W −→ G/H : X 7→ eX mod H.

We will show that this is injective when W is sufficiently small. Then we can take W to be
a smooth chart for G/H at H.
We start with two morphisms exp : g −→ G and

β : g = L × h −→ G : (u, v) 7→ eu × ev .

It is easy to see that differentials of both maps are the identity map on g. Thus there is
neighborhood U0 = W0 × V0 of 0 in g = L × h on which both are diffeomorphism. Since

55
H is closed in G, we may further shrink U0 so that H ∩ exp(U0 ) = exp(V0 ). Now take
neighborhood W ⊂ W0 in of 0 in L such that exp(W ) · exp(−W ) ⊂ exp(U0 ). We claim that
W −→ G/H is injective. In fact, for any w1 , w2 ∈ W1 and h ∈ H the identity

h = e−u2 eu1 ,

implies h = ev for a unique v ∈ V0 . Thus

eu2 ev = eu1 .

By the assumption that β is a diffeomorphism on U0 , we have that u1 = u2 , v = 0.

Example 2.3.4 (Iwasawa decomposition). We want to give description of GLn (R)/Bn (R) and
GLn (R)/On (R) as the manifolds using Iwaswa decomposition

GLn (R) = Bn (R)On (R) = On (R)Bn (R).

The last identity is obtained by inverse map g 7→ g −1 . So we need only prove the second
one.
Let g ∈ GLn , its rows (v1 , · · · , vn ) can be viewed as a base of Rn . Then g is in On if and
only if (v1 , · · · , vn ) is an orthonormal base. Now we use Gram-Schmidt process to make an
orthogonal base form g:

u1 = v1 /∥v1 ∥, u2 = (v2 − (v2 · u1 )u1 )/∥(v2 − (v2 · u1 )u1 )∥, · · · ,

X X
un = (vn − (vn · ui )ui )/∥(vn − (vn · ui )ui )∥.
i<n i<n

If we write out ui as a linear combination of vi , then we have

X
ui = bij vj ,
j

with b = (bij ) ∈ Bn . Let k denote the orthogonal matrix whose rows are ui . Then we get
k = bg. Thus g = b−1 k gives the Iwasawa decomposition.
Now we apply this to the coset manifold GLn (R)/Bn (R) use Iwasawa decomposotion.
We know that every coset is presented by an element in On . Thus we have

GLn (R)/Bn (R) = On (R)/(On (R) ∩ Bn (R)).

Now it is easy to see that On ∩ Bn = O1n where O1 = {±1}. The O1n is embedded into
GLn (R) as the subgroups of diagonal matrices with entries ±1. So we obtain
∼
GLn (R)/Bn (R) −→ On /O1n .

This implies that GLn (R)/Bn (R) is compact.

56
 Using Iwasawa decomposition, we also get
∼
GLn (R)/On −→ Bn (R)/O1n = Bn (R)+

where Bn (R)+ is the group of upper triangular matrices with positive entries in the diagonal.
We can also do the computation for GLn (C). Then we have Iwasawa decomposition

GLn (C) = Bn (C) · Un = Un · Bn (C).

It follows that
∼ ∼ ∼
GLn (C)/Bn (C) −→ Un /U1n , GLn (C)/Un −→ Bn (C)/U1n −→ Bn+ (C).

One special case is when H is a regular normal subgroup.

Corollary 2.3.7. A normal subgroup H of G is a regular normal subgroup if and only if H

is the kernel of a homomorphism of Lie groups G −→ G′ .

Example 2.3.5. For G = GLn (R), SLn (R) is the kernel of det : GLn (R) −→ R× .
Notice that G acts on G/H multiplication. It is clear that this action is smooth in the
following sense:

Lemma 2.3.8. Let m : G × X −→ X be a smooth action of a Lie group acting on a smooth

manifold. Then for any x ∈ X, the stabilizer Gx = {g ∈ G, gx = x} is a regular subgroup.

Proof. Consider the continuous map mx : G −→ X : g 7→ gx. Then Gx = m−1

x (x). Since x
is closed in X, Gx is closed in G.
Combining above theorem and lemma, we get

Corollary 2.3.9. A subgroup H of a Lie group G is regular if and only if it is a stabilizer

subgroup for smooth action of G on a homogenous space.

Example 2.3.6 (Complete flag manifolds). For a positive integer n, let Fn (R) denote the set
of complete flags on R: the set of filtrations F :

F : 0 = F0 ⊂ F1 ⊂ · · · ⊂ Fn = Rn , dim Fi = i

The action of GLn on Rn gives an action of GLn on Fn :

gF : 0 = gF0 ⊂ gF1 , · · · , gFn = Rn .

Let F 0 be the standard flag:

F 0 : 0 ⊂ Re1 ⊂ Re1 + Re2 ⊂ Re1 + · · · Ren−1 ⊂ Rn .

Then it is easy to see the following:

57
 1. for every flat F , there is a g ∈ GLn (R) such that F = gF 0 ;

2. the stabilizer of F 0 is Bn .

It follows that the map g 7→ gF 0 induces a bijection:

∼
GLn (R)/Bn (R) −→ Fn (R).

This shows that Fn (R) has a manifold structure isomorphic to On /O1n . In particular, Fn (R)
is compact. In fact, we can write down isomorphism Fn (R) −→ On /O1n geometrically. This
will give a geometric description of Iwaswa decomposition.
Let En (R) be the set of full decomposition of Rn into orthogonal sums of lines:

E : Rn = E1 ⊕ E2 ⊕ · · · ⊕ En .

Then the orthogonal group On acts on En (R) transitively with stabilizer O1n on the standard
decomposition Fi = Rei . Thus we have

En (R) = On /O1n .

Now we can connect these two sets by a map

En (R) −→ Fn (R) : E = (Ei ) 7→ 0 ⊂ E1 ⊂ E1 ⊕ E2 ⊂ · · · Rn .

This is actually bijiective: for each flat F , we take Ei the orthognal complement of Fi−1 in
Fi .
∼ ∼ ∼
GLn (R)/Bn (R) −→ Fn (R) −→ En (R) −→ On /O1n .
We may do the same theory for GLn (C) to get a complex manifold of full flags:
∼
Fn (C) −→ GLn (C)/Bn (C).

We can also identify Fn (C) with the set En of decompositions of Cn into orthogonal direct
sums of lines. Thus we have a diffeomorphism of real manifolds:
∼
Fn (C) −→ Un /U1n .

This at least show that Fn (C) is a compact complex manifold.

Example 2.3.7 (Partial flag manifold). Consider an increasing sequence of integers:

0 = n0 < n1 < n2 ≤ nr−1 < nr = n

Let Fn1 ,n2 ,··· ,nr (R) denote the set of flags on R of multiple dimensions ni : the set of filtrations
F:
F : 0 = F0 ⊂ F1 ⊂ · · · ⊂ Fr = Rn , dim Fi = ni .
For the special case n = 1 + (n − 1), this is space of lines, thus F (1, n) = Pn−1 (R).

58
 The action of GLn on Rn gives an action of GLn on F (n1 , · · · , nr )(R):

gF : 0 = gF0 ⊂ gF1 , · · · , gFr = Rn .

Let F 0 ∈ F (n1 , · · · , nr ) be the standard flag

F 0 : 0 ⊂ Rn1 ⊂ Rn2 ⊂ · · · ⊂ Rnr−1 ⊂ Rn .

Then one can show that for every flat F , there is a g ∈ GLn (R) such that F = gF 0 . Let
Pn1 ,n2 ,··· ,nr be the stabilizer of GLn (R) at F 0 . Then we have a bijection:
∼
GLn (R)/Pn1 ,··· ,nr (R) −→ Fn1 ,··· ,nr (R).

This bijection gives a manifold structure on the flag Fn1 ,··· ,nr .
Similarly, let di = ni − ni−1 , consider the set Ed1 ,d2 ,···dr (R) of full decomposition of Rn
into orthogonal sums of subspaces:

E : Rn = E1 ⊕ E2 ⊕ · · · ⊕ Er , dim Ei = di

Then the orthogonal groupPOn acts transitively on Ed1 ,··· ,dr (R). Let E 0 be the standard
decomposition with Ei0 = ni=n i
i−1 +1
Rei . Then the stabilizer of On on E 0 is Od1 × Od2 ×
· · · × Odr . Thus we have a bijection
∼
On /(Od1 × Od2 × · · · Odr ) −→ Ed1 ,d2 ,··· ,dr (R) : g 7→ gE 0 .

This bijection gives a manifold structure on Ed1 ,··· ,dr (R).

We can now define a map
X
Ed1 ,··· ,dr (R) −→ Fn1 ,n2 ,··· ,nr (R) : (Ei ) 7→ (Fi = Ej ).
j≤i

Again, this map is bijective. The inverse map takes F = (Fi ) to E = (Ei ) with Ei the
orthogonal complement of Fi−1 in Fi . In this way, we obtain a homeomorphism:
∼
GLn (R)/Pn1 ,n2 ,··· ,nr −→ On /(Od1 × Od2 × · · · Odr ).

In the spacial case d1 = 1, d2 = n − 1, we get a description of Pn−1 (R):

Rn−1 (R) = On /(O1 × On−1 ).

Let S n−1 be the unit sphere of norm 1 vectors in Rn . Define a map On −→ S n−1 by g 7→ ge1 .
Then this is surjective with stabilizer On−1 . Thus we get

Rn−1 (R) = S n−1 /{±1}.

This agrees with our previous description of the various structure of Pn−1 (R).

59
2.4 Solvable Lie algebras
In this subsection, we assume that g is Lie algebra over a field K characteristic 0, and define
the notions “nilpotent”, “solvable”, “simple”, “semi-simple” and “radical”. Notice the Lie
bracket defines the adjoint representation:

ad : g −→ gl(g), ad(X)[Y ] := [X, Y ], ∀X, Y ∈ g.

Definition 2.4.1. Let g be a Lie algebra and a linear subspace a of g is called an ideal of g
if [g, a] ⊂ a. In other words, a is a subrepresentation of ad.
Example 2.4.1. Let φ : g → h be a morphism of Lie algebras. Then Kerφ ,→ g, Imφ ,→ h
are Lie subalgebras and Kerφ is also an ideal. Conversely, if a is an ideal of g, then we can
make a quotient Lie algebra g/a. Notice that kernel of ad is the center Z(g) of g.
Example 2.4.2. If a, b are ideals of Lie algebra g, then so are a + b, a ∩ b and [a, b].
Example 2.4.3. For the Lie algebra gln , we have the following subalgebras:
• Borel subalgebra bn : the algebra of upper triangular matrices (xij ) with xij = 0 for all
j < i;
• un nilpotent algebra: the algebra of strictly upper triangular matrices (xij ) with xij = 0
for all j ≤ i;
• Let g be a Lie algebra. [g, g] is Lie subalgebra of commutators, and is acturally an
ideal of g such that g/[g, g] is the maximal abelian quotient
• [gln , gln ] = sln = [sln , sln ]
∼
• [bn , bn ] = un , bn /un −→ K n .
Definition 2.4.2 (Commutator series). Let g be a finite dimensional Lie algebra. Define

g0 := g, g1 := [g, g], gn+1 := [gn , gn ]

Then the decreasing sequence

g0 ⊃ g1 ⊃ g2 ⊃ · · ·
is called the commutator series for g. We call g is solvable if gn = 0 for some n. So
commutative Lie algebra is solvable.
Definition 2.4.3 (Nilpotent series). Let g be a finite dimensional Lie algebra. Define

g0 := g, g1 := [g, g], gn+1 := [g, gn ]

Then the decreasing sequence

g0 ⊃ g1 ⊃ g2 ⊃ · · ·
is called the lower central series for g. We call g is nilpotent if gn = 0 for some n. Inductively
we see that gn ⊂ gn , and it follows that nilpotent Lie algebras are solvable.

60
Example 2.4.4. An 1-dimensional Lie algebra g over K must have [X, X] = 0 if {X} is a
basis. Thus g must be abelian and nilpotent.
Example 2.4.5. Let g be a 2-dimensional Lie algebra, and {X, Y } a basis. The expansion of
[X, Y ] in terms of X and Y determines the Lie algebra up to isomorphism. If [X, Y ] = 0,
then g is abelian. Otherwise let [X, Y ] ̸= 0. We claim that g is isomorphic to the Borel
subalgebra b2 of sl2 . In orther words g had a base {H, E} with relation [H, E] = 2E. Set
Write E = [X, Y ] = αX + βY . Then one of α, β ̸= 0. Then [X, E] = βE and [Y, E] = −αE.
So we can take H = 2E/β if β ̸= 0, or H = −2E/α if α ̸= 0. We conclude that only possible
2-dimensional Lie algebras g over the field K up to isomorphism are
∼
• g −→ K 2 is abelian thus nilpotent;
∼
• g −→ b2 is solvable but not nilpotent.

Example 2.4.6. • for n ≥ 2, gln , sln are neither solvable nor nilpotent;

• bn is solvable but not nilpotent;

• nn is nilpotent.

Lemma 2.4.4. If a is a solvable ideal in Lie algebra g and if g/a is solvable, then g is
solvable.

Proof. Let ρ : g → g/a be the quotient homomorphism, and suppose that (g/a)n = 0 for
some positive integer n. Since ρ(g) = g/a, ρ(gn ) = (g/a)n = 0 we conclude that gn ⊂ a. By
assumption am = 0 for some m. Hence gm+n = (gn )m ⊂ am = 0, and g is solvable.
For finite-dimensional Lie algebra, there will be a “maximal” solvable ideal; we first prove
the following property:

Lemma 2.4.5. Le g be a finite dimensional Lie algebra. Then the set of solvable ideals has
a maximal element.

Proof. Let a, b ,→ g be two solvable ideals. Then claim that a + b is also a solvable ideal. In
fact [g, a + b] = [g, a] + [g, b] ,→ a + b so there is an exact sequence

0 → a → a + b → b/(a ∩ b) → 0.

Since a and b are solvable, so are b/(a ∩ b). Thus a + b is solvable.

Definition 2.4.6. We call the maximal solvable ideal the radical of g and denote it by
Rad(g).

Example 2.4.7. The Rad(gln ) = Nil(gln ) is the center of gln :

gln = K · Id ⊕ sln

61
Simple and semi-simple Lie algebras
Definition 2.4.7. A finite-dimensional Lie algebra g is simple if dim g > 1 and has no
proper nonzero ideals; we say g is semi-simple if g has no nonzero solvable ideals, i.e., if
Rad(g) = 0.

Example 2.4.8. Consider the Lie algebra

  
a b
sl2 (K) = , a, b, c ∈ C .
c −a

It has a distinguished basis

     
1 0 0 1 0 0
h= , e= , f= .
0 −1 0 0 1 0

In this basis the bracket relations are

[h, e] = 2e, [h, f ] = −2f, [e, f ] = h.

In particular, e, f, h are eigen vectors for ad(h) with eigenvalues 2, −2, 0. Let a be a non-zero
ideal of sl2 . Then a is a sum of eigen spaces under ad(h). It follows that one of e, f, h ∈ a.
From the above formula, we have that e, f, h are all in a. Thus sl2 (K) is simple.

Proposition 2.4.8. Let g be a simple Lie algebra, then [g, g] = g and g is semi-simple.
Every semisimple Lie algebra has 0 center.

Proof. Since g is simple and [g, g] is also an ideal, hence is 0 or g. Since g is not commutative
it has to be g. For the second argument consider the ideal Rad(g), it is either 0 or g.
If Rad(g) = g, then g is solvable and [g, g] is a nonzero proper ideal, contradiction. So
Rad(g) = 0, and g is semi-simple. Finally Z(g) is an abelian ideal so must be 0.
For any finite dimensional Lie algebra g there is an exact sequence

0 → Rad(g) ,→ g → g/Rad(g) → 0

Proposition 2.4.9. Let g be a finite-dimensional Lie algebra, then g/Rad(g) is semi-simple.

Proof. Let ρ : g → g/Rad(g) be the quotient homomorphism. Let h be a solvable ideal

in g/Rad(g). The preimage a = ρ−1 (h) ⊂ g, then ρ(a) = h is solvable, and Kerρ|a is
solvable in Rad(g). So a is solvable. Hence a ⊂ Rad(g) and h = 0. Therefore g/Rad(g) is
semisimple.

62
Representations of solvable Lie algebras
This section assumes that K is an algebraically closed field of characteristic 0.
Theorem 2.4.9 (Lie). Let g be a solvable Lie algebra. The following are equivalent:
1. g is solvable;

2. every irreducible representations of g is one dimensional;

3. every representation ρ : g → gl(V ) triangularizable: there is a basis e1 , · · · , en of V

such that ρ(g) is included in the Borel subgroup bn .
Proof. Let us first prove (1) =⇒ (2). Assume that g is solvable. Let V be an irreducible
representation of g. We want to show that V is one-dimensional. We will prove this by
induction on dim g. There is nothing need to prove if g = 0. Otherwise [g, g] ̸= g.
So we have a linear space h ⊂ g of codimension 1 and containing [g, g]. Then h is an
ideal of g:
[g, h] ⊂ [g, g] ⊂ h.
Let X ∈ g \ h. Then g = h + KX.
We apply induction to h to get a non-zero vector v ∈ V such that hv ⊂ Kv. Thus there
is a homomorphism λ : h −→ K of Lie algebras such that Y v = λ(Y )v for all Y ∈ h. We
claim that V has a basis given by v, Xv, · · · , X n−1 v with n = dim V . Since V is irreducible,
it suffices to show that for any Y ∈ g, Y X i v is a linear combination of X j v. This is obvious
for Y = X. So we may assume that Y ∈ h. We can do better by proving the following
formula
X
(2.4.1) Y X i v = λ(Y )X i v + aij λ(Yi−j )X j v, aij ∈ K, Yk = ad(X)k (Y ) ∈ [g, g].
j<i

To prove this formula, we use induction on i and the formula in End(V ):

X
Y X i = X iY − X a ad(X)(Y )X b .
a+b=i−1

By formula (2.4.1), for all Y ∈ h, using base v, Xv, · · · , X n−1 v, we find that

tr(ρY ) = nλ(Y ).

Apply this for [g, g] ⊂ h, we get

1
λ(Y ) = tr(ρY ) = 0, ∀Y ∈ [g, g]
n
Now the above formula (2.4.1) show that h acts on whole V by character λ. Take w ∈ V be
any eigen vector for X. Then Kw is an invariant subspace of V . By irreducibility, we have
V = Kw.

63
 Now let us prove that (2) =⇒ (3). We will use induction on n = dim V . Let W be an
irreducible subrepsentation of V . Then by part (2), W is generated by one element, say en
as in and apply the theorem for V ′ = V /Ken if n > 1. Then we will get a base e′1 , · · · e′n−1 as
in the theorem for V ′ . Lifting e′1 , · · · , e′n−1 ∈ V ′ to some elements e1 , · · · , en−1 ∈ V . Then
e1 , · · · , en is the base as in part (3).
Finally, it remains to prove (3) =⇒ (1). Consider the representation ad : g −→ gl(g).
Then we have an exact sequence:

0 −→ Z(g) −→ g −→ ad(g) −→ 0.

By part (3), ad(g) is triangularizable: with ad(g) ⊂ nn for some isomorphism g −→ Cn (as
linear spaces). Thus ad(g) is solvable. By Lemma 2.4.4, since Z(g) is solvable, ad(g) =
g/Z(g) is solvalble, g is solvable.

2.5 Nilpotent Lie algebras

Let g be a finite dimensional Lie algebra. Then g is nilpotent if and only if there is a k > 0
such that gk = 0. More precisely, this means

[X1 , [X2 , [· · · [Xk−1 , Xk ] · · · ]]] = 0, ∀X1 , · · · , Xk ∈ g.

Using the adjoint representation

ad : g → gl(g), X 7→ (Y 7→ [X, Y ]),

we can write the above condition as

ad(X1 )ad(X2 ) · · · ad(Xk−1 )(Xk ) = 0.

Lemma 2.5.1. The g is nilpotent if and only if there is some k ≥ 1 such that

ad(X1 )ad(X2 ) · · · ad(Xk ) = 0

in End(g) for all X1 , · · · , Xk ∈ g.

This implies in particular that ad(X)k = 0 for all X ∈ g.
Definition 2.5.2. A representation ρ : g −→ gl(V ) is called a nilpotent representation if
for any X ∈ g, ρ(X) is nilpotent in End(V ).
Definition 2.5.3. Let g be a Lie algebra. We say X ∈ g is ad-nilpotent if ad(X) ∈ gl(g) is
nilpotent.
Theorem 2.5.1 (Engel). Let g be a Lie algebra. If ad : g −→ gl(g) is nilpotent, then g is
nilpotent.
In view of Lemma 2.5.1, it suffices to prove the following for ad(g) ⊂ gl(g):

64
Theorem 2.5.4. Let g −→ gl(V ) be a nilpotent representation of Lie algebras. Then there
is a basis e1 , · · · , en of V with which g is included into nn (V ). In other words, for any X ∈ g,
we can write X
Xei = aij ej , aij ∈ K.
j>i

In particular, for any X1 , · · · , Xm ∈ g with m > dim V , their product in End(V ) vanishes:

X1 X2 · · · X m = 0

As in the case of solvable Lie algebra that Theorem 2.4.9 is equivalent to Corollary ??,
the above theorem is equal to the following:

Corollary 2.5.5. Let V be a positive dimensional space and g ⊂ gl(V ) be a Lie subalgebra
consisting of nilpotent elements in End(V ). Then there is a non-zero vector v ∈ V such that
Xv = 0 for all X ∈ g.

Proof. As in the proof of Theorem 2.4.9, we may do induction dim g. There is nothing to
prove if g = 0. Otherwise, we want to construct an ideal h with codimension 1. If we have
this h, then we can write g = h + KX. Then by induction the null space

W := {v ∈ V : Y v = 0, Y ∈ h} =
̸ 0.

On the other hand, this space is invariant under X, thus containing a non-zero eigenvector
v of X. Since X is nilpotent, the only eigenvalues are 0. Thus Xv = 0. It follows Av = 0
for all A ∈ g.
To construct h, we take h to be the largest Lie subalgebra not equal to g. We claim that h
is an ideal of g which then immediately implies codh = 1. It suffices to that Ng (h) ̸= h which
implies that Ng (h) = g. We consider the adjoint action of ad : h −→ gl(g/h). Notice that
this representation is a subquotient of the representation h −→ gl(V ) −→ gl(End(V )). By
Lemma A.1.4, every element ad(h) is nilpotent in gl(End(V )). Thus by induction, there is an
w ∈ g/h, such that ad(h)(w) = 0. Take X ∈ g representing w, then we have ad(X)(h) ⊂ h.
This show that Ng (h) ̸= h.

Cartan’s Criterion
In this section, we work with field K = C. In this the following, we translate Engle’s theorem
in terms of traces.

Theorem 2.5.6. Let V be a finite dimensional space over C and g ⊂ gl(V ) be a Lie subal-
gebra. Assume that tr(XY ) = 0 for all X ∈ [g, g] and Y ∈ g. Then g is solvable.

Proof. It suffices to show that every element X ∈ [g, g] is nilpotent. Then by Theorem 2.5.4
we will have that [g, g] is nilpotent and that g is solvable. Take the a Jordan–Chevelley
decomposition as Lemma A.1.3: X = Xs + Xn with Xs semisimple and Xn nilpotent and
commute to each other. We need to show Xs = 0.

65
 Take a base e1 , · · · , en for V for which Xs is diagonal and Xn ∈ nn in strictly upper-
triangular. So we have Xs ei = λi ei with λi ∈ C. Then we have
X
tr(X̄s X) = |λi |2 .

Thus it suffices to show that tr(X̄s X) = 0. Since X ∈ [g, g], it suffices to prove that for all
A, B ∈ g.
tr(X̄s [A, B]) = 0.
Rewrite this equation in a different way,

tr(X̄s [A, B]) = tr([X̄s , A]B) = tr(adgl(V ) (X̄s )(A) · B)

where adgl(V ) : gl(V ) −→ gl(gl(V )) denote the adjoint map for Lie group gl(V ).
We claim that there are polynomials Q, R ∈ T C[T ] such that

adgln (X̄s ) = Q(adgln Xs ), adgln (Xs ) = R(adgln X).

Assume this claim and take P = QR, then

adgln (X̄s ) = P (adgln (X)).

Thus adgln (X̄s )(A) ∈ [g, g]. Since tr([g, g] · g) = 0, we have

tr(adgln (X̄s )(A) · B) = 0

This finishes the proof of the Theorem.

It remains to prove our claim. For polynomial Q, we notice that adgln (Xs ) and adgln X̄s ),
have eigenvalues for the bases Eij with eigenvalues λij := λi − λj and λ̄ij := λ̄i − λ̄j
respectively. Let Q ∈ T C[T ] be a polynomial which takes λij to λ̄ij . Then we have
adgln (X̄s ) = Q(adgln Xs ). For polynomial R, we Lemma A.1.5.

Theorem 2.5.7. A Lie algebra g is solvable if and only if Bg (X, Y ) = 0 for all X ∈ g and
Y ∈ [g, g].

Proof. We use representation ad : g −→ gl(g). The kernel of this representation is the center
of g. Thus g is solvable if and only if Im(ad) is solvable. Thus we can apply Theorem
2.5.6
In the following, we want to deduce a criterion for semisimple Lie algebra in terms of
Killing forms. The primary tool is radical for the Killing form.

66
Killing forms
In the rest of this section, we work with field K = C. Let g be a Lie algebra and ad : g −→
gl(g) the associated adjoint representation.

Definition 2.5.8. The Killing form Bg on g is a bilinear form defined by

Bg : g × g −→ C, (X, Y ) 7→ tr(adX · adY ).

It is clear that B is bilinear and symmetric. Thus B can be considered an element in

Hom(g ⊗ g, C).

Notice that the action ad of g on itself defines an action of g on g ⊗ g. We claim that B is

invariant under ad. This means

Lemma 2.5.9. For any X, Y, Z ∈ g,

B([X, Y ], Z) = B(X, [Y, Z]) = B([Z, X], Y ).

Proof. Since ad is a Lie algebra homomorphism, we need only prove the following:

tr([adX, adY ]adZ) = tr(adX[adY, adZ]) = tr([adZ, adX]adY ).

This is a clear identity for End(g).

Another property of the Killing form is as follows:

Lemma 2.5.10. Let g be a Lie algebra and h an ideal. Then

Bh (X, Y ) = Bg (X, Y ), ∀X, Y ∈ h.

Proof. Let X, Y ∈ h and P := adXadY ∈ End(g). Then P takes g −→ h. Thus we can

compute the trace using the exact sequence

0 −→ h −→ g −→ g/h −→ 0

trg (P ) = trh (P ) + trg/h (P ) = trh (P ).

Example 2.5.2. Consider sl2 (C) in Example 2.4.8, with respect to base e, f, h, the Killing
form is represented by a matrix
   
B(e, e) B(e, f ) B(e, h) 0 4 0
Bg = B(f, e) B(f, f )
 B(f, h) = 4 0 0 .
B(h, e) B(h, f ) B(h, h) 0 0 8

67
Definition 2.5.11. Let β : V × V −→ C be a symmetric bilinear form. Then the radical of
β is define to be
Radβ = {x ∈ V : β(x, V ) = 0}.

By Lemma 2.5.9, it is clear that Rad(Bg ) is an ideal. We have the following inclusion
relations among radicals. Recall that Nil(g) is the maximal nilpotent ideal of g as in Exercise
2.7.12.

Theorem 2.5.12. Let g be a Lie algebra. Then the radical Rad(Bg ) of the B satisfies the
following relation:

[Rad(g), Rad(g)] ⊂ Nil(g) ⊂ Rad(Bg ) ⊂ Rad(g).

Proof. The first inclusion is proved in Exercise 2.7.11.

For the second inclusion, let k > 0 be an integer such that Nil(g)k = 0. Then for any
X ∈ Nil(g), Y ∈ g, we have

(adX · adY )k+1 (g) ⊂ (adX · adY )k Nil(g) ⊂ (adX · adY )k−1 Nil(g)1 ⊂ · · · ⊂ Nil(g)k = 0

For the last inclusion, apply the above lemma to the ideal h = Rad(Bg ) of g which shows
that Bh = 0. Thus by Theorem 2.5.6, h = Rad(Bg ) is solvable.
By Exercise 2.7.12, we have the following criterion of semisimple Lie algebras:

Theorem 2.5.13. A Lie algebra g is semisimple if the Killing form Bg is non-degenerate.

2.6 Semisimple Lie algebras

For a collection of Lie algebras gi (i ∈ I), we can define their direct sum g := ⊕i∈I gi as the
direct sum of underlying vector spaces with component-wise Lie bracket. It is easy to see
that the Killing form of g is then the direct sum of the Killing forms of gi .

Theorem 2.6.1. Let g be a semi-simple Lie algebra, and H the set of simple ideas (which
are also minimal non-zero ideals). Then the natural homomorphism φ : ⊕h∈H h −→ g of
linear space is an isomorphism of Lie algebras.

Proof. First we show that φ is a homomorphism of Lie algebras. This follows from the fact
that any different simple ideals h1 , h2 commute to each other:

[h1 , h2 ] ∈ h1 ∩ h2 = 0.

Thus we need to show that φ is bijective.

Let Bg be the Killing form of g. Then its restriction on each h ∈ H is Bh . First, let us
prove that these ideals are orthogonal to each other. This will show that φ is an injective
homomorphism of Lie algebras.

68
 Let X1 ∈ h1 , X2 ∈ h2 , then we have that adX1 · adX2 : g −→ g has image in [h1 , h2 ] = 0.
Thus adX1 · adX2 = 0 and

Bg (X1 , X2 ) = tr(adX1 · adX2 ) = 0.

P
This shows that φ is injective: if 0 = φ(⊕h∈H xh ) = h∈H xh , then for any fixed h0 ∈ H, and
y ∈ h0 , we have X X
0 = Bg (y, xh ) = Bg (y, xh ) = Bh0 (y, xh0 ).
h∈H h∈H

Thus x0 is perpendicular to h0 . Since h0 is simple, x0 = 0.

To show that φ is surjective, let h = Imφ and h⊥ be the orthogonal complement of h.
Then h⊥ is an ideal: for any X ∈ g, Y ∈ h⊥ , Z ∈ h,

Bg ([X, Y ], Z) = Bg (Y, [Z, X]) = 0

because [Z, X] ∈ h. It is clear that h is semisimple, so Bg |h = Bh is non-degenerate. Thus

we have
g = h ⊕ h⊥ .
If h⊥ ̸= 0, then it has a simple subideal h not included into H. This is impossible.

Corollary 2.6.2. Let g be a semisimple Lie algebra.

1. g = [g, g];

2. every ideal of g is semisimple;

3. every quotient of g is semisimple.

Jordan–Chevalley decompositions
For the following discussion, we need to have a Jordan–Chevalley decomposition for semisim-
ple algebras.

Theorem 2.6.3. Let g be a semisimple Lie algebra and X ∈ g be an element. Then there
is a unique decomposition X = Xs + Xn such that for any representation ρ : g −→ gl(V ),
ρ(X) = ρ(Xs ) + ρ(Xn ) is the Jordan–Chevalley decomposition of ρ(X).

Proof. Since g is a direct sum of its simple ideals hi , it is easy to reduce the case g is simple.
Also, since ρ is completely reduced, we need only prove the case where ρ is irreducible. This
follows from Exercises 2.7.15, 2.7.16. More precisely, these exercises show that for every
irreducible representation of simple g there is a decomposition X = Xs + Xn with following
two properties:

1. ρ(X) = ρ(Xs ) + ρ(Xn ) is the Jordan–Chevalley decomposition of ρ(X) in gl(V );

2. ad(X) = ad(Xs ) + ad(Xn ) is the Jordan–Chevalley decomposition of ad(X) ∈ gl(g).

69
Since ad : g −→ gl(g) is injective, we see that the decomposition X = Xs + Xn does not
depend on the choice of ρ.

Definition 2.6.4. Let g be a semisimple Lie algebra and X ∈ g an element. Then the sum
X = Xs + Xn as in the above theorem is called the Jordan–Chevalley decomposition of X.

Corollary 2.6.5. For any homomorphism φ : g −→ h to a semisimple Lie algebras and any
Jordan–Chevalley decomposition X = Xs + Xn in g, φ(X) = φ(Xs ) + φ(Xn ) is a Jordan–
Chevalley decomposition in h.

Reductive Lie algebras

Definition 2.6.6. A Lie algebra g is called reductive its radical is equal to its center:

Rad(g) = Z(g).

Proposition 2.6.7. Let g be a Lie algebra over R or C. The following are equivalent.

1. g is reductive;

2. g is completely reducible an module over ad(g);

3. [g, g] is semisimple, and g = [g, g] ⊕ Z(g);

4. [g, g] ∩ Rad(g)] = 0.

Proof. For 1 =⇒ 2, we notice that the action of ad(g) = g/Z(g). If g reductive then g/Z(g)
is certainly semisimple. So g is completely reducible under ad(g).
For 2 =⇒ 3, if g is completely reducible under ad(g), it is a direct sum of irreducible
ideals. WeP separate them as two groups: aPi of dimension 1 and bj of dimension > 1. It is
clear that j bj = [g, g] is semisimple and ai = Z(g).
The implications 3 =⇒ 4 are clear.
For implication 4 =⇒ 1, we notice that the assumption 4 implies that [g, Rad(g)] ⊂
[g, g] ∩ Rad(g) = 0. Thus Rad(g) ⊂ Z(g). Other direction is trivial.
Example 2.6.1. gln (C) = C + sln (C).
Let g be a Lie algebra. Define the following quotients:

gss := g/Rad(g), gab = g/[g, g], gred := g/([g, g] ∩ Rad(g)).

Corollary 2.6.8. The gred is the maximal reductive quotient of g. More precisely, the natural
map induces isomorphisms:
∼
gred −→ gss ⊕ gab .

One main motivation for introducing reductive Lie algebra is the following Proposition:

70
Proposition 2.6.9. Let g be a Lie algebra. Then every semisimple representation V of
g is induced from a semisimple representation of gred . Moreover if V is irreducible, then
V = U ⊠ W where U is an irreducible representation of gss and W is an character of gab .

Proof. By decomposing V into a direct sum of irreducible representations, we need only

consider the case V is irreducible and show that Rad(g))∩[g, g] acts trivially on V . Let (V, ρ)
be an irreducible representation of g. Then its restriction to Rad(g) is triangularizable. This
∼
means that there is an isomorphism V −→ Cn such that ρ(Rad(g)) ⊂ bn . It follows thasat
ρ[Rad(g), Rad(g)] ⊂ nn . Thus the subspace V 0 of vectors killed by ρ[Rad(g), Rad(g)] is non-
zero. Since [Rad(g), Rad(g)] is an ideal of g, we see that V 0 is invariant under g. It follows
that VL= V 0 . This shows in particular that V is the direct sum of weight spaces of Rad(g):
V = λ∈Λ Vλ indexed by a finite set Λ of Lie algebra homomorphisms λ : Rad(g) −→ C,
and Vλ is the subspace of V consisting of vectors v such that

Vλ := {v ∈ V : Y v = λ(Y )v, ∀Y ∈ Rad(g)}.

Now we claim that Λ is, in fact, a singlet. We need only show that for any λ ∈ Λ and
X ∈ g, XVλ ⊂ Vλ . Let v ∈ Vλ be a non-zero element and V ′ be the subspace of V generated
by X i v’s. Let d = dim V ′ . Then V ′ has base v, · · · X d−1 v. By formula (2.4.1), we see that
for any Y ∈ Rad(g), the action of Y on V ′ is triangularizable with eigen values λ(Y ) in the
k ′
P It follows that (Y − λ(Y )) = 0 on V for all k ≥ d . In particular, if we write
diagonal.
Xv = µ∈Λ vµ for vµ ∈ Vµ . Then for all k ≥ d,
X X
0 = (Y − λ)k Xv = (Y − λ)d vµ = (µ(Y ) − λ(Y ))k vµ = 0.
µ∈Λ µ∈Λ

This identity shows that for each µ with vµ ̸= 0, µ(Y ) − λ(Y ) = 0. Since this is true for all
Y , we must have µ = λ. Thus we have shown that Xv ∈ Vλ , and then that Λ is singleton.
Now for any Y ∈ Rad(g) ∩ [g, g], we have that

0 = tr(ρ(Y )) = nλ(Y ).

It follows that λ(Y ) = 0. Thus a acts trivially on V .

2.7 Exercises
Exercise 2.7.1. Let G be a connected Lie group. Then G is generated by any open neigh-
borhood U of e.
Exercise 2.7.2. Let G be a connected Lie group and H a discrete normal subgroup. Then H
is in the center of G. Moreover if Z(G) is discrete, then G/Z(G) has trivial center.
Exercise 2.7.3. Let Mn be the ring of n × n matrices in R or C, and GLn be the group
of invertible matrices in Mn . In this exercise, we want to define exponential map and its
inverse.

71
 qP
1. Let ∥ · ∥ : Mn −→ R be the norm function defined by ∥A∥ = 2
ij |Aij | . Show for all
A, B ∈ Mn ,
∥A + B∥ ≤ ∥A∥ + ∥B∥, ∥AB∥ ≤ ∥A∥∥B∥.

2. Define a map exp : Mn −→ Mn by

n
X Xk
exp(X) = , X ∈ Mn .
k=0
k!

Show that this series is absolutely convergent with limits in GLn (R), and defines a
smooth map of manifolds:
exp : Mn −→ GLn .

3. Let V be a neighborhood of In ∈ Mn defined by

V := {X ∈ Mn : ∥In − X∥ < 1}
and define a map log : V −→ Mn by
X (In − X)k
log X := − .
k=1
k

Show that V is included into GLn , and that log is absolutely convergent and defines a
smooth map log : V −→ Mn .
4. Let U = exp−1 V as an open neighborhood of 0 ∈ Mn . Show that exp defines a
∼
diffeomorphism U −→ V with inverse log.
Exercise 2.7.4. Let G be a Lie subgroup of GLn (R) with Lie algebra g, then g as a the
tangent space of G at e is a subspace of gln = Mn (R) ot Mn (C).
1. If G = SLn (R), show that
g = sln := {x ∈ gln | tr(x) = 0}.

2. If G = On consists of the matrices X satisfying X tX = X tX = I, show that

g = on := {x ∈ gln | x + tx = 0}.

3. Compute the Lie algebras of SOn , Un , SUn , and Spn (R).

Exercise 2.7.5. Let A and B be two element in Mn . Show that
eA+B = lim (eA/n eB/n )n ,
n→∞
2
e[A,B] = lim (eA/n eB/n e−A/n e−B/n )n .
n→∞
Hint: take logarithms.

72
Exercise 2.7.6. Let G be a Lie group with Lie algebra g. Let X, Y ∈ g. Then ad(X)(Y ) =
[X, Y ] = 0 if and only if for any s, t ∈ R,

esX etY = etY esX .

Hint: Rewrite the last equality formula as etY = esX etY e−sX and apply functoriality in
Remark 2.2.2.
Exercise 2.7.7. Let G be a Lie group with Lie algebra g. Let X, Y ∈ g.

1. The [X, Y ] = 0 if and only if esX etY = etY esX for all s, t ∈ R;

2. eX+Y = limn→∞ (eX/n eY /n )n .

2
3. e[X,Y ] = limn→∞ (eX/n eY /n e−X/n e−Y /n )n .

Exercise 2.7.8. Let Homcont (R, G) be the set of continuous homomorphisms from R to G.
Then the correspondence X 7→ exp(tX) defines a bijiection:
∼
g −→ Homcont (R, G).

Hint: For inverse map, using logarithm and continuity to construct bijections
log
∼ ∼
Homcont (R, G) −→ [Link] (R, g) −→ Homgroup (Q, g) = g.

Exercise 2.7.9. Let ϕ : G → G′ be a morphism of connected Lie groups. Show the following:

1. If dϕ : Lie(G) → Lie(G′ ) is surjective, then ϕ is surjective.

2. If dϕ : Lie(G) → Lie(G′ ) is bijective, then ϕ is a covering map.

Exercise 2.7.10. Show that sln (K) is simple for any n ≥ 2 by the following steps

1. Let Eij ∈ Mn (K) be the matrixes with entry 1 at position (i, j) and 0 elsewhere. Show
that
[Eij , Ekℓ ] = Eiℓ δjk − Ekj δℓi .

2. Let h be the
P subalgebraP of diagonal matrices. Thus each element of h can be written
as ha := i ai Eii with ai = 0. Show that under action of ad : h −→ gl(sln (K)),
sln (K) is the direct sum of eigenspaces of h of eigen character λ0 (ha ) = 0, and KEij
for eigen character λi,j (ha ) = ai − aj .

3. Show that the sum of ad(Eij ) gives an injection:

⊕i̸=j ad(Eij ) : h −→ ⊕i̸=j KEij .

73
 4. Let a be an ideal of sln (K). Then under ad(h), it can be written as a direct sums of
eigen spaces: M
a = a0 ⊕ aij
i̸=j

with a0 ⊂ h, and aij ⊂ KEij for some i ̸= j.

5. Show that all aij ̸= 0 for i ̸= j.

6. Show that for all i ̸= j, Eii − Ejj ∈ a0 . Thus a = sln (K).

Exercise 2.7.11. Let g be a Lie algebra with radical Rad(g). Then [Rad(g), Rad(g)] is nilpo-
tent.
Hint: let h = [Rad(g), Rad(g)]. Show that adh ⊂ gl(g) is nilpotent and then use exact
sequence
0 −→ Z(g) ∩ h −→ h −→ adh −→ 0.
Exercise 2.7.12 (Nilpotent radical). Let g be a finite dimensional Lie algebra.

1. Let a and b be ideals of g. Show by induction the following inclusion relation for lower
series: X
(a + b)i ⊂ ai + bi + aj ∩ bk .
j+k=i−1

2. Show that the set of nilpotent ideals has a maximal element. We call this element the
nilpotent radical of g and denoted it as Nil(g).

3. Nil(g) = 0 if and only if Rad(g) = 0. More precisely, if Rad(g) ̸= 0, then there is a k

such that 0 ̸= Rad(g)k ⊂ Nil(g).

Exercise 2.7.13. Let g be a nilpotent Lie algebra with center Z(g). Let h ⊂ g be an ideal.
Show that h ∩ Z(g) ̸= 0.
Hint: Consider the ideals h ∩ gi ’s, where gi are lower series of g.
Exercise 2.7.14 (Symmetric matrices). Let Hn be the manifold of positive definite n × n
symmetric matrices. Define a map π : GLn (R) −→ Hn by g · g T . Show that this map induces
a diffeomorphism of manifolds
∼
GLn (R)/On −→ Hn .
Can you construct similar quotients for GLn (C)/Un , GL2n (R)/Sp2n (R) by replacing g 7→ g T
by some other involutions?
Exercise 2.7.15. Let g ⊂ sl(V ) be a Lie subalgebra. Let h = Nsl(V ) (g) be the normalizer of
g in sl(V ). Let X ∈ g be an element with the Jordan–Chevalley decomposition in sl(V ):

X = Xs + Xn .

74
 1. Show that Xs , Xn ∈ h by using the Lemma A.1.5:
adgl(V ) (X) = adgl(V ) (Xs ) + adgl(V ) (Xn )
is the Jordan–Chevalley decomposition of adgl(V ) (X) in End(gl(V )).
2. Let ρ : h −→ gl(g) be the induced representation ρ(A)(B) = [A, B]. Show that
ρ(X) = ρ(Xs ) + ρ(Xn )
is the Jordan–Chevelley decomposition of adg (X) = ρ(X).
Exercise 2.7.16. Let g be a simple Lie algebra and ρ : g −→ gl(V ) be an irreducible repre-
sentation. Let h = Nsl(V ) (ρ(g)) be the normalizer of ρ(g) in sl(V ). Show ρ(g) = h by the
following steps.
1. Show that with respect to the Killing form Bh , there is an orthogonal decomposition
h = ρ(g) + a
with a an ideal of h.
2. Show that a = 0 using Schur’s lemma.
Exercise 2.7.17. With notation as in the previous example but with R replaced by C.
1. show that for every F ∈ Fn1 ,··· ,nr (C) there is a g ∈ GLn (C) such that F = gF 0 ;
2. computing the stabilizer Pn1 ,··· ,nr (C);
3. gives a description of Ed1 ,··· ,dr (C) in terms of group Un ;
4. give a description of Pn−1 (C) in terms of sphere S 2n−1 .

3 Representions of semisimple Lie algebras

This section assumes that our base field is C. But many results work for any field of
characteristic 0.

3.1 Basic definitions and examples

Definition 3.1.1. Let g be a Lie algebra.
1. A representation of a g on a finite dimensional vector space V over k is Lie homomor-
phism
ρ : g −→ gl(V ).
This can also be interpreted as a g-module structure on V :
g × V −→ V : (X, v) 7→ ρ(X)v.

75
 2. A homomorphism φ : V −→ W of two g-modules is a linear map compatible with
action by g:
φ(Xv) = Xφ(v).
Let Homg (V, W ) denote the space of g-morphisms from V to W .

Let Rep(g) denote the category of representations of g.

Example 3.1.1. The most basic representation of Lie algebra is the adjoint representation

ad : g −→ gl)(g), ad(X)(Y ) = [X, Y ].

Example 3.1.2. As in the case of group representation, we can construct from from old
representations V, W by constructing some new representations:

1. V ⊕ W , V ⊗ W , Hom(V, W ), where the action of g on the last two cases are given by

X(v ⊗ w) = Xv ⊗ w + v ⊗ Xw, ∀v ⊗ w ∈ V ⊗ W.

X(φ) = X ◦ φ − φ ◦ X, ∀φ ∈ Hom(V, W ).

2. Ker(f ), Im(f ), Coker(f ) for each f ∈ Homg (V, W ).

3. ∧i V , Symi V .

Example 3.1.3. For any Lie algebra g, there is a simply connected Lie group G with Lie(G) =
g. For any representation ρ : G −→ GL(V ), we obtain a representation dρ : g −→ gl(V ).
Conversely given a representation π : g −→ gl(V ), we can construct a representation ρ :
G −→ GL(V ) such that ρ(eX ) = eπX . The correspondence ρ −→ dρ gives an equivalence
between the categories of representations of G and g.
If H is another connected Lie group with Lie algebra g, then there is a covering G −→ H
with center Z a discrete abelian group. We have the following functors:
∼
Rep(H) −→ Rep(g) ←− Rep(G).

The image of the composition map is the representation of G, which is trivial over Z.
In the last section, we have show that every Lie algebra g is an extension of a semiple
Lie algebra by a solvalble Lie algebra:

0 −→ Rad(g) −→ g −→ g/Rad(g) −→ 0.

We have also shown that all representations of solvable Lie algebra are triangularizable. This
section will focus on the representation theory of semisimple Lie algebra over C. The first
example in our mind is the representations on the spaces of homogenous functions:

76
Example 3.1.4 (Representation Vm (Cn )). As one example, we want to calculate the represen-
tation of gln on the space Vm (Cn ) of homogenous functions on Cn of the homogenous function
of degree m. For convenience, we view Cn as row vectors and define the representation of
GLn on Vm (Cn ) is given by the right matrix multiplciation:
g · f (x) := f (xg).
It follows that the action of gln on Vm (Cn ) is given by
d d
A · f := |t=0 f (xetA ) = |t=0 f (x + txA)) = ∂xA f (x).
dt dt
If A = (aij ), then XX X
∂xA = ( xi aij )∂xj = xi aij ∂xj .
j i ij

Irreducible representations of sl2 (C).

In the following, we will show that all irreducible representations of sl2 (C) are given by
Vn (C2 ), the space of homogenous functions on C2 of degree n. Recall that sl2 (C) is generated
by the following three base
     
1 0 0 1 0 0
H= , E= , F =
0 −1 0 0 1 0
with relation
[H, E] = 2E, [H, F ] = −2F, [E, F ] = H.
For any representation V of sl2 (C), then we have a subspace V of eigenspaces of H:
M
V′ = Vλ
λ∈C

where Vλ = {v ∈ V : Hv = λv}. It is easy to check that the operators E and F shift these
spaces:
E : Vλ −→ Vλ+2 , F : Vλ −→ Vλ−2 .
This shows that V ′ is a subrepresentation of V .
Now we assume that V is irreducible. Then V = V ′ and there is a λ ∈ C such that Vλ ̸= 0
but Vλ+2 = 0. Pick one nonzero v0 ∈ Vλ . Let vi = F i v0 . Since V is finite dimensional, there
is a least n ∈ N such that vn ̸= 0 but vn+1 = 0. Then for each i we have Hvi = (λ − 2i)vi .
We want to calculate Evi for i ≥ 1:
Evi = EF i v0 = [E, F ]F i−1 v0 + F EF i−1 v0 = (λ − 2(i − 1))vi−1 + F Evi−1 .
By induction, this gives
i−1
X
Evi = (λ − 2j)vi−1 = i(λ − i + 1)vi−1 .
j=0

77
Taking i = n + 1, we get λ = n. Thus have shown that V has base v0 , · · · , vn over which
the action of H, E, F are given by

Hvi = (n − 2i)vi , Evi = i(n − i + 1)vi−1 , F vi = vi+1

where we treat v−1 = vn+1 = 0. Conversely above formula gives a representation of sl2 (C).
Theorem 3.1.2. The set of isomorphism classes of irreducible representations are given by
the action of sl2 (C) on the space Vn (C2 ) ⊂ C[x, y] of homogenous functions of degree n:

∂E = x∂y , ∂F = y∂x , ∂H = x∂x − y∂y .

Proof. We have shown that for each n, up to isomorphism, there is precisely one irreducible
representation of dimension n + 1. Moreover, the representation is given by the above
formula. We need only show that Vn (C2 ) has the same formula. Now Vn (C2 ) has the base
xi y j (i + j = n) with action by

∂H (xi y j ) = (i − j)xi y j , ∂E (xi y j ) = jxi+1 y j−1 , ∂F (xi y j ) = ixi−1 y j+1 .

Thus we can set

n!
v0 = xn , vi = ∂Fi v0 = xn−i y i .
(n − i)!
Then we obtain the same formula as above.
Corollary 3.1.3. The set of irreducible representations of SL2 (C) is given by Vn (C2 ).
Remark 3.1.1. In Exercise 1.7.22, we have also shown that all reprepresentations of SU(2)
is also given by Vn (C2 ). This is a not coincidence. In fact since both SU(2) and SL2 (C) are
simply connected, we have the following bijections:
∼ ∼ ∼
Rep(SU(2)) −→ Rep(su(2)) −→ Rep(sl2 ) ←− SL2 (C).

The middle isomorphism is induced by following isomorphism:

∼
su(2) ⊗ C −→ sl2 .

3.2 Decomposition of representations

Casimir element
Let g be a semisimple Lie algebra. In this subsection, we want to show that every represen-
tation V of g is completely reducible. The primary tool is to construct many elements in
Endg (V ) using Casimir element Cg ∈ ⊗2 (g), the dual element of the Killing form Bg .
To define this element we let X1 , · · · , Xn be a base of g and Y1 , Y2 , · · · , Yn be its dual
base with respect to the Killing form

Bg (Xi , Yj ) = δij .

78
Definition 3.2.1. The Casimir element Cg of g is defined to be
X
Cg := Xi ⊗ Yi ∈ ⊗2 g.
i

Proposition 3.2.2. The Casimir element is symmetric, g-invariant, and independent of the
choicePof basis. Moreover if g = ⊕i hi is an decomposition in to a direct sum of ideals, then
Cg = Chi .

Proof. The second part is clear by choosing an appropriate base for g. For the first part, let us
first make a connection between Bg and Cg . The bilinear pairing Bg defines an isomorphism
α : g −→ g∗ of g-modules by

α(X)(Y ) = Bg (X, Y ), X, Y ∈ g.

Doubling α, we obtain an isomorphism β : g ⊗ g −→ g∗ ⊗ g∗ of g-modules by

(X ⊗ Y, β(X ′ ⊗ Y ′ )) = B(X, X ′ )B(Y, Y ′ ), X, X ′ , Y, Y ′ ∈ g.

Now we view Bg as an element (g ⊗ g)∗ = g∗ ⊗ g∗ , and claim that

β(Cg ) = Bg .

Indeed, for any X, Y ∈ g,

X
(X ⊗ Y, βCg ) = B(X, Xi )B(Y, Yi ).
i
P P
Write X = i ai Y i , Y = j bj X i , then we have
X
(X ⊗ Y, βCg ) = ai bi = B(X, Y ).
i

Other properties follows from the corresponding property for Bg .

Now we want to apply to Cg to representations.

Corollary 3.2.3. Let ρ : g −→ gl(V ) be a representation of semisimple Lie algebra and h

an ideal of g, then ρ(Ch ) commutes with ρ(g). Thus it defines an element in Endg (V ).

For any representation ρ : g −→ gl(V ), we can define Killing form

Bρ : g × g −→ C, Bρ (X, Y ) = tr(ρ(X)ρ(Y )).

The following lemma gives a simple way to compute the Killing form and Casimir operator
when g is simple.

79
Proposition 3.2.4. Let g be a simple Lie algebra and ρ : g −→ gl(V ) a non-zero represen-
tation. Then Bρ is non-degenerate and proportional to Bg in the sense there is a nonzero
constant c such that Bρ = cBg . Moreover

tr(ρCg ) = c dim g.

Proof. It is easy to see that Bρ is invariant under g, and thus defines as an element in
Homg (g, g∗ ). We know this space is one dimensional and generated by Bg . Thus there is a
constant c ∈ C such that
Bρ = cBg .
We claim that c ̸= 0. Otherwise, Bρ = 0, and by Cartan’s criterion 2.5.6, Imρ is solvable,
thus zero.
In this way, we can calculate tr(ρ(Cg )) use the bases Xi , Yj of g:
X X X
tr(ρCg ) = tr(ρ(Xi )ρ(Yi )) = Bρ (Xi , Yi ) = c B(Xi , Yi ) = c dim g ̸= 0.
i i i

Example 3.2.1 (Killing form for sln ). We want to illustrate how to easily compute the Killing
form for sln (C) using the standard representation std : g −→ gln (C) instead of adjoint
representation. The Killing form for identity representation of gln is given by
(
1, (i, j) = (ℓ, k),
Bstd (Eij , Ekℓ ) = tr(Eij Ekℓ ) =
0, otherwise.

For sln , we need only write down this form for the diagonal matrices
X X X
Bstd ( xi Eii , yj Ejj ) = xi y j .
i j i

We need P
to compute the constant c in the formula Bg = cBstd . For this we apply diagonal
matrices i xi Eii which have eigencalues xi − xj on Eij ’s. Thus
X X X X
Bg ( xi Eii , yj Ejj ) = (xi − xj )(yi − yj ) = 2n x i yj .
i j i,j i

Thus we have Bg = 2nBstd . If we take a base for sln by Hi = Eii − Ei+1,i+1 for i = 1, · · · n − 1
and Eij for i ̸= j. Then Eij ’s are orthogonal to Hi and

4n
 i=j
Bg (Hi , Hj ) = −2n |i − j| = 1

0 |i − j| ≥ 2.


80
Example 3.2.2 (Killing form for sp2n ). Now we want to compute the Killing form for sp2n .
Recall that sp2n is the subalgebra of sl2n of matrices X satisfying the equation
 
t 0 1n
XJ + J X = 0, J := .
−1n 0
 
A B
Let us write X = with entries in Mn (C). Then the abvoe equation is equivalent
C D
to the following equations:

D = −tA, B = tB, C =t C.

In this way, we have base for sp2n given by

Xij = Ei,j − En+j,n+i , 1 ≤ i ≤ n, 1≤j≤n

Yij = Ei,n+j + Ej,n+i , Zij = En+j,i + En+i,j , 1≤i≤j≤n

We can compute the non-zero values of the Killings forms among these as follows:

Bstd (Xij , Xji ) = 2,

Bstd (Yij , Zij ) = 2.
P
To compute the Bg , we consider the diagonal subalgebra h = i CXii . Then P the above
base consists of eigen vectors for representation ad. More precisely, for H = xk Xk,k , we
have

ad(H)Xij = (xi − xj )Xij , ad(H)Yij = (xi + xj )Yij , ad(H)Zij = −(xi + xj )Zij .

Thus we have
X X X X
Bg ( xi Xii , yj Xjj ) = (xi − xj )(yi − yj ) + 2 (xi + xj )(yi + yj )
ij i≤j
X X
= [(xi − xj )(yi − yj ) + (xi + xj )(yi + yj )] + 4 xi yi .
ij i
X X X
= (2xi yi + 2xj yj ) + 4 xi yi = 4(n + 1) xi y i .
ij i i

Now we compare this with

X X X
Bstd ( xi Xii , yj Xjj ) = 2 xi yi
i

to get
Bg = 2(n + 1)Bstd .
Notice that Xii ’s is an orthogonal base with norm 4(n + 1).

81
Decomposition of representations
We start with Schur’s lemma, whose proof is the same as in the case of group representations:
Theorem 3.2.3 (Schur’s lemma). Let V and W be irreducible finite dimensional represen-
tations of a Lie algebra g, then
• If V is not isomorphic to W then Homg (V, W ) = 0.
• If V ∼
= W , then
dimC Homg (V, W ) = 1.
In particular Endg (V ) = C.
Theorem 3.2.5. Let ρ : g −→ gl(V ) be a representation of a semsimple Lie algebra. Then
ρ is completely reducible. In other words, for any subrepresentation W ⊂ V , there is a
subrepresentation W ′ ⊂ V such that W + V ′ = V and W ∩ V ′ = {0}.
Proof. We will prove the theorem by several reduction steps. First of all, let U = V /W ,
then we have an exact sequence
α β
0 −→ W −→ V −→ U −→ 0.

We need to find a g-homomorphism s : U −→ V of g-modules so that β ◦ s = idU . There is

nothing to prove if U = 0. So we assume that dim U > 0.
∼
Step 1: Reduce the proof to the case U −→ C. We form a new exact sequence of
g-modules:
α∗ β∗
0 −→ Hom(U, W ) −→ Hom(U, V ) −→ Hom(U, U ) −→ 0
where α∗ (φ) = α ◦ φ and β∗ (φ) = β ◦ φ are compositions of operators. Then the question
becomes lifting the identity element idU to an element s ∈ Homg (U, V ). Thus we define new
g-modules W ′ = Hom(U, W ), V ′ = β∗−1 (CidU ), and U ′ = C, to get a new exact sequence of
g-modules:
α′ β′
0 −→ W ′ −→ V ′ −→ C −→ 0.
Step 2: reduce to the case where W is irreducible by induction on dim W . If W is not
irreducible, let W ′ be a non-zero irreducible submodule of W . Then we have a new exact
sequence:
0 −→ W/W ′ −→ V /W ′ −→ C −→ 0.
By induction, we can lift C to a g-submodule V ′ /W ′ of V /W ′ of dimension 1. Now we have
an exact sequence
0 −→ W ′ −→ V ′ −→ C −→ 0
with W ′ irreducible.
Step 3: prove the theorem for W irreducible and U = C. Let h be an ideal of g such that
ρ(h) ̸= 0. Then we have an element

ρ(Ch ) ∈ Endg (V ).

82
Then by Proposition 3.2.4, trρ(Ch ) ̸= 0. On other hand, this element maps the exact
sequence
0 −→ W −→ V −→ C −→ 0.
Thus
trCh |V = trCh |W + trCh |C .
As ρ(Ch ) = 0 on C. Thus trCh |W ̸= 0. Since W is irreducible, by Schur’s lemma, ρ(Ch ) is
given by a scale map αIdW on W . This gives a decomposition

V = W ⊕ Ker(ρ(Ch )).

We may apply the proof of the canonical decomposition Theorem 1.1.16 to semisimple
Lie algebra to get:
Theorem 3.2.6. Let V be a finite-dimensional representation of a semisimple Lie algebra
g. Then there is a canonical isomorph:
∼
M
Homg (W, V ) ⊗C W −→ V,
W ∈b
g

where b
g is the set of isomorphism classes of irreducible representations of g.
The following theorem explains the name “semisimple algebra”.
Theorem 3.2.7. Let g be a Lie algebra. Then g is semsimple if and only if every represen-
tation ρ is semisimple, i.e., ρ is a direct sum of irreducible representations.
Proof. The “only if” part is the previous theorem. Now assume that every representation ρ
of g is semisimple. Take ρ = ad. Then g is the direct sum of minimal ideals hi : g = ri=1 hi .
L
We decompose g into the sum two parts: the sum g1 of ideals hi with dimension > 1, and the
sum h2 of hi with dimension 1. Then it is clear that g1 is semisimple, and g2 is commutative.
∼
If g2 ̸= 0, then it has some non-semisimple representation V . The projection g/g1 −→ g2
makes V a non-semisimple representations of g. Thus a contradiction. Thus g = g1 is
semisimple.

Jordan–Chevalley decomposition
Let V be a finite dimensional vector space over C and let X ∈ End(V ). By Lemma A.1.3,
there is a unique decomposition
X = X s + Xn
with Xs semisimple (i.e., Xs is diagonalizable), and Xn nilpotent (i.e., Xnk = 0 for some
k > 0), and Xs Xn = Xn Xs . Moreover, there are polynomials P (T ), Q(T ) ∈ C[T ] with
vanishing constant term such that

Xs = P (X), Xn = Q(X).

83
 Now let g be a semisimple Lie algebra. Then we have a representation

ad : g −→ gl(g).

This is in fact injective. Thus we can the following definition:

Definition 3.2.8. An element X ∈ g is called semisimple (resp. nilpotent) if ad(X) is

semisimple (resp. nilpotent).

Theorem 3.2.9 ( Lemma A.1.3, Corollary 2.6.5). Let g be a semisimple Lie algebra. Then
for any element X ∈ g, there is a unique decomposition

X = X s + Xn

with Xs semisimple, Xn nilpotent, and [Xs , Xn ] = 0. Moreover, for any homomorphism

φ : g −→ h to a semisimple Lie algebra,

φ(X) = φ(Xs ) + φ(Xn )

is the Jordan–Chevalley decomposition in h.

Cartan subalgebras
Definition 3.2.10. Let g be a semisimple Lie algebra over C and h a commutative subalgebra.

1. h is called toral if every elements in h is semisimple.

2. h is called a Cartan subalgebra if h is a maximal toral subalgebra of g.

Example 3.2.4. In Exercise 2.7.10, we have shown that sln (C) is a simple algebra. It is clear
that the diagonal algebra h is toral: for any H ∈ h with eigenvalues λi for the standard basis:
Hei = λi . Then ad(H) is semisimple with eigenvalues λi − λj for the base Eij for i ̸= j and
0 on h. We want to show that h is a Cartan subalgebra: for any X ∈ sln (C) commuting
with h, write X
X = X0 + aij Eij , X0 ∈ h, aij ∈ C.
ı̸=j
P
Then for any H ∈ h and any (λ1 , · · · , λn ) with i λi = 0,
X
0 = [H, X] = aij (λi − λj )Eij .
i̸=j

This implies that aij = 0 for all i ̸= j. Thus X = X0 ∈ h.

Example
P 3.2.5. Similarly, we can show that sp2n is a simple Lie algebra with Cartan subal-
gebra i CXii as in Example 3.2.2.

84
Weight decompositions
Let g be a semisimple Lie algeba and h a toral subalgebra with linear dual h∗ = Hom(h, C).
Let ρ : g −→ gl(V ) be a representation. Since h is commutative with every elements
semisimple, ρ(h) is diagonalizable. So we can write
M
V = Vλ , Vλ = {v ∈ V : H · v = λ(H)v, ∀H ∈ h} .
λ∈h∗

Definition 3.2.11. Let λ ∈ h∗ . If Vλ ̸= 0, then call it a weight space and λ a weight of h

on V . Let Λ(ρ, h) denote the set of weights.
n
Example 3.2.6. Consider
P the standard representation of sln on C . Then the standard base
n
decomposition C = i Cei is in fact a weight decomposition with weight λi given by
X
λi ( xj Ejj ) = xi .

More generally, for a wedge product ∧m (Cn ), the weight decomposition is the standard
decomposition X
∧m (Cn ) = Cei1 ∧ ei2 · · · eim
i1 <i2 ···im

with weights
λi1 + λi2 + · · · λim .
We will show late that these representations are irreducible. Notice that ∧m (Cn ) and
∧n−m (Cn ) are dual to each other, see Remark A.1.2.
Example 3.2.7. Similarly, we may considerP the standard represenation of sp2n on C2n . Then
the standard base decomposition C = 2n2n
i=1 Cei is in fact a weight decomposition with
weight λi given by (
n
X xi i≤n
λi ( xj Xjj ) =
j=1
−xi−n i > n.
More generally,
Vm 2nfor each m between 0 and 2n, we have a represenation on the m-th wedge
product (C ), the weight decomposition is the standard decomposition
X
∧m (C2n ) = Cei1 ∧ ei2 · · · eim
i1 <i2 ···im

with weights
λi1 + λi2 + · · · λim .
The matrix J defines an invariant bilinear form ψ on C2n by ψ(u, v) = uJ tv, and thus a self-
dulity of the standard representation std of sp2n . By Remark A.1.2 and A.1.3, this defines
an isomorphism of sp2n -modules:
∼
αm : ∧m (C2n ) −→ ∧2n−m (C2n ).

85
Remark 3.2.1. The represention ∧m (C2n ) is not irreducible in general. To decompose it as a
direct sum L
of irreducible representation, we define a new represenation sl2 on the their sum
∗
∧ (C ) = m ∧m (C2n ) as follows:
2n

X
E(v1 ∧ v2 · · · vm ) = (−1)i+j−1 (vi , vj )v1 ∧ · · · ∧ vbi ∧ · · · ∧ vbj · · · ∧ vm ,
i<j

n
X
F (v1 ∧ v2 · · · vm ) = ( ei ∧ ei+n ) ∧ v1 ∧ v2 · · · vm ,
i=1

H(v1 ∧ v2 · · · vm ) = n − m.
These three operators are communicating with the action of sp2n .
To show that that is in fact an action by sl2 , we decompose C2n as the direct sum of Vi
(i = 1, · · · , n) with Vi = Cei ⊕ Cei+n . It follows that
n
O
∗ 2n
∧ (C ) = ∧∗ (V i )
i=1

and that H, E, F come from the corresponding action on ∧∗ V i . Thus we need only check
if the above formula define an action of sl2 in the case n = 1. In this case, we have a
decomposition:
∧∗ C2 = (Ce1 ⊕ e2 ) ⊕ (C ⊕ Ce1 ∧ e2 ),
where E, F, H act trivialy on the first summand, and act like standard representation of sl2
on the second command. Thus we have construct a representation sp2n ⊕ sl2 on ∧∗ (C2n ).
Now we want to decompose ∧∗ (C2n ) into a direct sum of irreducible prepresentations of
sp2n ⊕ sl2 . We start with the following canonical decompositiion for a general representation
V of sl2 :
M∞
V = Wk ⊗ Symk (C2 ), Wk = Homsl2 (Symk (C2 ), V ).
k=0

In order to compute Wk practically, we compute Ker(E):

∞
M
Ker(E) = Wk ⊗ Ker(E|Symk (C2 )).
k=0

From Theorem 3.1.2, we know that Ker(E|Symk (C2 )) is one dimensional with weight k. Thus
the above decomposition is also a weight decomposition of Ker(E), thus we have another
canonical decomposition
∞
∼
M
V −→ Ker(E)k ⊗ Symk (C2 ).
k=0

Now we apply the above formula for V = ∧∗ (C2n ). In this case,

Ker(E)k = Ker(E| ∧n−k (V 2n ), 0 ≤ k ≤ n.

86
These are actually irreducible representations of sp2n . Thus, we obtain a canonical decom-
position into irreducible representations of sp2n ⊕ sl2 :
n
M
∗ 2n
∧ (C ) = Ker(E| ∧m (C2n )) ⊠ Symn−m (C2 ).
m=0

It follows that for any m, we have decompostion of irreducible representations for sp2n :
X M
∧m (C2n ) = F t Ker(E| ∧m−2t (C2n )) = Ker(E| ∧m−2t (C2n )).
(m−n)/2≤t≤m/2 (m−n)/2≤t≤m/2

3.3 Root space decompositions

Root space decompositions
This subsection will fix a semisimple Lie algebra g and a Cartan subalgebra h. Now we apply
the weight decomposition to ad : g −→ gl(g). It is clear that g0 = Zg (h) the centralizer of h.
Let Φ the set of nonzero α ∈ h∗ such that gα ̸= 0. We call Φ the set of roots of g relative to
h. Thus we have a decomposition:
M
g = Zg (h) ⊕ gα .
α∈Φ

It is clear that Φ generates h∗ : the kernel of all α ∈ Φ will be the center of g, zero!
P
Example 3.3.1. For g = sln with diagonal subalgebra h and weights λi ( xi Eii ) = xi , then
the decomposition is given by M
g=h⊕ CEij .
i̸=j

By Example 3.2.1, the set of roots consists of λi − λj on Eij .

P
Example 3.3.2. For g = sp2n with diagonal subalgebra h and weights λi ( xi Xii ) = xi , then
the root decomposition is given by
M M M
g=h⊕ CXij ⊕ CYij ⊕ CZij .
i̸=j i≤j i≤j

By Example 3.2.2, the set Φ of roots consisting of λi − λj on Xij , λi + λj on Yij and −λi − λj
on Zij .

Lemma 3.3.1. Let ρ : g −→ gl(V ) be a representation with weight decomposition

M
V = Vλ .
λ∈Λ

Let α, λ ∈ h∗ .

87
 1. ρ(gα )Vλ ⊂ Vλ+α ;

2. if α ̸= 0, every element in ρ(gα ) is nilpotent.

Proof. For the first one, let H ∈ h, X ∈ gα , and v ∈ Vλ we have

HXv = [H, X]v + XHv = α(H)Xv + λ(H)Xv = (λ + α)X(v).

For the second one, for any k ≥ 0,

ρ(gα )k (Vα ) ⊂ Vλ+kα .

As the set Λ of weights is finite, for k >> 0, Vλ+kα = 0 for all λ ∈ Λ(V, h). Thus ρ(gα )k =
0.

Lemma 3.3.2. Let α, β ∈ h∗ .

1. [gα , gβ ] ⊂ gα+β ;

2. if α ̸= 0, every element in gα is nilpotent;

3. if α + β ̸= 0, then Bg (gα , gβ ) = 0;

4. the restriction of the form Bg : gα × g−α −→ C is non-degenerate;

5. if α ∈ Φ, then −α ∈ Φ.

Moreover, h∗ is generated by Φ.

Proof. The first two parts follows from previous lemma to representation ad.
The third part follows from invariance of B under ad(h): for X ∈ gα , Y ∈ gβ , H ∈ h,

0 = B([H, X], Y ) + B(X, [H, Y ]) = (α(H) + β(H))B(X, Y ).

The last two parts follow part 3 and the nondegeneracy of Bg .

One main result of this subsection is following:

Theorem 3.3.3. Let g be a semisimple Lie algebra and h is a Cartan subalgebra. Then

h = Zg (h) = Ng (h).

Thus we have a root space decomposition:

M
g=h⊕ gα .
α∈Φ

88
Proof. We will prove the theorem in several steps by carefully analyzing the Lie algebra
Zg (h). Let n is the subset of nilpotent elements of g included into Zg (h).
Step 1: The Zg (h) is a nilpotent subalgebra with a global Jordan–Checalley decomposition

Zg (h) = h + n.

Indeed for any X ∈ Zg (h) with Jordan–Chavalley decomposition X = Xs + Xn in g.

Then it is clear that ad(Xs ), ad(Xn ) are both commuting with ad(h). Thus both are in
Zg (h). Since h is Cartan, Xs ∈ h. It is clear that adZg (h) (X) is nilpotent as both adZg (h) (Xs )
and adZg (h) (Xn ) are nilpotent.
Step 2: The n is an ideal of Zg (h) and the the decomposition Zg (h) = h+n is an orthogonal
decomposition with respective to Bg . Moreover Bg |h and Bg |n are both non-degenerate.
For any X ∈ h and Y ∈ n, it is clear that ad(X) · ad(Y ) is nilpotent in End(g). Thus

Bg (X, Y ) = tr(ad(X)ad(Y )) = 0.

Since Bg |Zg (h) is non-degenerate by the above lemma, so Bg |h is non-degenerate. It follows

that n is exactly the orthogonal complement of h in Zg (h). Thus n is an ideal of Zg (h) over
which Bg is non-degenerate.
Step 3: n = 0 and Zg (h) = h.
Otherwise Z(n) ̸= 0 since n is nilpotent. Thus for any X ∈ Z(n), Y ∈ n, ad(X)ad(Y ) is
nilpotent. Thus Bg (X, Y ) = 0. This contradicts to the non-degeneracy of Bg |n .
Step 4: Ng (h) = h. Let X ∈ Ng (h) be an element. Then write it in root decomposition
X
X = X0 + Xα , X0 ∈ h, Xα ∈ gα .
α∈Φ

Then for any H ∈ h, X

[H, X] = α(H)Xα ∈ h.
α∈Φ

Then must have all Xα = 0. Thus X = X0 ∈ h.

We use the Killing form to define a bijection t : h∗ −→ h: α 7→ tα by

α(H) = B(H, tα ), ∀H ∈ h.

Using this bijection, we can transform B to an inner product on h∗ :

(α, β) = B(tα , tβ ).

Our main purpose is to study the triple (h∗ , Φ, (·, ·)).

Example 3.3.3. For g = sln , by Example 3.2.1, the Killing form on h is given by
X X X
Bg ( xi Eii , yi Eii ) = 2n x i yi .
i i i

89
 xi λi ∈ h∗ with
P P
Thus for an α = i xi = 0,

1 X
tα = xi Eii .
2n i

Thus X X 1 X
( xi λi , yj λj ) = xi y i .
2n
Example 3.3.4. For g = sp2n , by Example 3.2.2, the Killing form on h is given by
X X X
Bg ( xi Xii , yi Xii ) = 4(n + 1) xi y i .
i i i

xi λi ∈ h∗ ,
P
Thus for an α = i

1 X
tα = xi Xii .
4(n + 1) i

Thus X X 1 X
( xi λi , yj λj ) = xi y i .
4(n + 1)

Standard sl2 -triples

We start with construction of some distinguished elements.

Lemma 3.3.4. Let α ∈ Φ.

1. For X ∈ gα , Y ∈ g−α , [X, Y ] = B(X, Y )tα .

2. α(tα ) = (α, α) ̸= 0; write Hα = 2tα /(α, α).

3. There are Eα ∈ gα , Fα ∈ g−α such that

[Hα , Eα ] = 2Eα , [Hα , Fα ] = −2Fα , [Eα , Fα ] = Hα .

Thus we have an sl2 (C)-triple sα := CEα + CFα + CHα .

Proof. The first one is equivalent to the following identity:

B(H, [X, Y ]) = B(X, Y )α(H).

This follows from the invariance of B: B(H, [X, Y ]) = B([H, X], Y ).

For the second one, consider the algebra

sα := gα + g−α + Ctα .

90
If α(tα ) = 0, then the above algebra is solvable with tα in the center. Thus by Cartan’s
criterion 2.5.6, the representation ad : sα −→ gl(g) is triangulizable with ad(tα ) nilpotent.
This contradict to the fact that tα ∈ g is semisimple.
For the third part, since B is non-degnerate on gα × g−α , we can find Eα ∈ gα , Fα ∈ g−α
such that
[Eα , Fα ] = Hα .

Corollary 3.3.5. Let V be a representation of g with weight decomposition with resepct to

h: M
V = Vλ .
λ

Then for each α ∈ Φ, λ(Hα ) ∈ Z.

Proof. Consider the V as a sα module and classfication of representations of sl2 .
Example 3.3.5. Take g = sln and h the diagonal subalgebra. For each pair i ̸= j, we have
sl2 (C) triple
Eij , Eji , Hij := [Eij , Eji ] = Eii − Ejj , i ̸= j.
Example 3.3.6. Take g = sp2n and h the diagonal subalgebra. Then we have sl2 (C) triples:

Xij , Xji , Hij := [Xij , Xji ] = Xii − Xjj , 1 ≤ i ̸= j ≤ n

Yij , Zij , Hij′ := [Yij , Zij ] = Xii + Xjj , 1≤i<j≤n

Yii , Zii /2, Hii′ := [Yii , Zii /2] = 2Xii , 1 ≤ i ≤ n.
Now fix one α, we consider the representation

ad : sα −→ gl(g).

For each β ∈ Φ, gβ is a weight space of Hα with eigenvalue β(Hα ). It has following invariant
subspaces:
M M
V (α) := CHα ⊕ gλα , V (β) := gβ+iα , β ̸= ±α.
c∈C× i∈C
cα∈Φ β+iα∈Φ

Lemma 3.3.6. The V (α) = sα and V (β) is irreducible.

Proof. The weight spaces of Hα are given as follows:

CHα = V (α)0 , gcα = V (α)2c , gβ+iα = V (β)β(Hα )+2i .

Let us decompose V (α) into a direct sum of irreducible presentations:

Vα = ⊕n≥0 mn ρn , ρn := Vn (C2 )

91
and mn is the multiplicity. Then we have
X
1 = dim V (α)0 = mn .
2|n

It follows that the sum of ρn for even n is fact irreducible, which must be sα . Thus we have
dim gα = 1. For the odd part we have that if cα ∈ Φ with c ̸= ±1, then 2c is an odd integer.
By changing the role of α and cα, we get that 2/c is an odd integer. This is impossible.
Thus we have shown that
V (α) = sα .
Now we consider the representation V (β). Since each gβ+iα is at most one dimensional with
different weight β(Hα ) + 2i but same parity, it is irreducible.

Theorem 3.3.7. Let α, β ∈ Φ. Then

1. dim gα = 1;

2. Φ ∩ Cα = {±α};

3. β(Hα ) ∈ Z;

4. β − iα ∈ Φ for any integer i between 0 and β(Hα );

5. if α + β ∈ Φ, then [gα , gβ ] = gα+β .

∼
Proof. The first three parts follow from the previous two lemmas. We assume V (β) −→
Vm (C2 ) for the last two parts. then β(Hα ) ∈ Z and any integer between β(Hα ) and −β(Hα )
with same parity as β(Hβ ) is a weight of V . These are exactly gβ−iα ’s for i between 0 and
β(Hα ). Moreover, if α + β ∈ Φ, then we have an isomorphism:

ad(Eα ) : gβ −→ gβ+α .

3.4 Root systems

Positivity
Now we let EQ be the Q-subspace of h∗ generated by Φ equipped with a non-degenerate and
symmetric product
EQ × EQ −→ C : (α, β) = B(tα , tβ )
as before.

Theorem 3.4.1. The space EQ has dimension dimQ EQ = dimC h∗ . Thus h∗ = EQ ⊗Q C.

Moreover the pairing (·, ·) is positive definite with values in Q.

92
Proof. We first prove that (·, ·) on EQ is positive definite with values in Q. Our main tool is
to use

1. the part 1 of Theorem 3.3.7: for any α, β ∈ Φ

(α, β)
∈ Q;
(α, α)

2. the formula: for any two element α, β ∈ h∗ ,

X X
(α, β) = B(tα , tβ ) = tr(ad(tα )ad(tβ )) = φ(tα )φ(tβ ) = (φ, α)(φ, β).
φ∈Φ φ∈Φ

From these facts, we have that for α, β ∈ Φ

(α, β)
∈ Q.
(α, α)(β, β)

This implies first (α, α) ∈ Q (by taking α = β), and then (α, β) ∈ Q. Thus (·, ·) is Q-valued
on EQ . To show it is positive definite use the formula again: for any nonzero α ∈ EQ :
X
(α, α) = (φ, α)2 > 0.
φ∈Φ

Now we want to show that dimQ EQ = dimC h∗ =: ℓ. Let ℓ = dim h and {α1 , · · · , αℓ } ⊂ Φ
be a base formed by elements in Φ. Then for any β ∈ Φ, we have a linear combination
ℓ
X
β= xi α i , xi ∈ C.
i=1

Now for any j ∈ {1, · · · , ℓ} we have

X
(β, αj ) = xi (αi , αj ).
i

Thus we can solve this equation in xi ’s by inverting the matrix (αi , αj ) with coefficient in
Q. In particular, xi ∈ Q. This shows that dim EQ = ℓ.
Let E = EQ ⊗ R. Then E is an Euclidean space with inner product (·, ·) which contains
the finite set Φ of non-zero vectors. The pair (E, Φ) forms a root system of g.

93
Root systems
By a root system, we mean a finite dimensional Euclidean space E with a finite set Φ call
the set of roots with following properties:

1. the roots generates E;

2. for α ∈ Φ, then Rα ∩ Φ = {α, −α};

3. for any α, β ∈ Φ,
2(α, β)
∈ Z.
(α, α)
2(α,β)
4. for any α, β ∈ Φ, and i between 0 and (α,α)
,

β − iα ∈ Φ.

Definition 3.4.2. Let E reg be the complement of E of union of hyperplanes (x, α) = 0 for
α ∈ Φ. Each connected component of E reg is called a Weyl chamber of (E, Φ).

Fix a Weyl chamber C of (E, Φ). Then we define a partial order on E by

λ ≤ µ ⇐⇒ (x, λ) ≤ (x, µ), ∀x ∈ C.

Let E + be the subset of E of elements x ≥ 0 which is the closed dual cone of C. Then C
is actually determined by E + : the closure C̄ is the dual of E + under the symmetric relation
(x, λ) ≥ 0.
Let Φ+ denote the intersection Φ ∩ E + and Φ− = {−α : α ∈ Φ+ }. We call Φ+ (resp. Φ− )
the set of positive (resp. negative) roots.
P
Example 3.4.1.
P In the case
P g = sln (C), we have that E is isomorphic to subspace of Rλi
of elements i xi λi with xi = 0 with the inner product
X X 1 X
( xi λ i , y j λj ) = xi y j ,
2n
reg
and ΦPis the set of roots λi − λj for all i ̸= j.. The set E of regular elements is given by
λ = i xi λi with xi ̸= xj for all i ̸= j. The Weyl chambers are indexed by permutations
σ ∈ Sn of [1, · · · , n]: for each σ ∈ Sn the corresponding Weyl chamber Cσ is the set of x ∈ E
such that
xσ(1) > xσ(2) > · · · > xσ(n) .
The corresponding set Φ+
σ of positive roots is given by λσ(i) − λσ(j) for i < j.

Example 3.4.2. In the case g = sp2n , we have that E is isomorphic ni=1 Rλi with inner
P
product
X X 1 X
( xi λ i , y j λj ) = xi yj ,
4(n + 1)

94
 P form ηi λi − ηj λj with ηi , ηj ∈ {±1}. The set
and Φ is given by nonzero elements of the
reg
E of regular elements is given by λ = xi λi such that xi ̸= ±xj for all i, j. The Weyl
chambers are indexed by elements in W := µn2 ⋊ Sn : for each element

w = (ϵ1 , · · · , ϵn , σ) ∈ W, ϵi = ±1, σ ∈ Sn ,

the corresponding Weyl chamber Cw is the set of x ∈ E such that

ϵ1 xσ(1) > ϵ2 xσ(2) > · · · > ϵn xσ(n) > 0.

The correspionding set Φ+

w of positive roots is given by

ϵi λσ(i) ± ϵj λσ(j) , i≤j

Lemma 3.4.3. The Φ+ generates E + with coefficients in R≥0 . Moreover Φ = Φ+ ∪ Φ− .

Proof. By definition, Φ+ is in fact determined by equation (x0 , α) > 0 for any single x0 ∈ C.
This implies that Φ is the disjoint union of Φ = Φ+ ∩ Φ− .
For the first part, let D = R≥0 Φ+ . The interior C ′ of its dual cone is defined by inequal-
ities: (x, α) > 0 for all x ∈ Φ+ . It is connected and contains only regular elements as Φ+
generates E. Also it is clear that C ′ ⊃ C. Thus C ′ = C. So D = E as their duals.

Simple roots
Let ∆ ⊂ Φ+ be a subset that can’t be written as a sum of two elements in ∆. We call ∆ a
set of simple roots in Φ+ .
Example 3.4.3. Take g = sln (C) with Weyl chamber Cσ corresponding to an σ ∈ Sn . Then
the corresponding set ∆σ of simple roots is given by

λσ(i) − λσ(i+1) .

Example 3.4.4. Take g = sp2n with Weyl chamber Cw corresponding to an w = (ϵi , σ) ∈ W .

Then the corresponding set ∆σ of simple roots is given by

λσ(i) − λσ(i+1) , 2λσ(n) , (i < n)

Theorem 3.4.4. The ∆ is a base of E. Moreover each α ∈ Φ+ (resp. α ∈ Φ− ) is a linear

combination of of ∆ with coefficients in Z≥0 (resp. Z≤0 ).

Proof. Let’s show that ∆ generates Φ+ with coefficients in Z≥0 : Φ+ ⊂ Z≥0 ∆. Let x0 ∈ C
be any element which defines a function Φ+ −→ R+ by α 7→ (x0 , α). If the set of elements
of Φ+ \ Z≥0 ∆ is not empty, then we have a α in this set with minimal (x0 , λ). Then
α = α1 + α2 . This implies that (x0 , α1 ) and (x0 , α2 ) are both smaller that (x0 , α). Thus both
α1 , α2 ∈ Z≥0 ∆. This implies that α ∈ Z≥0 ∆. A contradiction.
By multiplying everything by −1, it follows that ∆ generates Φ− with coefficients in Z≤0 .

95
 We want to show that ∆ is a base for E. Since Φ generates E, the above shows that ∆
generates E. Thus it suffices to show that ∆ is linearly independent. Assume in contrary
then we will have an equation of the form
X X
aα α = bβ β
α∈∆1 β∈∆2

where ∆1 , ∆2 are disjoint non-empty subset of ∆ and aα , bα > 0. Then we have

X X X
0<( aα α, bβ β) = aα bβ (α, β).
α∈∆1 β∈∆2 α,β

This is impossible by the following lemma.

Lemma 3.4.5. Let ∆ be the set of simple roots as above. Then for any α, β ∈ ∆, (α, β) < 0

Proof. Assume in contrary that (α, β) > 0. Then by Property (4), β − α ∈ Φ. Thus either
β − α ∈ Φ+ or α − β ∈ Φ− . Both of them are contradicting with the definition of ∆.

Corollary 3.4.6. The ∆ generates E + with coefficients in R≥0 . Moreover C is the open
dual cone of ∆ determined by inequality (x, α) > 0 for x ∈ C and α ∈ ∆.

Definition 3.4.7. The Weyl group W of (E, Φ) is the subgroup of O(E) generated by re-
flections on E defined by roots α ∈ ∆:

2(λ, α)
σα : λ 7→ λ − α.
(α, α)

By assumption, σα stabilizes the subset Φ. Thus W is a finite group.

Example 3.4.5. Take example
P g = sln (C) with standard base ∆ consisting of vectors αi =
λi − λi+1 . Thus for λ = i xi λi ,

(λ, α) X X
σαi (λ) = λ − 2 α= xk λi − (xi − xi+1 )(λi − λi+1 ) = xσ(i) λi
(α, α) k i

where σi,i+1 ∈ Sn is the element switching i and i + 1. This shows that Weyl group is
isomorphic to Sn .
Example 3.4.6. Take example g = sp2n with standard base ∆ consisting of vectors αi =
λi − λi+1 for i < n and αn = 2λn . Thus
P σi for i < n has the same formula as in sln thus
generated group Sn . For σαn , and λ = i xi λi , we have

(λ, αn ) X X
σαn (λ) = λ − 2 αn = xk λi − 2xn λn = xi λi − xn λn .
(αn , αn ) k k<n

This shows that Weyl group is isomorphic to µn2 ⋊ Sn .

96
Lemma 3.4.8. For any λ ∈ E, theres is w ∈ W such that wλ ∈ C̄. In particular, the Weyl
group acts transitively on the set of Weyl chambers.
Proof. For any λ ∈ E, we can choose w ∈ W so that µ := wλ is maximal above partial
order. If µ ∈
/ C̄, then there is an α ∈ ∆ such that (µ, α) < 0. Then we get a contradiction:
2(µ, α)
σα (µ) = µ − α > µ.
(α, α)

3.5 Simple Lie algebras

Dynkin diagrams
Lemma 3.5.1. Let α, β ∈ Φ with α ̸= β. Then

β(Hα )α(Hβ ) ∈ {0, 1, 2, 3}.

Moreover, if this is nonzero, then

 
(α, α) (β, β)
β(Hα )α(Hβ ) = max , .
(β, β) (α, α)
Proof. By Theorem 3.3.7, β(Hα ) is an integer. On the other hand,
4(α, β)2
β(Hα )α(Hβ ) = .
(α, α)(β, β)
Since the paring is positive and α not parallet to β, we have that β(Hα )α(Hβ ) < 4. This
proves the first part. From this part, we know that one of α(Hα ), β(Hα ) has absolute value
1: i.e. 2|(α, β)| is equal to either |(α, α)| or |(β, β)|.
The Dynkin diagram for a simple root system ∆ is a graph with one vertex for each
simple root α. Two simple roots α and β are joined by an edge with multiplicity

m(α, β) = α(Hβ )β(Hα ) ∈ {0, 1, 2, 3}.

If m(α, β) > 1, then the edge is oriented by an arrow pointed towards the short one of α
and β. The simple Lie algebras are completeley classified by their Dynkin diagrams.
Example 3.5.1. Let g = sln with simple roots αi = λi − λi+1 for i < n. Then for i < j
(
4(αi , αj )2 1, j = i + 1
m(αi , αj ) = =
(αi , αi )(αj , αj ) 0, otherwise

Thus the Dynkin diagram of g is given by

An−1 :

97
Example 3.5.2. Let g = spn with simple roots αi = λi − λi+1 for i < n and 2λn . Then for
i < j, we have 
4(αi , αj )2 1, j = i + 1 < n

m(αi , αj ) = = 2, j = i + 1 = n
(αi , αi )(αj , αj ) 
0, otherwise


Notice that
1 1
(αn−1 , αn−1 ) = , (αn , αn ) = .
2(n + 1) n+1
Thus the Dynkin diagram of g is given by

Cn :

The following are Dynkin’s diagrams for other simple Lie algebras.

so2n , Dn : , so2n+1 , Bn :

G2 : , F4 : , F6 : , E7 : , E8 : .

3.6 Highest weight representations

In this subsection, we want to classify its highest weight irreducible representations for a
semisimple Lie algebra.
Let g be a semisimple Lie algebra with a Cartan subalgebra h. Let (E, Φ) be the as-
sociated root system. We fix a Weyl chamber C of the root system (E, Φ). Then we have
a set Φ± of positive or negative
L roots and two sets ∆ ⊂ Φ+ of simple roots. In the root
± +
decomposition, we let n = α∈Φ± gα , and b = h ⊕ n then b is a solvable subalgebra of g.
For each α ∈ Φ+ , we fix a sl2 (C) triple sα = CHα + CEα + CFα with Eα ∈ gα , Fα ∈ g−α
such that
Hα = [Eα , Fα ], α(Hα ) = 2.
Example 3.6.1. Let g = sln (C) and the standard Weyl chamber corresponding to e ∈ Sn .
Then b is the Borel subalgebra of g upper triangular matrices, and n− is the space of strictly
lower triangular matrices.

Definition 3.6.1. Let V be a (possibly infinite dimensional) representation of g.

1. a non-zero vector v ∈ V in a weight space Vλ is called a maximal vector of weight λ if

gα v = 0 for all α ∈ ∆;

2. we say that V is a highest weight representation of g if V is generated as a g-module

by a maximal vector.

98
 By Theorem 3.3.7, every gα for α ∈ Φ+ is generated by gα for α ∈ ∆. Thus for maximal
vector v ∈ Vλ , gα v = 0 for all α ∈ Φ+ .
We will use the universal enveloping algebra U (g) defined as follows:

U (g) := ⊗∗ g/⟨X ⊗ Y − Y ⊗ X − [X, Y ], ∀X, Y ∈ g⟩.

Then a representation of g is identified with a module over U (g). We will use the following
expression of U (g).
Theorem 3.6.2 (Poincaré–Birkhoff–Witt). Let g be a Lie algebra over C and {x1 , · · · , xn }
be a base of g. Let i : g −→ U (g) be the natural embedding. Then U (g) has a base consisting
of ordered monomials:
xd11 · · · xdni , di ∈ Z≥0 .
The above theorem shows that i is injective. Thus we naturally consider g as a subset of
U (g) by omitting i.
For our application, we take g semisimple with a Cartan subalgebra h and a Weyl chamber
C in E as above. Take any order of Φ+ and ∆. Then we make an ordered base of g by

Fαi > Hβj > Eγk , αi , γk ∈ Φ+ , βj ∈ ∆

By Poincaré–Birkhoff– Witt theorem, we have an base for U (g). Now we use this base to
prove the following.
Lemma 3.6.3. Let V be a highest weight representation of g generated by a maximal vector
vλ of weight λ. Then V has a weight space decomposition
X
V = Vµ
Pℓ
with µ of the form λ − j=1 cj αj with ci ∈ Z≥0 . Moreover
1. dim Vµ < ∞ and dim Vλ = 1;

2. if V is irreducible then Vλ \ {0} is the set of all maximal vectors.

Proof. As an U (g)-module, V is a C-linear combination of

Fαd11 · · · Fαdm
m
vλ , di ∈ Z≥0 .

Thus V has a weight space decomposition V = ⊕µ Vµ where Vµ is the sum of Fαd11 · · · Fαdmm
vλ
P Pℓ
satisfying µ = λ− i≤ℓ di αi . As each αi is in Z≥0 ∆, we see that µ has the form λ− j=1 cj αj
with ci ∈ Z≥0 . For each such µ, the dimension of Vµ is bounded by the solution for di in
Pℓ Pm
j=1 cj αj = i=1 di αi . Thus

1. dim Vµ is of finite dimensional;

2. dim Vλ = 1.

99
 If V is irreducible and v ′ ∈ Vλ′ is a maximal vector. Then V is generated by v ′ . Thus we
have λ ≤ λ′ and λ′ ≤ λ. So v ′ ∈ Vλ .

Lemma 3.6.4. Let V be a highest weight representation of g generated by a maximal vector

vλ of weight λ. Then V has a unique irreducible quotient V ′ . Moreover V ′ is also a highest-
weight representation of weight λ.

Proof. We need only show that the sum U of all proper subrepresentation of V does not
include Vλ . Then V /U is irreducible with highest weight λ. By the above lemma, V as a
representation of h is a sum of weight
L spaces. Thus every subrepresentation W of V has a
weight space decomposition W L = Wµ . If W is a proper representation of g, then Vλ is
not in W . It follows that W ⊂
L µ<λ Vµ . Thus the sum U of all proper subrepresentation of
V is included into is in W ⊂ µ<λ Vµ .
Now we have the first main theorem of this subsection.

Theorem 3.6.5. For any λ ∈ h∗ there is a unique irreducible representation V (λ) of highest
weight λ up to isomorphism.

Proof. For the existence, let us construct an U (g)-module

X X
V = U (g)/I(λ), I(λ) := U (g)Eα + U (g)(Hα − λ(Hα )).
α∈∆ α∈∆

Then V is a highest weight module generated by the maximal vector vλ := 1 + I(λ) of weight
λ. The unique irreducible quotient V (λ) will be an irreducible representation of g of highest
weight λ.
For the uniqueness, assume that V1 , V2 are two irreducible highest weight represenations
of g of weight λ generated by maximal vectors v1 , v2 respectively. Then let V3 be the subrep-
resentation of V1 ⊗ V2 generated by v3 := (v1 , v2 ). Then V3 is a highest weight representation
∼
of weight λ with irreducible quotients V1 , V2 . By the above lemma, V1 −→ V2 .

Finite dimensional irreducible representations

In the following, we want to classify all its finite-dimensional irreducible representations for
a semisimple Lie algebra. We will fix a Cartan subalgebra h and a base set ∆ = {α1 , · · · αℓ }
of simple roots. We start with the following definition.

Definition 3.6.6. Let λ ∈ h∗ be a weight.

1. λ is called integral if λ(Hα ) ∈ Z for all α ∈ ∆;

2. λ is called dominant if λ(Hα ) > 0 for all α ∈ ∆.

Our starting point is as follows:

100
Theorem 3.6.7. Let V be a finite dimensional irreducible representation of g. Then V =
V (λ) with λ an integral and dominant weight λ. Moreover the set of weights is stable under
Weyl group.

Proof. Since b is solvable, b is diagonalizable in V . Thus there is a vector v ∈ Vλ in the

weight space of h and killed by n+ . Since V is irreducible, V is generated by v.
For the second part, we notice that for each α ∈ Φ+ , the subspace of V generated by v
under sα has the highest weight λ(Hα ) ∈ Z≥0 . The last part follows from Exercise 3.7.6.
The main theorem of this subsection is as follows:

Theorem 3.6.8. For any integral dominant weight λ, V (λ) is a finite-dimensional irre-
ducible representation. Moreover, the correspondence λ 7→ V (λ) gives a bijection between the
set of integral and dominant weights and the set of isomorphic classes of finite-dimensional
irreducible representations of g.

By the above lemma, we need only show that for any integral and dominant weight λ,
V = V (λ) is finite-dimensional. We need only show that the set M (λ) is finite. For this, in
the following, we will show that M (λ) is stable under Weyl reflection σ(α) for each α ∈ ∆.
So we start with some facts about the restriction of V under sα for a simple root α ∈ ∆.

Lemma 3.6.9. Let v be a maximal vector of V of weight λ and d = λ(Hα ). Then

1. w := Fαd+1 v = 0 and

2. Vα := di=0 CFαi v is stable under sα .

P

Proof. The second statement follows from the first one. For the first one, we first prove that
w is killed by Eβ for any β ∈ ∆:
X
Eβ w = [Eβ , Fαd+1 ]v = Fαi [Eβ , Fα ]Fαj v.
i+j=d

If α = β, then [Eα , Fβ ] = Hα and Hα Fαj v = (d − 2j)Fαj v. Thus we have

X
Eβ w = (d − 2j)Fαd v = 0.
i+j=d

If α ̸= β, then [Eα , Fβ ] ∈ gα−β . Since both α and β are simple, α − β is not a root. Thus
[Eα , Fβ ] = 0.
Now if w ̸= 0, then it is a maximal vector of weight λ − (d + 1)α < λ. This is impossible
by Lemma 3.6.3.

Lemma 3.6.10. The V = V (λ) as an sα -module is the sum of finite dimensional subrepre-
sentations.

101
Proof. Let V ′ be the sum of all finite-dimensional sα -submodules of V . If W ⊂ V ′ be a finite
dimensional sα -subrepresentation of V . Then it is clear that
X
W+ gβ W
β∈Φ

is also a finite dimensional sα -subrepresentation. Thus V ′ is closed under action g. Since

V ′ ̸= 0 by previous lemma, V is irreducible, V = V ′ .

Lemma 3.6.11. Let α ∈ ∆. There is an element wα ∈ GL(V ) such that for any weight
space Vµ in V ,
wα Vµ = Vσα (µ) .

Proof. By the lemma, the action of sα on V can be lifted

 to an action SL2 (C) on V . Then
0 1
we take wα ∈ GL2 (V )) be the image of w := ∈ SL2 (C). The identity follows from
−1 0
Exercise 3.7.6.
Proof of Theorem 3.6.8. Recall that we have a weight space decomposition
M
V = Vµ .
µ∈h∗

Let M (λ) denote the set of µ with Vµ ̸= 0. Then every µ ∈ M (λ) is integral and λ − µ is
dominant. We need only show that M (λ) is finite as each Vµ is finite-dimensional. By the
above lemma, M (µ) is stable under the action of Weyl group W . Thus, we partition M (λ)
into a disjoint union of W -orbits. As W is finite and each orbit contains a dominant weight
by Lemma 3.4.8, we need only show that the subset M (λ)+ of dominant weights is finite.
This is undoubtedly true since it is integral dominant and bounded by λ.

Reductive Lie algebras

For a reductive Lie algebra g over C, a subalgebra h is called a Cartan subalgebra if it is
∼
the preimage of a Cartan subalgebra of g/Z(g) −→ [g, g]: h/Z(g) is a maximal commitative
subalgebra of g/Z(g) consisting of semisimple elements. In other words, it has the form
h = Z(g) + h′ where h′ is a Cartan subalgebra of [g, g].
Let g be a reductive Lie algebra over C with a Cartan subalgebra h. Then g = Z(g)+[g, g],
and h = Z(g) + h′ with h′ = h ∩ [g, g], and we have a root space decomposition for [g, g]:
M
[g, g] := h′ ⊕ gα .
α∈Φ

Let us view Φ as a subset of h∗ via the decompostion

h∗ = h′∗ ⊕ Z(g)∗ .

102
Then we have a root space decomposition for g:
M
g=h⊕ gα .
α∈Φ

The notion of positice roots Φ+ , and simple roots ∆ naturally extend to the pair (g, h).

Definition 3.6.12. We say that a weight λ ∈ h∗ is dominant (resp. integral) if its projection
on h′∗ is dominant. Let P denote the set of integral weights, and P + denote the set of
dominant integral weights.

Then we have a one-to-one correspondence between the set of irreducible representations

of g and P + .

Fundamental weights
Let (g, ∆) be as in the last section. Then ∆ = (α1 , · · · , αℓ ) is base for h∗ . Then the lattice
of integral weights has a base Π = (π1 , · · · , πℓ ) consisting of fundamental weights defined by

2(πi , αj )
πi (Hj ) = = δi,j .
(αj , αj )

Now every dominant integral weight can be written as a positive integral combination:
X
λ= ai π i

with ai are non-negative integers. Let V (πi ) be the irreducible representation of g with
highest weight πi . Then

S(λ) := Syma1 V (π1 ) ⊗ Syma2 V (π2 ) ⊗ · · · ⊗ Symaℓ V (πℓ )

is generated by a heighest weight λ vector v, and V (λ) is the subrepresentation generated

by v. Thus the construction of all irreducible representations becomes the contruction of
fundamental representations and the subrepresentations V (λ).

Weyl construction for sln

simple roots αi = λi − λi+1 for i = 1, · · · , n − 1. Then the fundamental
Let g = sln with P
weights are πi = j≤i λi . The fundamental
P representations are V (πi ) := ∧i (Cn ). Thus for
any dominate integral weight λ = ai πi , the V (λ) is quotient of

T m,n := ⊗m Cn
Pn
with m = i=1 ai i.

103
 Notice that this space has the actions by both sln and Sm . Since both actions are
semisimple, we have a decomposition into the direct sum of irreducible representations of sln
and Sm :
M
T m,n = U ⊠ V ⊠ W, U := Homsln ×Sm (V ⊠ W, T m,n ).
b n ×S
(V,W )∈∈sl bm

We need to understand the pairs (V, W ) ∈ sl

b n × Sbm that appear in this decomposition.

Proposition 3.6.13. For each pair of irreducible representations W of Sm , there is at most

one irreducible representation V of sln appearing in the above decompisition. In other words,

T (W ) := HomSm (W, T m,n )

is an irreducible representation of sln .

Proof. By Schur’s lemma, it suffices to show that Endsln (T (V )) = C. Since Sm is finite, we
have a decomposition M
T m,n = T (W ) ⊠ W
V ∈S
bm

Thus by Schur’s lemma applying to W ’s, we have

M M
EndSm (T m,n ) = End(T (W )) ⊠ EndSm (W ) = End(T (W )).
W ∈S
bm

The left hand side can be computed directly:

End(T m,n )Sm = (⊗m (End(Cn )))Sm = Symm (End(Cn )).

This last space is generated by diagonal element g · · · g for g ∈ SLn (C). Thus we have shown
that End(T (W )) is generated by elements in SLn (C). It follows that

Endsln (T (W )) = EndSLn (C) (T (V )) = EndEnd(T (W )) (T (W )) = C.

The Weyl
P construction will give a precise construction of irreducible representation W
m,n
for λ = ai πi . We view T as a right C[Sm ] module and left sln -module. First, such a
λ defines a partition of [1, · · · , m] by an subsets of length n, an−1 subsets of length n − 1,
... and a1 subsets of length 1. We put this partition in a Young tableau with subsets in
columns. Here is a example for the dominant weight λ = π1 + 2π2 + π3 :

1 4 6 8
2 5 7
3

104
 Let R (resp. C) be the subgroup of Sm fixing every rows (resp. columns). We define
X X
aλ = g, bλ = sgn(g)g, cλ = aλ · bλ .
g∈R g∈C

Then we have an irreducible representation of Sm .

W (λ) = cλ C[Sm ].

The corresponding representation V (λ) of sln is given by

V (λ) = T m,n cλ .

Example 3.6.2.
P Consider the fundamental weight λ = πm corresponding the singlet partition.
Thus cλ = g∈Sm sgn(g)g. The resulting space is

T m,n cλ = ∧m (Cn ).

Weyl construction for sp2n

Let g = sp2n with simple rootsPαi = λi − λi+1 for i = 1 · · · , n − 1 and αn = 2λn . The the
fundamental weights are πi = j≤i λj . The space ∧i (C2n P) has a maximal weight πi -vector
e1 ∧ · · · ei . Thus for any dominant integral weight λ= ai πi , the representation V (λ) is
m,2n
P
again a subrepresentation of T with m = ai i.
Notice that this last space T m,2n has action by sp2n , Sm and also some action by idem-
ponent defined by θp,q for all 1 ≤ p < q ≤ m defined by the following formula at the product
at factors at p, q
n
1 X
θp,q (vp ⊗ vq ) = ψ(vp , vq ) (ei ⊗ ei+n − ei+n ⊗ ei )
2n i=1

where ψ is the symplectic form defined by J. Let Θ be the Lie subalgebra in End(T m,2n )
generated by θp,q . Then the Θ is normalized by action of Sm and commutative with action
by spsn . Let T0m,2n = KerΘ be the intersection of kernel of θp,q ’s. Since sp2n and Sm are both
semisimple, we have a decomposition:
M
T0m,2n = U ⊠ V ⊠ W, U = Homsp2n ×Sm (V ⊠ W, T0m,2n ).
b 2n ×S
(V,W )∈sp bm

We need to understand the pairs (V, W ) ∈ sl

b n × Sbm that appeared in this decomposition.

Proposition 3.6.14. For each irreducible representation W of Sm , there is at most one

irreducible representation V of sp2n appearing in the above decompisition. In other words,

T0 (W ) := HomSm (W, T0m,2n )

is an irreducible representation of sp2n .

105
Sketch of proof. By Schur’s lemma, it suffices to show that Endsp2n (T (V )) = C. Since Sm is
finite, we have a decomposition
M
T0m,2n = T (W ) ⊠ W
V ∈S
bm

Thus by Schur’s lemma applying to V ’s, we have

M M
EndSm (T0m,n ) = EndSm (W ) ⊠ End(T (W )) = End(T (W )).
W ∈S
bm

To compute the left-hand side, we make an embedding

EndSm (T0m,n ) ⊂ EndΘ×Sm (T m,n )

as follows. First of all, with respect to the standard hermitian form on C2n , one can show
that the decomposition
T m,2n = Ker(θp,q ) + Im(θp,q )
is orthogonal. This implies that

T m,2n = T0m,2n ⊕ ImΘ.

Thus for any endomorphism on T0m,2n , we can extend it to an endomorphism with value 0
on Im(Θ). For endomorphisms on T m,2n , we have the following:

End(T m,n )Θ×Sm = (⊗m (End(Cn )))Θ×Sm = (Symm (End(Cn )))Θ .

By invariant theory of sp2n , this last space is generated by diagonal element g · · · g for
g ∈ Sp2n (C). Thus we have shown that End(T (W )) is generated by elements in Sp2n (C). It
follows that

Endsp2n (T (W )) = EndSp2n (C) (T (V )) = EndEnd(T (W )) (T (W )) = C.

P
For a given dominanant weight λ = ai πi the same element cλ will give the representa-
tion V (λ) = T0m,2n cλ .
Example 3.6.3.
P Consider the fundamental weight λ = πm corresponding the singlet partition.
Thus cλ = g∈Sm sgn(g)g. The resulting space is

T0m,n cλ = Ker(ψ : ∧m (C2n ) −→ ∧m−2 (C2n )).

106
3.7 Exercises
Exercise 3.7.1. Let V be a representation of g := sl2 (C) with a weight space decomposition

V = ⊕λ∈Z Vλ , Vλ := {v ∈ V : Hv = λv}.

1. Show that V is isomorphic to a direct sum of irreducible representations ρi := Vi (C2 )

for i ≥ 0 with multiplicity given by

dim Vi − dim Vi+2 .

2. Assume that V = Symn ρ2 . Decompose V into a direct sum of irreducible representa-

tions.

3. Assume that V = Sym2 ρn . Decompose V into a direct sum of irreducible representa-

tions.

Exercise 3.7.2. Let n ∈ N and Dn = (δi+j,n+1 ) the anti-diagonal identity matrix. Let g be
the subset of gln (C) consisting of matrices X satisfying the following equation:

X + Dn X T Dn = 0.

1. Show that g is conjugate to son of skew symmetric matrices in gln (C): there is a
T ∈ GLn (C) such that
T gT −1 = on (C).
In particular, g is a Lie subalgebra of gln (C) isomorphic to on (C).

2. Show that for an X = (Xij ) ∈ gln (C), X ∈ g if and only if the following condition
holds:
Xij + Xn+1−j,n+1−i = 0, ∀1 ≤ i, j ≤ n.
Thus g has a base
eij = Eij − En+1−j,n+1−i , i + j ≤ n.

3. Describe the Killing form on g with the base eij .

P
4. Show that the algeba h = i≤n/2 Ceii of diagonal matrix is a Cartan subalgebra.

5. For an integer m between 0 and n, describe the weight decomposition for ∧m (Cn ).

Exercise 3.7.3. With notation as in Exercise 3.7.2, the matrix Dn defines a g-invariant
quadratic quadratic form on Cn and thus a self-duality of the standard representation Cn
which takes e∗i to en−i . By Remarks A.1.2, A.1.3, we have the following isomorphisms of
g-modules:
∼ ∼
αm : ∧m (Cn ) −→ ∧m (Cn )∨ −→ ∧n−m (Cn ).

107
 1. Show that αm takes
ei1 ∧ · · · eim 7→ ±ej1 ∧ · · · ∧ ejn−m
such that
±en+1−i1 ∧ · · · ∧ en+1−im ∧ ej1 ∧ · · · ∧ ejn−m = e1 ∧ · · · ∧ en .

2. Assume that n = 2m. Then αm is a g-endomorphism of ∧m (Cn ). Show that αm ̸= ±1

2
but αm = 1. Thus ∧m (Cn ) can be desomposed into two nonzero subrepresentations of
g corresponding to ±1 eigen spaces of αm .
Exercise 3.7.4. With notation as in Exercise 3.7.2.
P
1. Let h = i≤n/2 Ci eii be the subalgebra of diagonal matrices with dual base (λi )i≤n/2
for h∗ . Show that h is a Cartan subalgebra with root space decomposition
M
g=h⊕ Ceij
i+j≤n
i̸=j

with Ceij the root space of adh corresponding to the root

λij := (λi − λn+1−i ) − (λj − λn+1−j )
where λj = 0 if j > n/2.
2. For each α ∈ h∗ , write tα ∈ h.
3. For each α ∈ Φ, write an sl2 (C) triple sα = [Hα , Eα , Fα ].
Exercise 3.7.5. Let notation be as in Exercises 3.7.2 and 3.7.4. Let g be the orthogonal
Lie algebra defined byLanti-diagonal matrix Dn and h the Cartan subalgebra of diagonal
matrices. Let g = h ⊕ α∈Φ gα be the root space decomposition. Let Φ+ be the set of roots
λij for i < j.
1. Describe the Weyl chamber which defines the set of positive roots Φ+ ;
2. Show that the set ∆ of simple roots {α1 , · · · , αm } in Φ+ is given as follows:
n = 2m : αi := λi − λi+1 (i ≤ m − 1), αm = αm−1 + αm .
n = 2m + 1 : αi := λi − λi+1 (i ≤ m − 1), αm = λm .

3. Show that the Dynkin diagram for ∆ is given by Dm when n = 2m and Cm when
n = 2m + 1.
Exercise 3.7.6. Let G be a complex semisimple Lie group with Lie algebra g. Let h be a
Cartan subalgebra and Φ the corresponding roots. For each α ∈ Φ with a sl2 (C) triangle
(Hα , Eα , Fα ), let wα ∈ G be the element of the form
wα = exp(π(Eα − Fα )/2) ∈ G.

108
 1. let iα : SL2 (C) −→ G be a Lie group homomoirphism
 induced by the isomorpbhism
∼ 0 1
sl2 (C) −→ sα determined by triangle. Let w = . Show that wα = iα (w).
−1 0
2. Show that for Ad(wα ) stabilizes h with formula
Ad(wα )(H) = H − α(H)Hα .
Thus wα induces an action on h.
Hint: check the identity separately for H ∈ Ker(α) (which commutes with sα ) and
H = Hα (which is in sα = Imd(iα )).
3. Show that wα the induces the reflection σα on h∗ : for any µ ∈ h∗ , H ∈ h
µ ◦ Ad(wα ) = σα (µ) := µ − µ(Hα )α.

4. Let ρ : G −→ GL(V ) be a finite dimensional representation of G. Let V = ⊕Vµ be

the weight space decomposition of the induced representation dρ of g on V . Show that
ρ(wα )Vµ = Vσα (µ) .
Exercise 3.7.7. Let g be a semisimple Lie algebra with a Cartan subalgebra h and a set ∆ of
simple roots. Let V be a finite-dimensional representation
L of g, W its subspace of maximal
vectors with the weight decomposition W = λ∈L Wλ .
1. For each λ ∈ L, show that λ is integral and dominant. Let V (λ) denote an irreducible
representation generated by a maximal vector vλ of weight λ.
2. Show that V has a decomposition into irreducible representations:
∼
M
Wλ ⊗ V (λ) −→ V,
λ∈L

where g acts trvially on Wλ ’s on the left hand side.

Hint: apply canonical decomposition formula in the Theorem 3.2.6, then compare the
spaces of maximal vectors for both sides.
Exercise 3.7.8. Let notation be as in Exercises 3.7.5.
1. Show that the fundamental weights are given as follows:
(a) if n = 2m is even, then
X
± 1
πi = λj (i ≤ m − 2), πm−1 = (λ1 + · · · + λm−1 ± λm ).
j≤i
2

(b) if n = 2m + 1, then
X 1
πi = λi (i ≤ m − 1), πm = (λ1 + · · · λm ).
j≤i
2

109
 2. Find one maximal vector vk in ∧k (Cn ) for k < n/2, and two linearly independent
maximal vectors vk± for k = n/2.
±
3. Check if the weights for vk and vn/2 are fundamental.

Exercise 3.7.9 (Clifford algebras). With notation as in Exercise 3.7.2, let Q(u, v) denote the
quadratic form on V := Cn defined by Dn : Q(ei , ej ) = δi+j,n+1 . Define the Clifford algebra
C(V ) by
C(V ) := ⊗∗ V /⟨u ⊗ v + v ⊗ u − 2Q(u, v), ∀u, v ∈ V ⟩.
Let xy denote the resulting multiplication for x, y ∈ C(V ).
1. Show that C(V ) is generated by eI := ei1 ei2 · · · eik for a subset I = {i1 ≤ i2 · · · ik },
e∅ = 1. Thus dim C(V ) ≤ 2n .
2. Assume n = 2m and write V = W ⊕ W ′ with W = i≤m Cei and W ′ = i>m Cei .
P P
Define a linear map ρ : V −→ End(∧∗ W ) by the following formulae:

ρ(w)v1 ∧ v2 · · · vm = w ∧ v1 ∧ v2 · · · ∧ vm , w∈W
m
X
′
ρ(w )v1 ∧ v2 · · · vm = (−1)i−1 Q(w′ , vi )v1 ∧ · · · vbi · · · vm , w′ ∈ W ′ .
i=1

This map induces an algebra homomorphism

ρ : ⊗∗ V −→ End(∧∗ W ), ρ(v1 ⊗ v2 · · · vm ) := ρ(v1 ) · · · ρ(vm ).

(a) Show that this algebra homomorphism facts through C(V ):

ρ : C(V ) −→ End(W ).

(b) Let I be a subset of {1, · · · , m} with complement I c ⊂ {1, · · · , m}, and dual
I ∗ ⊂ {m + 1 · · · , 2m} consisting of elements n + 1 − i for all i ∈ I. Define the
element θI,J = eJ eI c∗ eI c eI ∗ . Show that ρ(θI,J )eK = ±eJ δI,K for any subset K of
{1, · · · , m}.
(c) Show that ρ is surjective and thus bijective by comparison of dimensions.
(d) Let C even (V ) be the image of ⊗even V of subalgebra of even degree. Show that ρ
induces an isomorophism
∼
ρeven : C even (V ) −→ End(∧even W ) ⊕ End(∧odd W ).

Assume n = 2m + 1 and write V = W ⊕ Cem+1 ⊕ W ′ with W = i≤m Cei , W ′ =

P
3. P
+ − ∗ ∗
i>m+1 Cei . Define a map ρ = ρ ⊕ ρ : V −→ End(∧ W ) ⊕ End(∧ W ) so that its
′ ± i i
restriction on W ⊕ W is given by ρ and ρ (em+1 ) = ±(−1) on ∧ W . Again this map
induces an algebra homomorphism

ρ : ⊗∗ V −→ End(∧∗ W ) ⊕ End(∧∗ W ).

110
 (a) Show that this algebra homomorphism factors through C(V ):

ρ : C(V ) −→ End(∧∗ W ) ⊕ End(∧∗ W ).

(b) Define an element in η ∈ C(V ) by

Y
η = 2−m em+1 (ei + en+1−i )(ei − en+1−i ).
i≤m

Show that η is in the center of C(V ) and that ρ± (η) = ±1.

(c) Show that ρ is surjective and thus bijective by comparison of dimensions.
(d) Show that ρ± induces the same isomorphism
∼
ρeven : C even (V ) −→ End(∧∗ (W )).

Exercise 3.7.10 (Spin Lie algebra). With notation as in Exercise 3.7.9, and define the spin
Lie algebra spin(V ) as the follows:

spin(V ) := {x ∈ C even (V ), x∗ + x = 0, ad(x)(v) := xv − vx ∈ V, ∀v ∈ V }

where x 7→ x∗ is the antilinear involution on C even (V ) defined by

(v1 v2 · · · vr−1 vr )∗ = vr vr−1 · · · v2 v1 .

1. Show that spin(V ) is Lie subalgebra of C even (V ) with a base given by

ei ej − ej ei = 2(ei ej − δi+j,n+1 ), i < j.

Thus dim spin = n(n − 1)/2.

2. Show that the action ad : spin(V ) −→ gl(V ) is injective and preserves the form Q:

Q(ad(x)v, v) = 0, ∀x ∈ spin(V ), ∀v ∈ V.

Thus Im(ad) ⊂ g = so(Q).

Hint: use formula Q(u, v) = uv + vu for u, v ∈ V .

3. Show the following identity in End(V ):

ad(ei ej − ej ei ) = 4Xi,n+1−j := 4(Ei,n+1−j − Ej,n+1−i ), 1 ≤ i < j ≤ n.

∼
4. Show that ad is an isomorphism of Lie algebras: ad : spin(V ) −→ g.
Exercise 3.7.11 (Spin representations). With notation be as in Exercises 3.7.9 and 3.7.10, let
∼
φ : g −→ C even (V ) be the embedding induced by the isomorphism ad : spin(V ) −→ g. Thus
1 1
φ(Xij ) = (ei en+1−j − en+1−j ei ) = (ei en+1−j − δi+j,n+1 .).
4 2
111
 1. Assume n = 2m. The composition ρeven ◦ φ defines two representations of g on ∧even W
and ∧odd W which we call spin representations and denote by σ + and σ − . Show that
±
σ ± are irreducible with the highest weight πm−1 , and more precisely that the standard
base for ∧even W (resp. ∧odd W ) gives weight decompositions with weights in a single
+ −
orbit of πm−1 (resp. πm−1 ) under the Weyl group.
2. Assume that n = 2m + 1. The composition ρeven ◦ φ defines a presentation of g on ∧∗ W
which we call a spin representations and denote by σ. Show that σ is an irreducible
representation of sln of highest weight πm , and more precisely that the standard base
for ∧∗ W gives a weight decomposition with weights in a single orbit of πm under the
Weyl group.
Exercise 3.7.12. Let G be a connected complex Lie group with the semisimple Lie algebra
g. FixLa Cartan subalgebra h of g and a positive root system Φ+ . Then we have subalgebras
n+ = α∈Φ+ gα and b = h + n+ . Taking exponentials, we get a Borel subgrpup B = exp(b).
Let ρ : G −→ GL(V ) be an irreducble finite dimensional representation of G.
1. Show that the representation dρ : g −→ gl(V ) is also irreducible thus V = V (λ) for
some dominant integral λ ∈ h∗ .
2. Show that B fixes a unique line Cv ∈ V and that the action of B on Cv is given by
ρ(b)v = χ(b)v where χ is a character of B of the form χ(eX+H ) = eλ(H) for X ∈ n+
and H ∈ h.
3. Let ρ∗ : G −→ GL(V ∗ ) be the dual represention of ρ. Show that B fixes a unique line
Cv ∗ ⊂ V , and that the action of B on Cv is given by ρ∗ (b)v = χ∗ (b)v, where χ∗ is a
character of B of the form χ∗ (eX+H ) = e−w0 λ(H) , where w0 is an element in the Weyl
group WG (∆) which brings Φ+ to Φ− .
4. Let O(G, V ) denote the image of the matrix coefficient map A : V ∗ ⊠ V −→ C(G).
Show that as a representation of G × G, O(G, V ) is generated by a unique nonzero
function fV up to scale with the following property
fV (b1 gb2 ) = χ∗ (b1 )χ(b2 )f (g), ∀b1 , b2 ∈ B, g ∈ G.

4 Representations of compact Lie groups

Let G be a connected and compact Lie group. This section will describe the set G
b of
isomorphism classes of irreducible representations of G.

4.1 Lie algebra of compact Lie groups

Real Lie algebras
For any real Lie algebra gR , we get a complex Lie algebra gC = gR ⊗ C = gR + igR with a
complex conjugation θ : x + iy 7→ x − iy. Conversely for any complex Lie algebra gC with a

112
complex conjugation θ : X 7→ θ(X) in the sense that

θ[X, Y ] = [θX, θY ], θ2 X = X, θ(aX) = āθX, ∀a ∈ C, X, Y ∈ gC

we can define a real Lie algebra by

gR = {X ∈ gC , θX = X}.

Example 4.1.1. The following are some examples of pairs of real Lie algebras and their
complexifications:

1. (gln (R), gln (C)) where θ(X) = X̄ is just the usual complex conjugation. We can apply
this complex conjugation to get more pairs:

(bn (R), bn (C)), (nn (R), nn (C)), (on (R), on (C)), (sp2n (R), sp2n (C)).

The last two pairs are defined by the following two equations respectively:
 
t t −1 0 1n
X + X = 0, X + J XJ = 0, J := .
−1n 0

2. (u(n), gln (C)), where u(n) is the real subalgebra of gln (C) defined by the equation:

(4.1.1) X + tX̄ = 0.

In this example, θ on gln (C) is the operator θ(X) = − tX̄. This operator also applies
to another pair:
(u(2n) ∩ sp2n (C), sp2n (C)).

The above correspondence gR 7→ (gC , θ) construction gives an equivalence between the

categories of real Lie algebra and complex Lie algebra with an involution. Using these
constructions, most results over complex numbers can be easily transformed into those over
real numbers:

1. gR is solvable iff gC is solvable;

2. gR is nilpotent iff gC is nilpotent;

3. gR is semi-simple iff gC is semi-simple;

4. gR is reductive iff gC is reductive;

5. BgR = Bg ⊗ C;

6. if gR is semisimple, CgR ⊗ C = CgC .

Using the same proof for complex Lie algebra, we obtain the following:

113
Theorem 4.1.1. Let g be a semisimple Lie algebra over R. Then g is the direct sum of its
simple ideals.
Theorem 4.1.2. Let g be a semisimple Lie algebra over R and ρ : g −→ gl(V ) be a repre-
sentation over R. Then V is completely reducible.
Proof. The argument is the same as a complex case with some modifications. So we start
with an exact sequence
0 −→ W −→ V −→ U −→ 0.
We want to construct a section for V −→ U . First, we reduce to the case U = R, then
W is irreducible. Then we use the Casmier operator ρ(Cg ). The main difference is that
the restriction of ρ(Cg )|W is unnecessary in R · Id. But it is still invertible. Thus we have
V = W ⊕ Kerρ(Cg ).
Example 4.1.2 (Symplectic algebra in terms of quaternion algebra). We want to give a more
uniform description for the pair (u2n ∩sp2n (C), sp2n (C)) in terms of embedding of quaternions:
 
a b
H = C + Cj −→ M2 (C), a + bj 7→ .
−b̄ ā
We will show that the pair (u2n ∩sp2n (C), sp2n (C)) is isomorphic to the pair (uH (n), uM2 (C) (n))
of matrices in (Mn (H), Mn (M2 (C))) satisfying the equation
(4.1.2) X + tX ∗ = 0
where (Xij )∗ = (Xij∗ ) with Xij∗ = trXij − Xij the main involution on M2 (C). The complex
conjugation θ on un (M2 (C)) is given by
θ(Xij ) = (θXij )
where θ on M2 (C) is given by
d¯ −c̄
   
a b
θ = .
c d −b̄ ā
More precisely, we will describe these groups with respect to a new base in C2n :
(e′1 , · · · , e′2n ) = (e1 , en+1 , e2 , en+2 , · · · en , e2n ).
With this base the Hermitian matrix In does not change but the sympletic matrix J is change
 
0 1
−1 0 
 

 0 1 

′
J =
 −1 0 .

 .. 

 . 

 0 1 
−1 0

114
Thus u2n does not change, but sp2n (C) is changed to a subalgebra sp2n (C)′ of M2n (C) defined
by the equation:
′
X + J ′ tXJ −1 = 0.
If we write M2n (C) = Mn (M2 (C)), the algebra  of n× n -matrices with entries in M2 (C),
0 1
then J ′ is the diagonal matrix with entry w = . For X = (Xij ) ∈ Mn (M2 (C)) with
−1 0
Xij ∈ M2 (C), we have
(J ′ tXJ ′−1 )ij = w tXji w−1 =: Xji∗ .
Thus the above equation for becomes equation 4.1.2. This shows that sp2n (C)′ = uM2 (C) (n).
Recall that u2n is subalgebra of Mn (M2 (C)) defined by as composition of two transposes
on Mn and M2 . Thus equation 4.1.1 for X = (Xij ) ∈ Mn (M2 (C)) is equivalent to the
following equation:
Xij + tX̄ji = 0.
Thus u2n ∩ sp′2n (C) is a subalgebra of sp′2n (C) defined by
X̄ij = Xij∗ .
t

This exactly means that Xij is the image of H = C + Cj in M2 (C). Thus u2n ∩ sp′2n (C) =
uH (n). The complex conjugation on uM2 (C) (n) is given by
θ(Xij ) = ( tX̄ij∗ ) = (θXij ).
Lemma 4.1.3. Let g be a reductive Lie algebra over R and h is a subalgebra of g consisting
of semisimple elements modulo Z(g). The following are equivalent:
1. h is Cartan subalgebra of g;
2. hC is a Cartan subalgra of gC ;
3. NgC (hC ) = hC ;
4. Ng (h) = h.
Proof. Since g is a direct sum of the semisimple Lie algebra [g, g] and the center Z(g), all
above conditions imply that Z(g) ⊂ h and Z(gC ) ⊂ hC . Thus we can reduce to the case g is
semisimple.
For the implication (1) =⇒ (2), we assume that h is a Cartan subalgebra of g. Then hC
is commutative consisting of semisimple elements. Thus there is Cartan subalgebra h′ of gC
inluding hC . This this implies
h′ ⊂ ZgC (hC ) = Zh (g) ⊗ C.
By the same proof of Step 1 of Theorem 3.3.3, we know that Zg (h) is a nilpotent algebra
with a global Jordan–Checalley decomposition Zg (h) = h + u. Thus ZgC (hC ) = hC + uC . As
h′ is a subalgebra of ZgC (hC ) containing hC and consisting of only seimsimple elements, we
must have h′ = hC . Thus (1) =⇒ (2).
By Theorem 3.3.3, we have (2) =⇒ (3). Othe implications (3)=⇒(4) and (4) =⇒ (1) are
both trivial.

115
 Now let g be a real reductive Lie algebra with a Cartan subalgebra h, then the root space
for (gC , hC ) has a roots space decomposition
M
gC = hC ⊕ gα .
α∈Φ

The complex conjugation of θ on gC takes gα to gθα where θ acts on (hC )∗ by the identification:

HomC (hC , C) = HomR (h, C) = HomR (h, R) ⊗ C.

Taking invariants for the complex conjugation θ on gC , we obtain a root space decomposition
for g: M
g=h⊕ gα,R , gα,R := (gα + gθα )θ .
α∈Φ/θ

Notice that gα,R here is one dimensional if θα = α and two dimensional if θα ̸= α.

There are two special cases: θα = α and θα = −α. We say that g is split in the first
case, and compact in the second case. Here is the existence theroem for split and compact
form of a complex semisimple Lie algebra

Theorem 4.1.4. Let g be a semisimple Lie algebra over C. Then g is the complexifications
of a split subalgebras gs and compact subalgebra gc . Moreover gs and gc are unique up to
isomorphisms.

Finally, we discuss the set of common elements.

Definition 4.1.5. Let g be a reductive Lie algebra over C or R with a Cartan subalgebra h.
An X ∈ h is regular if α(X) ̸= 0 for every α ∈ Φ.

Lemma 4.1.6. Let g be a reductive algebra (over R or C) and h a Cartan subalgebra of g.

1. The set of regular elements in h is open;

2. for each H ∈ h regular, Zg (H) = h.

Proof. For the first one, by definition, the set of regular elements in h is equal to
[
h\ Kerα.
α∈Φ

Since Φ is finite, this set is open and dense. P

For the second one, if X ∈ Zg (H), then we can write X = X0 + Xα in the root space
decomposition. Then we have
X
0 = [H, X] = α(H)Xα .

This implies that all Xα = 0. Thus X = X0 ∈ h.

116
Lie algebras of compact Lie groups
In the rest of this section, we will focus on the representation theory of compact groups.
Example 4.1.3 (Classical compact Lie groups). The following are classical compact Lie
groups.

1. Orthogonal group O(n) as the subgroup of GLn (R) defined by the equation g tg = 1.
This group is not connected. The connected connected component is SO(n) = O(n) ∩
SLn (R) which is semsimple with Lie algebra so(n).

2. The unitary groups U(n) as the subgroup of GLn (C) defined by the equation g tḡ = 1.
This group is connected, reductive, with Lie algebra u(n). We also have a subgroup
of special unitary group SU(n) = U(n) ∩ SLn (C) which is semisimple with Lie algebra
su(n) = u(n) ∩ sln (C) = [u(n), u(n)].

3. Quaternion unitary group UH (n) as the subgroup defined by equation g tḡ = 1. This
group is connected and semisimple with the Lie algebra uH (n).

Theorem 4.1.7. Let G be a compact Lie group and g be its Lie algebra. Then

1. g is reductive;

2. every element in [g, g] is semisimple.

Proof. Consider the adjoint representation Ad : G −→ GL(g). Since G is compact, there is

an inner product on g invariant under ⟨·, ·⟩ on Ad(G):

⟨Ad(g)X, Ad(g)Y ⟩ = ⟨X, Y ⟩.

For any Z ∈ g, taking differential of the above equation for g = etX at t = 0, we obtain the
invariance of the inner product under ad(g):

⟨ad(Z)X, Y ⟩ + ⟨X, ad(Z)Y ⟩ = 0.

From this, it is easy to see that for any ideal h of g, the orthogonal complement h⊥ is also
an ideal. Thus we have a decomposition

g = h ⊕ h⊥ .

This shows that g is reductive.

For the second part, notice that [g, g] −→ g/Z(g) is the Lie algebra of Ad(G). Thus
replacing G by Ad(G), we may assume that g is semisimple and that Ad : G ⊂ GL(g) is in
injective. With inner product on g as above, we may even assume that G ⊂ Un and g ⊂ un .
This implies in particular that for any X ∈ g and t ∈ R, etX is diagonalizable. If X ∈ g
be an nilpotent element, then all eigenvalues of etX is 1. Thus etX = Id. This implies that
X = 0.

117
 We come to the main theorem about the Cartan subalgebras of compact Lie groups.

Theorem 4.1.8. Let G be a compact Lie group with Lie algebra g. Let h1 and h2 be two
Cartan subalgebras of g. Then there is a g ∈ G such that

Ad(g)h1 = h2 .

Proof. Let X1 ∈ h1 and X2 ∈ h2 be regular elements. It suffices to show that there is a g such
that [Ad(g)X1 , X2 ] = 0. Since then Ad(g)(X1 ) ∈ Zg (X2 ) = h2 , and then X1 ∈ Ad(g)−1 h2 .
So Ad(g)−1 h2 ⊂ Zg (X1 ) = h1 . Thus h2 ⊂ Ad(g)h1 . Since h2 is a maximal Cartan subalgebra,
thus h2 = Ad(g)h1 .
Since G is compact, we can take G-invariant inner product on g. Then the equation
[Ad(g)X1 , X2 ] = 0 is equivalent

0 = ⟨Z, [Ad(g)X1 , X2 ]⟩, ∀Z ∈ g.

Using invariance of the inner product, we get

d
⟨Z, [Ad(g)X1 , X2 ]⟩ = ⟨[Z, Ad(g)X1 ], X2 ⟩ = |t=0 ⟨Ad(etZ g)X1 , X2 ⟩.
dt
Thus equation [Ad(g)X1 , X2 ] = 0 for g ∈ G is equivalent to that g ∈ G is a critical point for
the function f (h) := ⟨Ad(h)X1 , X2 ⟩. Since G is compact, we have at least two way to find
a critical point, the maximal point and minimal point.

Corollary 4.1.9. Let G be a compact Lie group with Lie algebra g. Let h be a Cartan
subalgebra of g. Then [
g= Ad(g)h.
g∈G

Proof. For any X ∈ g, by Theorem 4.1.7, X is included into a Cartan subalgebra h′ of g.

By Theorem 4.1.8, there is a g ∈ G such that Ad(g)h′ = h. Thus Ad(g)X ⊂ h.

4.2 Maximal tori of compact Lie groups

Definition 4.2.1. Let G be a compact Lie group. By a torus, we mean a connected commu-
tative subgroup of G. By a maximal torus, we mean a maximal one for the order of inclusion
relation.

Here is a basic property:

Lemma 4.2.2. Let T ⊂ G be a torus with Lie algebra h.

1. T is isomorphic to U (1)n ;

2. The exponential map exp : h −→ H is surjective.

118
 3. T is a maximal torus if and only if h is a Cartan subalgebra of g.

Proof. For the first two, we notice that exp : h −→ T is a homomorphism of Lie groups
which is homeomorphic in a neighborhood of 0. Thus the image is an open subgroup of T .
Since T is connected, this must be surjective. The kernel of exp is thus a lattice in h. We
∼ ∼
may choose an isomorphism h −→ iRn so that Ker exp −→ 2πiZn . Then T −→ U (1)n .
For the last one, we notice that every non-zero element in [g, g] is semisimple; so h is
included into a Cartan subalgebra h′ . It is clear that exp(h′ ) is a torus containing T . Thus
T is maximal if and only if h is Cartan subalgebra of g.
Example 4.2.1 (Maximal tori for classical comapct groups). Here are standard maximal tori
for classical semisimple compact groups.

1. U(n): the standard maximal torus T is the group of diagonal matrices:

 iθ 
e 1
 eiθ2 
t(θ1 , · · · , θn ) :=  , θ1 , · · · , θn ∈ R.
 
. .
 . 
eiθn

Thus T ≃ U (1)n with Lie algebra h ≃ Rn :

 
iθ1
 iθ2 
H(θ1 , · · · , θn ) :=  , θ1 , · · · , θn ∈ R.
 
..
 . 
iθn

2. SU(n): the standard maximal torus T is the group of diagonal matrices:

 iθ 
e 1
 eiθ2  X
t(θ1 , · · · , θn ) :=  , θ1 , · · · , θn ∈ R, θi = 0.
 
 . .. 

eiθn

Thus T ≃ U (1)n−1 with Lie algebra h ≃ Rn−1 :

 
iθ1
 iθ2  X
H(θ1 , · · · , θn ) :=  , θ1 , · · · , θn ∈ R, θi = 0.
 
..
 . 
iθn

119
3. UH (n): the standard maximal torus T is the group of diagonal matrices:
 iθ 
e 1
 eiθ2 
g(θ1 , · · · , θn ) :=  , θ1 , · · · , θn ∈ R..
 
. .
 . 
eiθn
Thus T ≃ U (1)n with Lie algebra h ≃ Rn :
 
iθ1
 iθ2 
H(θ1 , · · · , θn ) :=  , θ1 , · · · , θn ∈ R..
 
..
 . 
iθn
4. SO(2n): the standard maximal torus T consists of n boxes of elements in SO(2) in
diagonal:
 
cos θ1 sin θ1
− sin θ1 cos θ1 
 

 cos θ2 sin θ2


t(θ1 , · · · , θn ) := 
 − sin θ2 cos θ2 ,

θ1 , · · · , θn ∈ R.

 . .. 

 
 cos θn sin θn 
− sin θn cos θn
Thus T ≃ SO(2)n ≃ U(1)n with Lie algebra h ≃ Rn consists of matrices:
 
0 θ1
−θ1 0 
 

 0 θ2


H(θ1 , · · · , θn ) := 
 −θ2 0 ,

θ1 , · · · , θn ∈ R.

 . .. 

 
 0 θn 
−θn 0
5. SO(2n + 1): the standard maximal torus T consists of n boxes of elements in SO(2) in
diagonal and 1 at the end:
 
cos θ1 sin θ1
− sin θ1 cos θ1 
 

 cos θ2 sin θ2


 − sin θ2 cos θ2 
t(θ1 , · · · , θn ) :=  , θ1 , · · · , θn ∈ R.
 
. .

 . 


 cos θn sin θn  
 − sin θn cos θn 
1

120
 Thus again T ≃ SO(2)n with the Lie algebra h ≃ R consisting of matrices:
 
0 θ1
−θ1 0 
 

 0 θ2 

 −θ2 0 
H(θ1 , · · · , θn ) :=  , θ1 , · · · , θn ∈ R.
 
..

 . 


 0 θn


 −θn 0 
0

Let G be a compact group with Lie algebra g. Let T be a a maximal torus G with Lie
algebra h, and the root space decomposition
M
gC = hC ⊕ gα .
α∈Φ

Then we have an action of T on g which stabilize the above decomposition with a unique
character ξα : T −→ C× whose differential dξα = α. In other words, Φ is naturally included
into Tb := Hom(T, C× ) under the differentianl map

d : Tb = Hom(T, C× ) ⊂ Hom(h, C).

Thus for X ∈ h, χ ∈ Tb, χ(eX ) = edχ(X) .

The following is the main theorem of the section.
Theorem 4.2.3 (Maximal Torus Theorem). Let G be a compact and connected Lie group
with Lie algebra g. Let T ⊂ G be a maximal torus. Then:
1. Every element in G is conjugate to an element in T :
[
G= gT g −1 .
g∈G

2. The exponential map exp : g −→ G is surjective.

Proof. Step 1: the equivalence of two statements. In fact, by Corollary 4.1.9,
[
g= Ad(g)h.
g∈G

Thus [ [
exp(g) = Ad(g) exp(h) = Ad(g)T.
g∈G g∈G

For the second equality, we have used Lemma 4.2.2.

121
 Step 2: closeness of exp(g). We will prove part 2 by showing that exp(g) is closed and
open in G. The closeness is easy by observing that exp(g) is the image of the map

f : G × T −→ G, f (g, t) = gtg −1 .

Now T, G are both compacts. Thus Imf is compact. So it is closed in G.

Step 3: It remains to show that exp(g) is open, i.e. exp(g) contains an open neighborhood
for every point eX ∈ exp(g). In the rest of this proof, we will fix such a point g = eX and a
Cartan subalgebra h of g containing X. Then T = exp(h) is a maximal torus of G containing
g. We first show that if g ∈ exp(g) ∩ Z(G), then

g exp(g) = exp(g).

Indeed,
[ [ [
g exp(g) = gh exp(h)h−1 = hg exp(h)h−1 = h exp(h)h−1 = exp(g).
h∈G h∈G h∈G

Here we have used fact that geY = eX+Y for Y ∈ h.

Step 4: Now, we consider the connected component of its centralizer A := ZG (g)0 . This
is proper subgroup of G with Lie algebra
M
a := Zg (g) = {Y ∈ g, Ad(g)Y = Y } = h ⊕ gα .
α∈Φ
ξα (g)=1

Since the curve A contains the curve etX , X ∈ a. Thus g ∈ Z(A) ∩ exp(a). The previous
step implies:
g exp(a) = exp(a).
Step 5: The ugly map φ. Now we have a decomposition g = a ⊕ b with b the complement
of a in the root space decomposition:
M
b := gα .
α∈Φ
ξα (g)̸=1

Let us define a map

φ : g −→ G : φ(Y + Z) = g −1 eZ geY e−Z , Y ∈ a, Z ∈ b.

We claim that dφ is invertible at 0. Assume this claim, then φ(g) contains an open neigh-
borhood of e. Thus we have an open neighborhood of g in φ(g):
[ [ [
gφ(g) = eZ g exp(a)e−Z = eZ exp(a)e−Z ⊂ exp(Ad(eZ )a) ⊂ exp(g).
Z∈b Z∈b Z∈b

122
 It remains to show that dφ is invertible at 0. We calculate this differential explicitly:
d
dφ0,0 (Y, 0) = |t=0 etY = Y
dt
d
dφo,o (0, Z) = |t=0 g −1 etZ ge−tZ = Ad(g −1 )Z − Z.
dt
Thus dφ0,0 = Ida ⊕ (Ad(g −1 ) − 1)b which is obviously invertible on g = a + b.
Corollary 4.2.4. Let G be a connected compact group with a maximal torus. Then the
following statements hold:
1. ZG (T ) = T ;

2. Z(G) ⊂ T .
Proof. For (1), it is clear that ZG (T ) includes T . For other direction, let g ∈ ZG (T ) and
consider the connected compact subggroup ZG (g)0 which includes T as a subtorus. Write
g = eX . Then g ∈ ZG (g)0 as g can be connected to e by an one parameter subgroup etX .
Thus g ∈ ZG (g)0 . By the maximal torus theorem for ZG (g)0 , there is an element h ∈ ZG (g)0
such that g = hgh−1 ∈ T .
Part (2) follows from part (1).

Weyl group WG (T )
Let G be a connected compact Lie group with Lie algebra g. Let T ⊂ G be a maximal torus
of G with Lie algebra h, and the root space decomposition:
M
gC = hC ⊕ gα .
α∈Φ

Let NG (T ) be the normalizer of T in G and WG (T ) = NG (T )/T the corresponding Weyl

group. Then W acts naturally on T , h, and h∗ . the adjoint action of NG (T ) on g stabilizer
the root decomposition. More precisely, for any element n ∈ NG (T ) with image w ∈ WG (T ),

Ad(n)gα = gwα .

Lemma 4.2.5. The WG (T ) is a finite group.

Proof. We need only show that NG (T )0 = T . For this, we need only show LieNG (T ) = h.
In fact, the action of NG (T ) on T induces an inclusion

LieNG (T ) ⊂ Ng (h) = h,

where the second identity follows from Lemma 4.1.3.

Lemma 4.2.6. Let X, Y be two regular elements in h. Then X is conjugate to Y if and
only if there is an element w ∈ WG (T ) such that wX = Y .

123
Proof. The part “if” is trivial. For part “only if ”, we suppose X and Y are conjugate by
one element g ∈ G:
Ad(g)X = Y.
Taking centralizers, we obtain
Ad(g)Zg (X) = Zg (Y ).
By Lemma 4.1.6, we know that Ad(g)h = h. Taking exponential of both sides, we obtain
Ad(g)T = T . Thus g ∈ NG (T ). So Ad(g)X = wX for the element w ∈ WG (T ) represented
by g.
Example 4.2.2 (Weyl group for classical compact groups). Here are Weyl groups for connected
compact Lie groups G with respect to the standard torus T as in Example 4.2.1.

1. U(n) and SU(n): the Weyl group WG (T ) is isomorphic to Sn which acts on T by

permutations.
wt(θ1 , · · · , θn ) = t(θw(1) , · · · , θw(n) ).

2. UH (n): the Weyl group WG (T ) is generated by Sn and diagonal subgroups with entries
1, j ∈ H. Thus WG (T ) can be represented again by the semi-product µn2 ⋊ Sn consisting
of elements:
w = (ϵ1 , · · · , ϵn , σ) : ϵi ∈ {±1}, σ ∈ Sn .
The induced action of w on T is given by

wt(θ1 , · · · , θn ) = t(ϵ1 θσ(1) , · · · , ϵn θσ(n) ).

3. SO(2n) and SO(2n + 1): the Weyl group WG (T ) is generated by the subgroup of A2n
or A2n+1 of elements σ stabilizing the partition

{1, · · · , 2n} = {1, 2} ∪ {3, 4} ∪ · · · ∪ {2n − 1, 2n}.

Thus WG (T ) can be represented by the semi-product µn2 ⋊ Sn consisting of elements:

Y
w = (ϵ1 , · · · , ϵn , σ) : ϵi = ±1, σ ∈ Sn , ϵi = 1 if G = SO(2n)

The induced action of w on T is given by

wt(θ1 , · · · , θn ) = t(ϵ1 θσ(1) , · · · , ϵn θσ(n) ).

4.3 Roots and weights

Analytically integral weights
Let G be a connected compact Lie group with a maximal torus T . Let g and h be their
Lie algebras. In the rest of this section, we will study representations of G. Then for

124
any finite dimensional complex representation V of G, its restriction to T give a character
decomposition:
M
V = Vξ : Vξ := {v ∈ V : tv = ξ(t)v, ∀t ∈ T },
ξ∈Tb

where Tb := Hom(T, U (1)) is the group of characters.

This decomposition has an easy connection with the induced representation on Lie algebra
gC := g ⊗ C and corresponding Cartan subalgebra hC = h ⊗ C. More precisely, h fix each Vξ
with weight
Hv = dξ(H)v.
Thus the set of weights of V as the induced representation of g is in the image

d : Tb −→ h∗C = h∗ ⊗R C.

Here h∗ = Hom(h, R) is the dual of h as a real vector space. This is an injective map with
image
A := {λ ∈ ih∗ , λ(Λ) ∈ 2πiZ} , Λ := Ker(exp : h −→ T ).
We call A the lattice in ih∗ of analytically integral weights. We denote the inverse map
d : Tb −→ A by ξ : A −→ Tb.

Real representations and root space decompositions

Now let VR be a real representation of G and VC = VR ⊗R C. Then the action of G commutes
with the complex conjugation θ : VC −→ VC by θ(w ⊗ a) = w ⊗ ā. It is clear to see that
θVµ = V−µ for any µ ∈ ih∗ . Thus Vµ + V−µ has a real structure

Vµ,R := (Vµ + V−µ ) ∩ VR = {X + θ(X), X ∈ gα } .

Vµ + V−µ = Vµ,R ⊗R C.
In this way, we get a weight deomposition
M
VR = Vµ,R .
µ∈ih∗ /±1

Now we apply this decomposition to the representation Ad : G −→ GL(g). Then the

root decompositions are given by
M M
gC = hC ⊕ gα , g=h⊕ gα,R .
α∈Φ α∈Φ/±1

Here we have used the fact that the complex conjugation θ on gC takes α to −α for α ∈ Φ
as Φ ⊂ ih∗ . It follows that for any α ∈ Φ, ξα ∈ Tb. Thus Φ ⊂ A.

125
su2 -triples
In the rest of this section, we assume that g is semisimple. Define a hermitian form on gC :

(X, Y )θ := −B(X, θY ),

where θ is the complex conjugation on gC :

θ : gC −→ gC , θ(X ⊗ a) = X ⊗ ā.

Lemma 4.3.1. The hermitian form (X, Y )θ on gC is Ad-invariant and positive definite.

Proof. The invariance under Ad(G) is clear. For the positivity, we take a nonzero X =
U + iV ∈ hC with U, V ∈ h′ . Then

(X, X)θ = −B(U + iV, U − iV ) = −B(U, U ) − B(V, V )

Thus need only treat the case X ∈ h. In this case, we have

X
(X, X)θ = − (α2 (U ) + α2 (V )) > 0.
α∈Φ

Now for each α ∈ Φ, we consider a subalgebra

sα = C[gα , g−α ] + Cgα + Cg−α .

This algebra is also stable under θ. Thus it has a real structure

uα := {X ∈ sα : θ(X) = X}.

Lemma 4.3.2. There is an sl2 (C) triangle Hα , Eα , Fα for sα such that

θ(Hα ) = −Hα , θ(Eα ) = −Fα .

Moreover the induced isomorphism

φ : sl2 (C) −→ sα

satisfies the following properties:

1. φ(−X̄ t ) = θφ(X);

2. φ(su2 ) = uα .

126
Proof. Recall that we have defined one element tα ∈ hC such that α(H) = B(tα , H) for any
H ∈ hC . Also we have proved that for any X ∈ gα , Y ∈ g−α

[X, Y ] = B(X, Y )tα .

As α ∈ ih∗ , we have tα ∈ ih. In particular Hα = 2tα

(tα ,tα )
∈ ih. Now the equations

θ(Eα ) = −Fα , [Eα , Fα ] = Hα

are equivalent to
2(Eα , Eα )θ (tα , tα ) = 1, Fα = −θ(Eα ).
By the previous lemma, we can always solve this equation for Eα . The other two properties
are obvious.
By this lemma, corresponding to the generators of su2 :
     
i 0 0 1 0 i
h= , e= , f=
0 −i −1 0 i 0

we can define the su2 -triple for uα

hα = iHα , eα := Eα − Fα , fα = i(Eα + Fα ).

Then we have
[eα , fα ] = 2hα , [hα , eα ] = 2fα , [hα , fα ] = −2eα .
Let Gα be the subgroup of G generated by exp(uα ). Recall that su2 is the Lie algebra of
SU2 , which is simply connected. Thus Gα is a quotient of SU2 .

Weight lattices
Assume that g is semisimple. Then besides A defined as above, we have two additional
lattices P and R in ih∗ already used in the representation theory of gC : the lattices P of
integral weights which are defined as λ ∈ ih∗ such that λ(Hα ) ∈ Z for each α ∈ Φ, and the
sublattices R generated by roots α ∈ Φ. We have a relation R ⊂ A and R ⊂ P .

Lemma 4.3.3. We have A ⊂ P . In other words, for any λ ∈ A and any α ∈ Φ,

λ(Hα ) ∈ Z.

Proof. Since Gα is the quotient of SU2 as above and exp(2πiH) = 1 in SU2 , we have
exp(2πiHα ) = 1 in G. Thus 2πiHα ∈ Λ := Ker exp. Thus 1 = e2πiλ(Hα ) . It follows
that λ(Hα ) ∈ Z.
In summary, we have a relation R ⊂ A ⊂ P . The relation between these lattices and the
topology of G is summarized as follows:

127
Theorem 4.3.4 (Center and π1 ). Let G be a connected compact Lie group a maximal torus
T . Assume that g is semisimple. Let R, A, P be lattices in ih∗ defined as above. Then we
have isomorphisms of groups:

1. Z(G) ≃ A/R.

2. π1 (G) ≃ P/A.

The proof of the first part will be an Exercise. The proof of the second part is omitted
here as it is somewhat involved. For any compact group G, we have a universal cover G e
ad ad
and quotient group G := G/Z(G). The groups G and G depends only on g. The above
e
theorem is equivalent to the following relations:
e −→ G) ≃ P/A,
Ker(G Ker(G −→ Gad ) ≃ A/R.

Before going to examples, we give following description of the integral dual lattices of P
and A in ih:
X 1
P ∗ := Hom(P, Z) = ZHα , A∗ = Hom(A, Z) = Ker(exp : h −→ T ).
α∈Φ
2πi

Example 4.3.1. In the following, we want to compute the lattice R, A, P for the classical
compact group U(n), SU(n) and UH (n) with maximal tori T given in Example 4.2.1.

1. U(n). This is reductive Lie group with the derived subgroup SU(n). Here we only
∼
describe A. The diagonal maximal torus is T = U (1)n . Thus Tb −→ Zn by exponent.
The diagonal Cartan subalgebra h = (iR)n with ususal exponential map h −→ T with
n
P
kernel (2πiZ) . Thus A = Zλi . . Recall that λj acts on h by the formula:

λj H(θ1 , · · · , θn ) = iθj .

∗
P
2. SU(n). Then P by H(θ1 , · · · , θn ) with θi ∈ R and i θi = 0. The lattice ih is
P h is given
given by i Rλi /R λi . Recall that λj acts on h by the formula:

λj H(θ1 , · · · , θn ) = iθj .

The root lattice is generated by λi − λj for i ̸= j. Thus

( )
X X
R= ai λ i , ai ∈ Z, ai = 0
i

For P , we notice that P ∗ is generated by: Eii − Ejj ∈ ih. Thus we have
nX X o X X
P = xi λi ∈ ih∗ : xi − xj ∈ Z xi = 0 ≃ Zλi /Z λi .

128
 For A, we can calculate directly A∗
nX X o
∗ 2πiH
ai = 0 = P ∗ .

A = H ∈ ih : e =1 = ai Eii , ai ∈ Z,

Thus
A = P.

The previous theorem implies that

Z(G) = A/R ≃ Z/nZ, π1 (G) = 1.

In other words, G is simply connected with the cyclic center of order n. We know that
∼
Z(G) −→ µn , the group of n-th roots of unity.

3. UH (n). We will use description in Example 4.1.2. Thus we consider it as a subalgeba

∗
Pn (M2 (C)). Then h is given by H(θ1 , · · · , θn ) with θi ∈ R. The lattice ih is given
of M
by i xi λi . The root lattice is generated by λi ± λj for i ̸= j. Thus
( )
X X
R= ai λ i , ai ∈ Z, ai ∈ 2Z .
i

 
∗ 1 0
For P , we notice that P is generated by Hii which is in the i-entry. Thus
0 −1
X
P = Zλi .
i

For A, we can calculate straightforward A∗

X
A∗ = H ∈ ih : e2πiH = 1 =

ZHii .

Thus X
A= Zλi = P.
i

The previous theorem implies that

Z(G) = A/R ≃ Z/2Z, π1 (G) = 1.

In other words, G is simply connected with the cyclic center Z(G) = {±1}.

129
4.4 Weyl integration formula
Let G be a connected compact Lie group and T a maximal torus. Let dg and dt are Haar
measures on G and T with volume 1.

Proposition 4.4.1. There is a unique measure dḡ for G/T such that for any continuous
function f on G, Z Z Z
f dg = f (gt)dtdḡ.
G G/T T

In fact, we can use the left-invariant differential form ωG , ωT , and ωG/T on G, T, G/T of top
degree to defined measures dg, dt, and dḡ.

Proof. For G and T the dt and dg are induced from invariant forms

ωG ∈ Γ(G, ΩnG ), ωT ∈ Γ(T, ΩℓT )

where n = dim G, ℓ = dim T . For G/T , we need some extra work. Let g = h ⊕ p be a
decomposition of g induced from root space decompsotion of g. Then p is stable under
ad(h). This decomposition induces a decomposition of its dual space

g∗ = h∗ ⊕ p∗ .

Take a highest wedge product to obtain

∧n (g∗ ) = ∧ℓ (h∗ ) ⊗ ∧m (p∗ ), m = n − ℓ = dim p.

Notice that ωG (e) ∈ ∧n (g∗ ) and ωT (e) ∧ℓ (h∗ ) are generators. Thus we have a unique element
α ∈ ∧m (p∗ ) such that ωG (e) = ωT (e) ⊗ α. Now we extend α to a section α e ∈ Γ(G, ΩmG ) using
left translation:
e(g) = ℓ∗g− 1 α.
α
Lemma 4.4.2. The α e is infact invariant under right translation rt by elements in t ∈ T ,
i.e., for any g ∈ G,
rt∗ ω e(gt−1 ).
e (g) = α
Proof. Unfold this identity using the definition, the identity in the lemma is equivalent to
the following identity:
rt∗ ℓ∗g−1 α = ℓ∗tg−1 α = ℓ∗g−1 ℓ∗t α.
Using the commtativity of left and right translation, we get equivalent identity:

rt∗ α = ℓ∗t α.

This is equivalent to the following identity: α = Ad(t)∗ α. Now we consider the action
Ad : T −→ GL(∧m (p∗ )) = R. Since T is compact and connected, thus ImAd = 1.

130
 By the lemma, we see that αe is left invariant under T . Thus, it is descent to a form ωG/T
m
on Γ(G/T, ΩG/T ): for the projection π : G −→ G/T ,
π ∗ ωG/T = α
e.
This form is certainly left-invariant as α
e is left-invariant.
On the other hand, by the maximal torus theorem, we have a surjective map
ψ : G/T × T −→ G : (ḡ, t) = gtg −1 .
Thus we can translate integrals on G to another integral over (G/T ) × T .
Theorem 4.4.3 (Weyl integration formula). Let f be a continuous function on G. Then
Z Z Z
1
f dg = D(t)dt f (gtg −1 )dḡ
G |W G (T )| T G/T

where WG (T ) = NG (T )/T , and D(t) = α∈Φ |1 − ξα (t)|.

Q

Example 4.4.1. Let G = U(n) or SU(n) and T the diagonal matrices:

( n )
X
iθk
T = e Ek,k . .
k=1

Then we have
Y Y
D(t) = (1 − ei(θj −θk ) ) = |1 − ei(θj −θk ) |2
j̸=k j<k
 
Y
2 n(n−1)
Y
2 θj − θk
= |2 sin(θj − θk )/2| = 2 sin .
j<k j<k
2

Similarly, for G = UH (n) and T the diagonal matrces:

( n )
X
T = eiθk Ek,k . .
k=1

we have
Y Y
D(t) = (1 − ei(θj −θk ) ) · (1 − ei(θj +θk ) )(1 − e−i(θj +θk ) )
j̸=k j≤k
Y Y
i(θj −θk ) 2
= |1 − e | · |1 − ei(θj +θk ) |2
j<k j≤k
Y Y
= |2 sin(θj − θk )/2|2 · |2 sin(θj + θk )/2|2
j<k j≤k
    Y
2n 2
Y
2 θj − θk 2 θj + θk
=2 sin sin · sin2 θj .
j<k
2 2 j

131
 We will prove the theorem by first replacing G and T with open subsets Greg , T reg of
regular points. Then the restriction of ψ gives a local homeomorphism

G/T × T reg −→ Greg

of degree WG (T ). What is left is to compute the Jacobian of this map:

φ∗ (dg)ḡ,t = D(t)dtdḡ.

Regular elements
Definition 4.4.4. Let G be a connected compact Lie group. An element g ∈ G is called
regular if ZG (g)0 is a maximal torus of G. Let Greg denote the subset of regular elements of
G.
Proposition 4.4.5. Let G be a connected compact Lie group.
1. Greg is open and dense in G.
2. If T is a maximal torus of G with root system Φ, t ∈ T reg if and only if D(t) ̸= 0.
3. Greg = ∪g∈G gT reg g −1 .
Proof. For parts (1), let g ∈ G and consider the characteristic polynomial of Ad(g) − 1 on g:
X
Q(g, λ) := det(λ − (Ad(g) − 1)|g) = di (g)λi .
i

Here again di (g) are smooth functions of g on G. Let T be a maximal torus of G containing
g with Lie algebra h = Lie(H). Then T ⊂ ZG (g)0 and we have an relation:

h ⊂ Lie(ZG (g)0 ) = Zg (g) := {X ∈ g : (Ad(g) − 1)X = 0}.

Thus we have
ordλ=0 Q(g, λ) = dim ZG (g) ≥ dim T =: ℓ.
Thus di (g) = 0 for i ≤ ℓ and g is regular if and only if dℓ+1 (g) ̸= 0. This shows Greg is
defined by the inquality dℓ+1 (g) ̸= 0. Thus Greg is open. We postpone the density to the end
of proof. For (2), we take t = g as above. Then
Y
Q(t, λ) = tℓ (λ − (ξα (t) − 1)).
α∈Φ

It follows that Y
dℓ+1 (t) = (1 − ξα (t)) = D(t).
α∈Φ

The part (3) follows from the identity G = ∪g∈G gT g −1 . Now by part (6), the density of
Greg follows from the density of T reg in T which is clear as T reg is the complement of finitely
many subtorus defined by ξα (t) = 1.

132
The map ψ
Recall that we have defined a smooth map

ψ : G/T × T −→ G, ψ(gT, t) = gtg −1 .

We want to compute its differential at one point (gT, t):

dψ : TgT (G/T ) × Tt (T ) −→ Tgtg−1 (G).

First of all, we reduce this problem to a morphism ψg,t near the identity point using the
left translations:
ψg,t
G/T × T / G , ψg,t (h, s) = gt−1 htsh−1 g −1 .
ℓg ×ℓt ≀ ≀ ℓgtg−1
 ψ 
G/T × T / G

L we identify Te,e (G/T × T ) with g using the decomposition g = h + p where

Secondly,
p = g ∩ α∈Φ gα . More precisely, we define a map

θ : g −→ G/T × T, θ(H + X) = (eX T, eH ), H ∈ h, X ∈ p.

It is easy to see that dθ is an identity at (e, e). In this way, we have a map

φg,t : g −→ G, φg,t (H, X) = ψg,t ◦ θ(H, X) = gt−1 eX teH e−X g −1 .

Lemma 4.4.6.
dφg,t (H + X) = Ad(g)[(Ad(t)−1 − 1)X + H].
Moreover
det(dφg,t ) = D(t).

Proof. For the first one, just compute the dφ(X) and dφ(H) separately. By the first part,
we than have
Y Y
det(dφg,t ) = det(Adt−1 − 1|p) = (ξα−1 (t) − 1) = (1 − ξα (t)) = D(t).
α∈Φ α∈Φ

Proposition 4.4.7. The map ψ on the regular points:

ψ reg : G/T × T reg −→ Greg

is a |WG (T )| to one diffeomorphism.

133
Proof. By Proposition 4.4.5 and Lemma 4.4.6, ψ reg is a diffeomorphism. Let g ∈ Greg be an
element, we want to show that ψ −1 (g) has WG (T ) elements:
|{(hT, t) : hth−1 = g}| = |WG (T )|.
By the Maximal Torus Theorem, ψ −1 (g) is not empty. So we can write g = h0 t0 h−1
0 . Thus
−1
the equation hth = g is equivalent to
t = nt0 n−1 , n := h−1 h0 ∈ G/T
This equaton implies that
ZG (t)0 = nZG (t0 )0 n−1 .
Since t and t0 are both regular, this means T = nT n−1 . Thus n ∈ NG (T )/T = WG (T ). Also
different n’s will produce different t’s. Thus we proved the above identity of cardinality.

Proof of Weyl integration formula

Finally, we are ready to prove the Weyl integration formula 4.4.3. By Proposition 4.4.5, Greg
is open and dense in G and T reg is open and dence in T . Thus it suffices to prove the formula
for regular elements:
Z Z Z
1
f dg = D(t)dt f (gtg −1 )dḡ
G reg |W G (T )| T reg G/T

We may replace dg, dt, and dḡ by differential forms ωG , ωT , and ωG/T . By proposition 4.4.7,
the left hand side is equal to
Z
1
D(t)f (gtg −1 )ψ ∗ (ωG ).
|WG (T )| G/T ×T reg
Thus it remains to prove the following identity:
ψ ∗ (ωG )(gT, t) = D(t)ωG/T (gT ) ∧ ωT (t).
Now we unfold both sides using definitions:
ψ ∗ (ω)(gT, t) = ψ ∗ (ω(gtg −1 ) = ψ ∗ ℓ∗gt−1 g−1 (ω(e))
ωG/T (gT ) = ℓ∗g (ωG/T (T )), ωT (t) = ℓ∗t−1 ωT (e).
Thus the above identity means
∗
ψg,t ωG (T, e) = D(t)ωG/T (T ) ∧ ωT (e)
where ψg,t is defined by the above commutative diagram. Also we can apply θ∗ to obtain
φ∗g,t ω(e) = D(t)ω(e).
This identity is equivalent to
det dφg,t (e) = D(t).
This is part of Lemma 4.4.6.

134
4.5 Weyl character formula
Let G be a connected compact Lie group and T ⊂ G a maximal torus. Let g = Lie(G) and
h the Cartan subalgebra. We fix a set Φ+ of positive roots.
Let V an irreducible representation. Then the induced representation on the of g is of
the form V (λ) where λ is a dominant analytically integral weight of g. InSthis section, we
want to give a formula for the character χλ of V (λ). Notice that since G = g∈G gT g −1 , the
χλ is determined by its restriction to T . So we need only describe the value χλ on T .

The element ρ and Weyl groups

In the following we fix a set Φ+ of positive roots. Then we have a set ∆ of simple roots, and
the Weyl group WG (∆) ⊂ gl(hC ) generated by reflextions σα for α ∈ ∆:

σα (β) = β − β(Hα )α.

Let us start with some simple facts about the action of WG (∆) on Φ.

Lemma 4.5.1. For any α ∈ ∆ and w ∈ WG (∆), we have the following identities:
+ )∩Φ− |
σα (Φ+ ) ∩ Φ− = {−α}, det(w|hC ) = (−1)|w(Φ .

σα (α) = −α, we need to show for any β ∈ Φ+ with β ̸= α

Proof. For the first identity, as P
+
that σα (β) ∈ Φ . Write β = δ∈∆ nδ δ. Then there is a γ ∈ ∆ \ {α} with multiplicity
nγ > 0. Now in in the formula,
σα β = β − β(Hα )α,
γ still has positive multiplicity nγ . Thus σα β ∈ Φ+ .
The second identity follows from the induction of the length r of a shortest expression
word
w = σα1 · · · σαr , αi ∈ ∆.

For further study of Weyl group, we define

1 X
ρ= α ∈ h∗C .
2 +
α∈Φ

Lemma 4.5.2. Let tρ ∈ ih be the element such that ρ(H) = B(H, tρ ).

P
1. For any α ∈ ∆, ρ(Hα ) = 1. Or equivalently, ρ = i the direct sum of fundamental
weights πi .

2. For any w ∈ WG (∆), wρ − ρ ∈ R ⊂ A.

3. tρ is regular.

135
Proof. The first two parts follow from the first identity in Lemma 4.5.1:

σα (ρ) = ρ − α,
1 X 1 X X
wρ − ρ = α− α= α ∈ R ⊂ A.
2 2
α∈wΦ+ ∩Φ− Φ+ ∩wΦ− α∈wΦ+ ∩Φ−

For the last part, take an α ∈ Φ+ with the decomposition α =

P
ni αi for αi ∈ ∆. Then we
have
X X X X
α(tρ ) = ni αi (tρ ) = ni ρ(tαi ) = ni (tαi , tαi )ρ(Hαi )/2 = ni (tαi , tαi )/2 > 0.
i i

In the following, we will use this element ρ to identify three different definitions of Weyl
groups acting on the Cartan subalgebra h for T :
1. WG (T ) := NG (T )/T which acts on h by adjoint representation.
2. WG (Φ): the subgroup of GL(h) generated by reflection σα for α ∈ Φ:

σα (H) = H − α(H)Hα .

3. WG (∆): the subgroup of WG (Φ) generated by σα for α ∈ ∆.

We have maps
WG (∆) ⊂ WG (Φ) ⊂ WG (T ).
The second one is as in Exercise 3.7.6: for any α ∈ Φ, σα is represented by

wα := exp(π(Eα − Fα )/2) ∈ G.

Theorem 4.5.3. As subgroups of GL(h), the above inclusions are equalities:

WG (∆) = WG (Φ) = WG (T ).

Moreover, the WG (T ) acts freely and transitively on the set C of Weyl chambers.
Proof. We have shown that W (∆) acts transitively on the set C of Weyl chambers by Lemma
3.4.8. So it remains to show that W (T ) acts freely on C . Let C be the Weyl chamber defining
∆ and w ∈ NG (T ) such that wC = C. We want to show that w ∈ T .
The identity Ad(w)C = C implies that Ad(w)Φ+ = Φ+ in ih∗ . Thus we have wρ = ρ.
Let tρ ∈ ih representing ρ. Then Ad(w) fixes tρ and thus w ∈ ZG (tρ ). It suffices to show
that w ∈ ZG (tρ )0 as then w ∈ ZG (tρ )0 = T as tρ is regular.
Now we consider the connected compact subgroup H := ZG (w)0 of G. Then H contains
one-parameter subgroup L = {eistρ , s ∈ R} and w in its center: w can be connected to e by
a maximal torus of G. Fix a maximal torus S of H containing L. Then by Corollary 4.2.4,
w ∈ S . This implies that w ∈ ZG (tρ )0 .

136
The functions Θλ
First of all, let us define the candidates of these functions for any analytically integral
character λ: P
w∈WG (T ) det(w)ξwλ+wρ−ρ (t)
Θλ (t) = Q .
α∈Φ+ (1 − ξ−α (t))

Here we need to check if wρ − ρ is analytically integral:

We also denote by the same symbol the induced function on h:
(w(λ+ρ)(H)
P
w∈W (T ) det(w)e θλ+ρ (H)
Θλ (H) = Q G α(H)/2 −α(H)/2
=
α∈Φ+ (e −e ) η(H)

where X Y
θλ (H) := det(w)ewλ(H) , η(H) := (eα(H)/2 − e−α(H)/2 ).
w∈WG (T ) α∈Φ+

From the second expression, it is easy to see that Θλ is invariant under the action of the
Weyl group: for any w ∈ WG (T ),

Θλ (wH) = Θλ (H).

It suffices to show both numerate and denominator are skew symmetric in the sense that

θλ (wH) = det w · θλ (H), η(wH) = det w · η(H).

In fact, it suffices to check this for w = σα for some simple root α ∈ ∆. Then we have shown
in a previous lemma that
σα (Φ+ ) = (Φ+ \ {α}) ∪ {−α}.
Then it is easy to see that both denominator and numerate change the sign.
Notice that η is essentially a square root of D(H) appearing in the Weyl integration
formula: Y
|η(H)|2 = D(H) = |1 − e−α(H) |.
α∈Φ

Lemma 4.5.4. Let λ, µ ∈ A be two dominated analytically integral weights.

1. Θλ has a Fourier expansion

X
Θλ = eλ + cν e ν , cν ∈ Z.
ν<λ

2. we have the following orthogonality condition in L2 (G):

Z
⟨Θλ , Θµ ⟩G := Θλ (g)Θ̄µ (g)dg = δλ,µ .
G

137
Proof. The first part is clear. For the second part, we compute this directly by Weyl inte-
gration formula:
Z Z
1
⟨Θλ , Θµ ⟩G = D(t)dt (Θλ Θ̄µ )(gtg −1 )dḡ
|W (G, T ) T G/H
Z
1
= D(t)Θλ Θ̄µ )(t)dt
|WG (T )| T
Z X
1
= det(ww′ )ξwλ+wρ−ρ (t)ξ¯w′ µ+w′ ρ−ρ (t)dt
|WG (T )| T w,w′ ∈W
1 X
= det(ww′ )⟨ξwλ+wρ−ρ , ξw′ µ+w′ ρ−ρ ⟩T
|WG (T )| w,w′ ∈W
1 X
= det(ww′ )δwλ+wρ−ρ,w′ µ+w′ ρ−ρ
|WG (T )| w,w′ ∈W

Since both µ and λ are dominate, µ + ρ and λ + ρ are dominate. Thus

wλ + wρ − ρ = w′ µ + w′ ρ − ρ

is equivalent to that λ = µ and w = w′ . The identity of the lemma follows quickly.

Weyl character formula

Theorem 4.5.5. Let V be an irreducible representation of G with highest weight λ = λV ∈
A+ . Then for all g ∈ G:
χV (g) = Θλ (g).

Proof. We need only prove this identity for g = eH for H ∈ hreg . With respect to the weight
decomposition M
V = Vµ
µ∈A

we have X
χV (t) = λ + nµ µ, nµ = dim Vµ ∈ Z≥0 .
µ<λ

We want to show that

χV (eH )η(H) = θλ+ρ (H).
The left hand side is given by
Y X
χV (eH )η(H) =eρ(H) (1 − e−α(H) ) nµ eµ(H)
α∈Φ+ µ∈A
X
(µ+ρ)(H)
= mµ e , mµ ∈ Z
µ∈A

138
Notice that the left hand side is skew symmetric. Thus for any α ∈ Φ, we have
X X
mµ eσα (µ+ρ)(H) = − mµ e(µ+ρ)(H) .
µ∈A µ∈A

Thus mµ = 0 if µ + ρ ⊥ α. So the only the regular term µ + ρ contributes to the sum. Thus
the orbit of µ + ρ contains exactly one dominant element, so we can write
X X
mµ e(µ+ρ)(H) = mµ θµ+ρ .
µ∈A µ∈A
µ+ρ>0

It follows that X
χV = mµ Θµ , mµ ∈ Z.
µ∈A
µ+ρ>0
P 2
Now we apply the identity ⟨χV , χV ⟩ = 1 to obtain mµ = 1. Thus we have a single
µ ∈ A such that
χλ = ϵΘµ , ϵ = ±1
Now comparing Fourier expansion, we get λ = µ, ϵ = 1.
Now we apply the theorem to the trivial representation to get

Corollary 4.5.6 (Weyl denominator formula). Let G be a connected compact Lie group with
a maximal torus T , Then
X
η(H) = θρ (H) = det(w)ewρ(H) , H ∈ h.
w∈W

Theorem 4.5.7 (Weyl dimension formula). Let G be a compact connected Lie group with a
maximal torus T . Let V (λ) be a irreducible representation with the highest weight λ ∈ A+ .
Then
Y (λ + ρ, α) Y  (λ, α)

dim V (λ) = = 1+ .
+
(ρ, α) +
(ρ, α)
α∈Φ α∈Φ

Proof. We will use Weyl formula:

dim V (λ) = χV (λ) (e) = Θλ (e).

However, since η(e) = 0, we need to use some limit of regular elements. Recall that we have
∼
use the Killing form B to give a bijection: h∗C −→ hC which take µ ∈ h∗C to an element tµ
such that µ(H) = B(H, tµ ). For our purpose, we will take H = ϵtρ . Then we have
ϵw(λ+ρ)(tρ )
P
w∈WG (T ) det w · e
dim V (λ) = lim .
ϵ→0 η(ϵtρ )

139
Then for any w ∈ WG (T ),
w(λ + ρ)(tρ ) = (w(λ + ρ), ρ) = (λ + ρ, w−1 ρ) = w−1 ρ(tλ+ρ ).
By Weyl denominator formula, we have
X
det w · eϵw(λ+ρ)(tρ ) = η(ϵtλ+ρ ).
w∈WG (T )

Thus we have
η(ϵtλ+ρ ) Y eϵ(λ+ρ,α)/2 − e−ϵ(λ+ρ,α)/2 Y (λ + ρ, α)
dim V (λ) = lim = lim = .
ϵ→0 η(ϵtρ )
+
ϵ→0 eϵ(ρ,α)/2 − e−ϵ(ρ,α)/2 +
(ρ, α)
α∈Φ α∈Φ

Theorem 4.5.8. The correspondence V 7→ λV defines a bijection between the set G b of iso-
morphism classes of irreducible representations of G and the set A+ of dominant analytically
integral weights of ih.
Proof. We need only show that this map is surjective. The Weyl character theorem shows
that this map induces an embedding of two sets of orthonormal central functions on G:
b ⊂ {Θλ : λ ∈ A+ }.
{χV : V ∈ G}
As the left is already a base, this map must be surjective.
Example 4.5.1. Let G = U(n) or SU(n)P with diagonal torus T . Let λ ∈ A+ , we can compute
Θλ (H)
P directly. In fact, write λ = ai λi with ai ∈ Z, and a1 ≥ a2 · · · ≥ an , and
H= xi Eii . Then
X P
θλ (H) = sgn(σ) · e i aσ(i) xi = det(eai xj ).
σ∈Sn

Applying this to λ + ρ and ρ with

1X 1X
ρ= (λi − λj ) = (n − 2i + 1)λi ,
2 i<j 2 i

we obtain
det(e(ai +(n+1)/2−i)xj ) det(e(ai +n−i)xj )
χV (λ) (H) = = .
det(e((n+1)/2−i)xj ) det(e(n−i)xj )
In terms of t = (ti ) with ti = exi , we have
det(tjai +n−i )
χV (λ) (t) = .
det(tn−i
j )

For dim V (λ), by Weyl dimension formula, we have

Y (λ, λi − λj
 Y
ai − aj

dim V (λ) = 1+ = 1+ .
i<j
(ρ, λi − λj ) i<j
j − i

140
 +
P 4.5.2. For G = UH (n) and λ ∈ A , we can compute
Example P Θλ (H) directly. In fact, write
λ= ai λi with ai ∈ Z, a1 ≥ a2 · · · ≥ an ≥ 0 and H = xi Hii . Then
X P
θλ (H) = sgn(ϵσ) · e i ϵi aσ(i) xi = det(eai xj − e−ai xi ).
(ϵ,σ)∈µn
2 ⋊Sn

Now we apply this formula to λ + ρ and ρ with

1X 1X X
ρ= (λi − λj ) + (λi + λj ) = (n − i + 1)λi .
2 i<j 2 i≤j i

Then we obtain
det(e(ai +n−i+1)xj − e−(ai +n−i+1)xj )
χV (λ) = ,
det(e(n−i+1)xj − e−(n−i+1)xj )
Y (λ, λi − λj )
 Y
(λ, λi + λj )

dim V (λ) = 1+ · 1+
i<j
(ρ, λi − λj ) i≤j
(ρ, λi + λj )
   
Y ai − aj Y ai + aj
= 1+ · 1+
i<j
j−i i≤j
2n + 2 − i − j

4.6 Representions of complex reductive groups

∼
A complex Lie group G is called reductive if its Lie algebra g is reductive and Z(G)0 −→
(C× )t . In this subsection, we want to translate the results about connected compact Lie
groups into connected reductive complex Lie groups.

Highest weight representation

Let G be a connected reductive complex Lie group with Lie algebra g. Then we have a root
space decomposition: M
g=h⊕ gα .
α∈Φ

We fix a Cartan subalgebra h and positive root system Φ+ . Then we have subalgebras:
M
n± = gα , b := h ⊕ n+ .
α∈Φ±

Then we have torus T with Lie algebra h, the unipotent subgroup N with Lie algebra n+ ,
and the Borel subgroup B = N T with Lie algebra b. Let Tb = Hom(T, C× ) denote the group
of complex characters on T , which can be considered as a character of B by quoting N .
Then we have an embedding
d : Tb −→ h∗ .
Let A denote the image and A+ denote the set of dominant weights in A. For any λ ∈ A,
let χλ ∈ Tb so that dχλ = λ.
Our first result is as follows:

141
Theorem 4.6.1. For any λ ∈ A+ , there is a unique irreducible representation V (χλ ) con-
taining a line Cv on which B acts by the character χλ . Moreover, the correspondence
λ 7→ V (λ) gives a bijection between A+ and the set of isomorphism classes of irreducible
representations of G.

It is easy to see that every irreducible representation V of G is of form V (λ). Any such a
V is also an irreducible representation of g; thus, it is of the form V (λ) and then contains a
line Cv killed by n+ and has weight λ by h. The weight decomposition of V for h is deduced
from the weight decomposition for T , we see that v is invariant under N with character χλ
under T .
For the remaining part of the theorem, we notice the representation V (λ) induces a
representation of the universal cover G e of G. We need to show that π1 (G) as a subgroup
of Z(G)e acts trivially on V (λ). For this, we need to use Weyl’s unitary trick. Recall that
g is the complexification of a compact subalgebra g0 with a cartan subalgebra h0 ⊂ h. Let
G0 = exp(g0 ) with the maximal torus T0 = exp(h0 ). Let T1 = exp(ih).

Proposition 4.6.2. We have following statements as real manifolds:

1. The group N and T0 are both isomorphic to Euclidean spaces Rr ’s.

2. The natural map

ψ : T0 \G0 −→ B\G
is an isomorphism.

3. The multiplication map

N × T1 × G0 −→ G
is an isomorphism.

4. π1 (G0 ) = π1 (G).

Proof. For the first part, we can take a faithful representation G ⊂ GL(Cn ) so that G0 ⊂
U(n) and B ⊂ Bn (C). Then N consists of strictly upper triangular matrices, h consists of
diagonal imaginary matrices, and ih0 consists of diagonal real matrices. Thus N and T1 are
both isomorphic to Rr ’s.
For the second part, first, we want to show that ψ is a diffeomorphism at every point.
Concerning the right multiplication of G0 on both T0 \G0 and B\G, it is clear that ψ is
G0 -equivariant. Thus it suffices to show that it is a diffeomorphism at e where dψ is given
by
dψ : g0 /h0 −→ g/b.
The bijectivity follows from the direct sum decomposition called Iwasawa decompsotion:

g = g0 + ih0 + n+ .

142
It suffices to show that n− is included into the right hand side. For this, let θ be the complex
conjugation on g, and X ∈ n− , then

X = (X + θ(X)) − θ(X) ∈ g0 + n+ .

For the second identity, we need only show that b ∩ θ(b) = h0 . This is clear.
Thus we have shown that ψ is a diffeomorphism at every point of T0 \G0 . This implies
that ψ is an open map. Since T0 \G0 is compact, Im(ψ) closed. Thus ψ is surjective. For
injectivity, we need only show that B ∩ G0 = T0 . Let g ∈ B ∩ G0 , then Ad(g) stabilizer b
and g0 and thus h0 . It follows that g ∈ NG0 (T0 ). The corresponding element w in the Weyl
group stabilizer b and thus Φ+ . It follows that w is trivial. This shows that g ∈ T0 .
It is easy to see that the second part implies the third and fourth parts.
Now it is easy to finish the proof of Theorem 4.6.1. Given a dominant λ ∈ A+ . Then λ is
also integral and dominant for the torus T0 . Thus V (λ) descends to a representation for G0 .
It follows that π1 (G0 ) as a subgroup of G0 acts trvially on V (λ). By the above proposition,
π1 (G) acts trivially on V (λ). Thus V (λ) decents to the representation V (χλ ) of G.
From the proof, we have proven the following

Theorem 4.6.3. Let G be a connected reductive complex Lie group, which is the complexifi-
cation of a connected compact Lie group G0 . The restriction map defines a bijection between
the sets of irreducible representations of G and G0 .

By this theorem, we can state the Weyl character formula, denominator formula, and
dimension formula for G without any change.

Peter–Weyl theorem
Let G be a reductive complex Lie group. For any finite-dimensional representation of V of
G, we have matrix coefficeint map

AV : V ∗ ⊠ V −→ C(G, C), ℓ ⊗ v 7→ (g 7→ (ℓ, gv)).

The image is in the subring O(G) of functions f ∈ C(G, C) satisfying the following property:

1. f is holomorphic;

2. G finite under translations on both sides: the subspace W of C(G, C) generated by

LR(x, y)f for all x, y ∈ G is finite dimensional;

Our second result as as follows:

Theorem 4.6.4 (Peter–Weyl Thoeorem). The matrix coefficients map induces a bijection
∼
M
V ∗ ⊠ V −→ O(G).
[V ]∈G
b

143
Proof. The map is injective. For surjectivity, we restrict both to a maximal compact sub-
group G0 . Then we notice that a holomorphic function f on G vanishes on G0 only if f
already vanishes on G.
In the following, we want to make it more precise in terms of the highest weight λ of V
with respect to a triple: (G, T, Φ+ ). Let’s start with the following lemma:

Lemma 4.6.5. Let V = V (λ) be finite-dimensional representation of G of the highest weight

λ ∈ A+ . Then V ∗ = V (−w0 λ) where w0 is the unique element in WG (T ) which brings Φ− to
Φ+ .
L
Proof. Let V = α∈Λ Vα be a L
weight space decomposition of V . Then we have a weight
space decomposition for V ∗ = ∗ ∗ ∗
−α∈−Λ V−α , where V−α := (Vα ) . Notice that −λ is the
lowest weight in −Λ. Since Λ is stable under WG (T ), −w0 λ ∈ −Λ. It is clear that −w0 λ is
the highest weight in −Λ. Thus V ∗ is a representation with highest weight −w0 λ.
Thus we can rewrite Peter–Weyl theorem as
M
O(G) = V (−w0 λ) ⊠ V (λ).
λ∈A+

Example 4.6.1. Let G = GL(n, C) with standard maximal torus and a set of positive roots
Φ+ . The element in WG (T ) which bring positive roots to negative roots is given by
 
1
w0 :=  . . .  .
 

A dominant integral weight can be written as: λ = a1 λ1 + a2 λ2 + · · · with ai ∈ Z and

a1 ≥ a2 · · · ≥ an . The dual of λ is

−w0 λ = −an λ1 − an−1 λ2 − · · · .

If we take a maximal vector v for V (χ−w0 λ ), then we obtain the following:

Theorem 4.6.6. For any λ ∈ A+ , the representation V (χλ ) can be realized as

B (χw0 λ ) := {f ∈ O(G),
IndG f (bg) = χw0 λ (b)f (g), ∀b ∈ B, g ∈ G} .

Notice that the representation V (λ) is generated by a maximal vector v of the highest
weight λ. This implies that ImAV is generated by a unque function fλ ∈ O(G) with the
following property:

fλ (b1 gb2 ) = χw0 λ (b1 )f (g)χλ (b2 ), f (w0 ) = 1.

Thus, this holomorphic function determines the representation V (χλ ).

144
Example 4.6.2. Let G = GLn (C) with diagonal torus T and standard positive root system Φ+
defined by the standard Borel group B of upper triangular matrices. Let λ = a1 λ1 +a2 λ2 +· · ·
be an integral and dominant weight of T . Then −w0 λ = −an λ1 − an−1 λ2 + · · · .The character
χλ on B is defined as follows:
 
t1
χλ 
 .. a a a
 := t11 t22 · · · tnn .

.
tn

Then by Peter–Weyl theorem, we have a relization of V (χλ ):

B (χw0 λ ) = {f ∈ O(GLn (C)) : f (bg) = χw0 λ (b)f (g)} .

IndG

Furthermore, this space is generated by a holomorphic function fλ on GLn (C) such that

fλ (b1 gb2 ) = χw0 λ (b1 )fλ (g)χλ (b2 ).

For a formula for fλ , lets decompose λ into fundamental weights:

n
X X
λ= mi λi , mi = ai − ai+1 , πi = λj .
i=1 j≤i

We actually have the following explicit formula for fλ :

Y
fλ (g) = fπmi i , fπi = deti (g)

where detk (gij ) = det(gi,j )1≤i,n+1−j≤k , the determinant of upper right k × k-block. This
function is the matrix coefficients of maximal vectors:
e1 ∧ e2 · · · ∧ en−i
∈ ∧n−i Cn ⊗ det−1 = V (πi )∗ , e1 ∧ · · · ∧ ei ∈ ∧i (Cn ) = V (πi )
e1 ∧ e2 · · · ∧ en
e1 ∧ e2 · · · ∧ en−i ∧ g(e1 ∧ · · · ∧ ei ) = ±fπi (g)e1 ∧ e2 · · · ∧ en .
Finally, we can interprete V (χλ ) the space of the line bundle on a complex manifolds:

Theorem 4.6.7 (Borel–Weil). Let L (χ) denote the line bundle on B\G defined by

L (χ) := C(χλ ) ×B G.

Then the representation V (χλ ) is isomorphic to the H 0 (B\G, L (χ)).

Example 4.6.3. Let G = GLn (C) with Borel group Bn (C) of upper triangular matrices. In
this case P := B\G is the variety parameterizing flags:

F0 = 0 ⊂ F1 ⊂ F2 · · · ⊂ Fn = Cn , dim Fi = i.

145
At each point x ∈ P, there are flags Fi (x) ⊂ Cn . Thus we have quotient space Ei (x) =
Cn /Fn−i (x) of dimension i and determinant line Li (x). These spaces form a vector bundle Ei
on P and determiant line bundle det Ei = Li on P. On the other hand, the maps Cn −→ Ei (x)
defines a section ∧i Cn −→ Li . In fact, we have an isomorphism
∼
∧i Cn −→ H 0 (P, Li ).

This is exactly the isomorphism

∼
V (πi ) −→ H 0 (P, Li )

predicted by the Borel–Weil theorem.

4.7 Exercises
Exercise 4.7.1. Let g a Lie algebra over R. Let gC = g ⊗R C be the base change Lie algebra
from g under extension C/R. Then
1. g is nilpotent iff gC is nilpotent;

2. g is solvable iff gC is solvable;

3. g is semi-simple iff gC is semi-simple;

4. BgL = Bg ⊗ C;

5. if g is semisimple, Cg ⊗ C = CgC .
Exercise 4.7.2. Let g ⊂ gln (C) be a complex Lie algebras which is stable under the anti-
involution X 7→ X ∗ := X̄ t . Let h be the (real) subalgebra of elements X ∈ g such that
X ∗ = −X.
1. Show that
∼
g = h + ih −→ h ⊗R C.

2. Show that the correspondence g 7→ h defines a bijection between the set of ∗-stable
complex subalgebras of gln (C) and the set of subalgbras of un .

3. Show that both g and h are reductive.

Exercise 4.7.3 (Weyl group). Let G be a connected compact Lie group with Lie algebra g.
Let T ⊂ G be a maximal torus of G with Lie algebra h, and the root space decomposition:
M
g=h⊕ gα .
α∈Φ

Let NG (T ) be the normalizer of T in G and W = NG (T )/T the corresponding Weyl group.

Then W acts naturally on T , h, and h∗ .

146
 1. Show that LieNG (T ) = LieT = h.

2. Show that W is a finite group.

3. Show that the adjoint action of NG (T ) on g stabilizer the root decomposition. More
precisely, for any element n ∈ NG (T ) with image w ∈ W ,

Ad(n)gα = gwα .

4. Let X, Y be two regular elements in h. Show that X is conjugate to Y if and only if

there is an element w ∈ W such that wX = Y .

Exercise 4.7.4 (Serpenski, Exercise 5.11). Let T be the maximal torus of SO(3) given by
  
 cos θ sin θ 0 
T = − sin θ cos θ 0 :
 θ∈R .
0 0 1
 

Find g ∈ SO(3) so that ZG (g)0 = T but with ZG (g) disconnected.

Exercise 4.7.5 (Serpenski, Exercise 5.12). Let G be a connected compact Lie group. Suppose
S is a connected abelian Lie subgroup of G.

1. If g ∈ ZG (S), show a maximal torus T of G exists so that g ∈ T and S ⊂ T .

2. Show that ZG (S) is the union of all maximal tori containing S and connected.

3. For g ∈ G, show that ZG (g)0 is the union of all maximal tori containing g.

Exercise 4.7.6 (Serpenski, 5.13). Let G be a connected compact complex Lie group. Show
that G is Abelian and isomorphic to Cn /Λ for some discrete subgroup Λ of Cn of rank 2n.
Hint: Study the holomorphic representation Ad : G −→ GL(Lie(G)).
Exercise 4.7.7 (Serpenski, 6.3.4). Let V be finite dimensional vectros spaces over R with
dual vector space V ∗ = Hom(V, R). By a lattice Λ ∈ V , we mean a discrete subgroup of V
that spans V . For such a lattice, we define its dual lattice by

Λ∗ := {ℓ ∈ V ∗ : ℓ(λ) ∈ Z, ∀λ ∈ Λ}.

1. For a lattice Λ in V , show that Λ∗ is also a lattice in V ∗ and Λ∗∗ = Λ.

2. If Λ1 ⊂ Λ2 in an inclusion of two lattices, then Λ∗2 ⊂ Λ∗1 .

3. If Λ1 ⊂ Λ2 , show that Λ2 /Λ1 ≃ Λ∗1 /Λ∗2 .

Exercise 4.7.8. This exercise is denoted to prove the first part of Theorem 4.3.4. Let G be
a connected Lie group with a maximal torus T . Consider the lattices R and A in ih∗ , and
their duald R∗ and A∗ in ih.

147
 1. Show that A∗ = 1
2πi
Ker(exp : h −→ T ).
2. For H ∈ h, the following are equivalent:
(a) eH ∈ Z(G),
(b) Ad(eH ) = 1 on g,
(c) α(H) ∈ 2πiZ,
(d) H ∈ 2πiR∗ .
∼ ∼
3. Show that Z(G) −→ 2πiR∗ /2πiA∗ −→ R∗ /A∗ . (Hint: use Corollary 4.2.4).
4. Show that G ≃ A/R.
Exercise 4.7.9. For the orthogonal group SO(n) with maximal torus T constructed in Ex-
ample 4.2.1, compute the lattices R, A, P and then the finite groups Z(G) and π1 (G).
Exercise 4.7.10 (Serpenski 7.8, p162). Let G be a connected compact Lie group G with a
maximal torus T . For H ∈ h, show that
 
H |Φ|
Y
2 α(H)
D(e ) = 2 sin .
+
2i
α∈Φ

Exercise 4.7.11 (Serpenski 7.9, p162). Let f be a continuous central function on SU(2). Use
Weyl integration formula to show that
2 π
Z Z  iθ 
e
f (g) = f sin2 θdθ.
SU(2) π 0 e−iθ

Exercise 4.7.12 (Serpenski 7.10, p162). Let G be a connected compact Lie group with a
maximal torus T .
1. If f is an L1 -central function on G, show that
Z Z
1
f (g)dg = D(t)f (t)dt.
G |WG (T )| T

2. Show that the map f 7→ |WG (T )|−1 D · f |T defines a norm preserving isomorphism
between the L1 -central functions on G and the WG (T )-invariant L1 -functions on T .
3. Show that f 7→ |WG (T )|−1/2 |D|1/2 f |T defines a unitary isomorphism between the L2 -
central functions on G to the WG (T )-invariant L2 -central functions on T .
Exercise 4.7.13. Recall the Lie algebras g and h defined by anti-diagonal matrix Dn denote
the defined in Exercise 3.7.2. Now we define a compact subgroup of SUn by

G := g ∈ SUn : g t Dn g = Dn .


Let T be subgroup of diagonal matrices. Let gR and hR be their Lie algebras.

148
 1. Show that gR + igR = g and hR + ihR = h.

2. Write down lattices R ⊂ A ⊂ P for the pair (gR , hR ).

3. Find the formula D(t).

Exercise 4.7.14. For each group G below, show that the listed representations V of G has
minimal dimension among non-trivial irreducible representations:

1. For G = SU(n), V is the standard representation on Cn or its dual;

2. For G = UH (n), V is the standard representation on Cn .

Exercise 4.7.15. Let G be a connected compact Lie group.

1. If G is semisimple, show that for any d ∈ N, there are only finitely many isomorphism
classes of irreducible representations of G of dimension d.

2. If G is not semisimple, show that there are infinitely many isomorphism classes of
representations of dimension 1.

Exercise 4.7.16. Let G = U(2) with diagonal maximal torus T . Let C2 be the standard
represention and det the one representation defined by derterminants.

1. Let λ = aλ1 + bλ2 be a dominant integral weight: a, b ∈ Z and a ≥ b. Show that the
Weyl charcater formula gives
 a1 +1 a1 +1 
t t2
det 1 a2 1 −a2
aX
t1 ta22
Θλ (t1 , t2 , t3 ) = = (t1 t2 ) a2
ti1 ta21 −a2 −i .
(t1 − t2 ) i=0

2. Show that
∼
V (λ) −→ Syma1 −a2 (C2 ) ⊗ deta2 .
Thus every irreducible representation of G is of form Symm (C2 ) detn .

3. For non-negative integer m ≥ n,

n
∼
M
m n
2
Sym (C ) ⊗ Sym (C ) −→ 2
Symm+n−2ℓ (C2 ) ⊗ detℓ .
ℓ=0

4. For any two dominant weights λ = a1 λ1 + a2 λ2 , µ = b1 λ1 + b2 λ2 with ai , bi ∈ Z and

a1 ≥ a2 , b1 ≥ b2 ,
min(a1 −a2 ,b1 −b2 )
M
V (λ) ⊗ V (µ) = V ((a1 + b1 − ℓ)λ1 + (a2 + b2 + ℓ)λ2 ).
ℓ=0

149
Exercise 4.7.17. Let G = U(3) with the standard diagonal maximal torus. Let H = U(2) be
embedded into U(3) by  
  a b
a b
7→  c d  .
c d
1
Let λ = a1 λ1 + a2 λ2 + a3 λ3 be an integral dominant weight. We want to decompose the
representation V (λ)|H using Weyl formula. We will use Θa1 ,a2 ,a3 (t1 , t2 , t3 ) for the charac-
ter functon of V (λ) evaluated at the diagonla matrix t1 E11 + t2 E22 + t3 E33 , and similarly
Θa1 ,a2 (t1 , t2 ) for representation of H on V (a1 λ1 + a2 λ2 ).

1. Show that the character of V (λ) can be expressed into the form:
 a1 +2 a1 +2 a1 +2 
t1 t2 t3
a2 +1 a2 +1
det t1
 t2 ta32 +1 
a3 a3
t1 t2 ta33
Θλ (t1 , t2 , t3 ) = − .
(t1 − t2 )(t2 − t3 )(t3 − t1 )

2. Show the followin identity:

1
Θλ (t1 , t2 , 1) = (Θa2 ,a3 (t1 , t2 ) − Θa1 +1,a3 (t1 , t2 ) + Θa1 +1,a2 +1 (t1 , t2 )).
(1 − t1 )(1 − t2 )

3. Show the following identity as power series of t1 and t2 :

∞
1 X
= Θk,0 (t1 , t2 ).
(1 − t1 )(1 − t2 ) k=0

4. Let α1 ≥ α2 be integers. Use previous exercise to show that

Θα1 ,α2 (t1 , t2 ) X

= Θβ1 ,β2 (t1 , t2 )
(1 − t1 )(1 − t2 ) β
1 ≥α1 ≥β2 ≥α2

where the is over all integers β1 , β2 satisfying the interleave condition.

5. Show the following identity of power series:

X X
Θλ (t1 , t2 , 1) = Θb1 ,b2 (t1 , t2 ) − Θb1 ,b2 (t1 , t2 )
b1 ≥a2 ≥b2 ≥a3 b1 ≥a1 +1≥b2 ≥a3
X
+ Θb1 ,b2 (t1 , t2 ))
b1 ‘≥a1 +1≥b2 ≥a+2+1
X
= Θb1 ,b2 (t1 , t2 ).
a1 ≥b1 ≥a2 ≥b2 ≥a3

150
 6. Show the following branching law:
M
V (λ)|H = V (b1 λ1 + b2 λ2 )
a1 ≥b1 ≥a2 ≥b2 ≥a3

Exercise 4.7.18. Let G be a connected compact Lie group with a maximal torus T . Suppose
V is a representation of G possesses a maximal vector of weight λ. If dim V = dim V (λ),
∼
show that V −→ V (λ), and, in particular, V is irreducible.
Exercise 4.7.19. Let G be the group SO(2n) or SO(2n + 1) with maximal torus T and Weyl
group in Examples 4.2.1 and 4.2.2.
1. Describe the fundamental Weyl chamber C.
2. Describe positive roots and its half sum ρ.
3. Describe dominant integral weights.
4. Write donw the Weyl character formula.
5. Compute dimension of V (λ) for a dominant integral weight λ.
Exercise 4.7.20. Let G be a connected complex Lie group with Lie algebra g. FixLa Cartan
+ +
subalgebra h of g and a positive root system Φ . Then we have subalgebras n = α∈Φ+ gα
and b = h + n+ . Taking exponentials, we get a torus T = exp(h) and Borel subgrpup
B = exp(b), and its unipotnet radical N := exp(n+ ). Let ρ : G −→ GL(V ) be an irreducble
representation.
1. There is a unique nonzero vector v ∈ V up to scale multiple fixed by N . Moreover the
action of T on v is given by a character χ : T −→ C× .
2. The representation dρ : g −→ gl(V ) is also irreducible with weght λ = dξ.

A Appendix
A.1 Linear algebra
Tensor algebra
Definition A.1.1. Let F be a field and V a vector space over F .
1. We define the tensor algebra of V , denoted ⊗∗ (V ), to be the algebra of tensors on V
with multiplication being the tensor product. In other words,
∞
M
∗
⊗ (V ) = ⊗k (V ), ⊗0 V = F, ⊗1 V = V
k=0

where k > 0, ⊗k (V ) is the tensor product V ⊗ V ⊗ · · · ⊗ V of V with itself k times. The

k-th component ⊗k (V ) of ⊗∗ (V ) is called the k-th tensor power of V . If u ∈ ⊗k (V )
and v ∈ ⊗s (V ), denote the multiplication in this algebra by u ⊗ v.

151
 2. We define symmetric algebra of V , denoted by Sym∗ (V ) to be the quotient of ⊗∗ (V )
modulo the two sided ideal I generated by (x ⊗ y − y ⊗ x). This ideal is in fact graded.
Thus we have
M
Sym∗ (V ) = Symk (V ), Sym0 V = F, Sym1 V = V
k≥0

The k-th component Symk (V ) of Sym∗ (V ) is called the k-th tensor power of V . If
u ∈ ⊗k (V ) and v ∈ ⊗s (V ), denote the multiplication in this algebra by u · v.

3. We define the exterior algebra ∧∗ (V ) to be the quotient algebra of ∧∗ (M ) modulo the

two-sided ideal J generated by all elements of the form x ⊗ x for x ∈ V . Again J will
be graded. Thus
M
∧∗ (V ) = ∧k (V ), ∧0 V = F, ∧1 V = V
k≥0

The k-th component ∧k (V ) of ∧∗ (V ) is called the k-th exterior power of V . If u ∈

∧k (V ) and v ∈ ∧s (V ), denote the multiplication in this algebra by u ∧ v, then u ∧ v =
(−1)ks v ∧ u. When characterisitc of F is not 2, J is also generated by elements of the
form
x ⊗ y + y ⊗ x = (x + y) ⊗ (x + y) − x ⊗ x − y ⊗ y, x, y ∈ V

Remark A.1.1. The construction ⊗∗ , Sym∗ , ∧∗ is functorial in the sense for any homomor-
phism ϕ : V −→ W , there are canonical homomorphisms:

⊗∗ ϕ : ⊗∗ V −→ ⊗∗ W, Sym∗ ϕ : Sym∗ V −→ Sym∗ W, ∧∗ ϕ : ∧∗ V −→ ∧∗ W.

Remark A.1.2. Assume V = F n with standard base e1 , · · · , en .

1. The algebra Sym∗ (V ) is the polynomial algebra F [e1 , · · · , en ].

2. The space ∧∗ V has a base consisting (eI : I ⊂ [1, · · · , n]) with eI := ei1 ∧ · · · ∧ eik .
Then eI ∧ eJ is no zero only
 ifI ∩ J = ∅. In this case eI ∧ eJ = ϵI,J eI∪J where ϵI,J = ±1.
n
In particular, dim ∧i V = and dim ∧∗ V = 2n .
i
3. The wedge product defines a natural duality for every i:
∼
∧i V ⊗ ∧n−i V −→ ∧n V −→ C

where the last step takes e1 ∧ e2 · · · en to 1.

4. If ϕ ∈ EndV , then the induced morphism on ∧n V is given by det ϕ.

Definition A.1.2. Let V be a vector space over a field F of characteristic 0, V ∨ = Hom(V, F )
be the space of linear forms on V .

152
 1. The space of k-forms on V is defined to be ⊗k V ∨ as a dual space of ⊗k V by the pairing:
Y
⟨·, ·⟩ : ⊗k V ∨ × ⊗k V −→ F, ⟨ℓ1 ⊗ · · · ⊗ ℓk , v1 ⊗ · · · ⊗ vk ⟩ = ⟨ℓi , vi ⟩.
i

2. The space of symmetric k-form is defined to be Symk V ∨ as the dual space of Symk V
by the following pairing:
XY
⟨·, ·⟩ : Symk V ∨ × Symk V −→ F, ⟨ℓ1 · · · ℓk , v1 · · · vk ⟩ = ⟨ℓi , vσ(i) ⟩,
σ∈Sk i

where Sk the group of k-letters.

3. The space of alternative k-forms on V is defined to be ∧k V ∨ as the dual space of ∧k V
by the following pairing:
X Y
⟨·, ·⟩ : ∧k V ∨ × ∧k V −→ F, ⟨ℓ1 ∧ · · · ∧ ℓk , v1 ∧ · · · ∧ vk ⟩ = ϵ(σ) ⟨ℓi , vσ(i) ⟩,
σ∈Sk i

where ϵ : Sk −→ {±1} is the sign function on Sk .

Remark A.1.3. Combining with Remark A.1.2, part 3, we have an isomorphism
∼ ∼
∧i (V ∨ ) −→ ∧i (V )∨ −→ ∧n−i (V ).
Concretely for V = Cn with base ei and V ∗ = Cn with dual base e∗i , this isomorphism takes
e∗I to ϵ(I)eJ where ϵ(I) = ±1 so that
eI ∧ eJ = ϵ(I)e1 ∧ · · · en .

Jordan–Chevalley deomposition
Let K be a field of characteristic 0.
Lemma A.1.3. Let V be a finite dimensional vector space over K and let X ∈ End(V ).
Then there is a unique decomposition
X = Xs + X n
with Xs semisimple (i.e., with separated minimal polynomial ), and Xn nilpotent (i.e., Xnk = 0
for k >> 0), and Xs Xn = Xn Xs . Moreover, there are polynomials P (T ), Q(T ) ∈ K[T ] with
vanishing constant term such that
Xs = P (X), Xn = Q(X).
Consider the map: ad : End(V ) −→ End(End(V )) by conjugugation:
ad(X)(Y ) = XY − Y X.
In the following we want to shoe that the map preserve the Jordan–Chevalley decomposition.

153
Lemma A.1.4. If X ∈ End(V ) is nilpotent, then ad(X) is ad-nilpotent.

Proof. Assume that X k = 0 for some k > 0. We want to show that for some ℓ > 0,

ad(X)ℓ (A) = 0, ∀A ∈ End(V ).

Notice thet ad(X) on End(End(V )) is given by left and right multiplication of X:

ad(X)(A) = XA − AX = L(X)(A) − R(X)A = (L(X) − R(X))(A).

Since L(X) commutes with R(X), we have

X ℓ
ℓ ℓ
ad(X) (A) =(L(X) − R(X)) A = L(X)i L(Y )j A
i+j=ℓ
i
 
X ℓ
= (−1)j L(X i )R(X j )A.
i+j=ℓ
i

If ℓ ≥ 2k, then one of i, j ≥ k. Thus we have ad(X)ℓ A = 0.

Lemma A.1.5. Let V be a finite dimensional space and X ∈ gl(V ) an element with the
Jordan–Chevalley decomposition
X = X s + Xn .
Then in ad(X) ∈ gl(gl(V )) has the Jordan–Chevelley decomposition:

ad(X) = ad(Xs ) + ad(Xn ).

Proof. We show this by checking all the properties of for Jordan–Checalley decomposition
as in Lemma A.1.3:
∼
1. ad(Xs ) is semisimple: take an isomorphism V −→ Cn so that Xs is diagonalized:
Xs ei = λi ei . Then adgl(V ) (Xs ) is diagonalizable for the base Eij :

ad(Xs )(Eij ) = λi − λj .

2. ad(Xn ) is nilpotent by Lemma A.1.4;

3. [ad(Xs ), ad(Xn )] = ad[Xs , Xn ] = 0.

154
A.2 Topological spaces
Basic definitions
Definition A.2.1. Let X be a topological space.
• We say X is Hausdörff if for any two points p ̸= q ∈ X, there are open neighborhoods
Up , Uq contains p, q respectively such that Up ∩ Uq = ∅;

• X is second-countable: there is a countably basis for the topology of X;

• X is compact,
S
S if for any open cover X = i∈I Xi , there is a finite subset J ⊂ I such
that X = j∈J Uj .

• X is locally compact, if for any point x ∈ X, there are neighborhood U of x with

compact closed closure Ū .
In this paper, if nothing is mentioned, a topological space always means Hausdörff, second
countable, and locally compact.
Definition A.2.2. Let X be a topological space.
1. A path on X from a point x to a point y is a continuous map γ : [0, 1] −→ X such that
γ(0) = x, γ(1) = y.

2. We say that two paths γ0 and γ1 on X from point x to y are in the same homotopy
class if γ0 can be continuously deformed to γ1 . More precisely, there is a continuous
map
F : [0, 1] × [0, 1] −→ M
such that for any s, F (0, s) = x, F (1, s) = y and that F (t, 0) = γ0 (t) and F (t, 1) =
γ1 (t).

3. For any two points x, y ∈ X, we let Π(X, x, y) be the set of homotopy classes on X
from x to y. Then we have an inverse map and a composition for any three points
p, q, r ∈ M :

i : Π(X, x, y) −→ Π(X, y, x) : γ 7→ (γ −1 : t 7→ γ(1 − t)),

m : Π(X, x, y) × Π(X, y, z) −→ Π(X, x, z), (γ1 , γ2 ) 7→ γ1 · γ2

given by where (
γ1 (2t) if t ∈ [0, 1/2],
γ1 · γ2 (t) =
γ2 (2t − 1) if t ∈ [1/2, 1].

Definition A.2.3. Let X be a topological space.

1. The path connectedness defines an equivalent relation on X. Each equivalent class is
called a connected component, and the set connected component is denoted by π0 (X).

155
 2. We say that X is connected if π0 (X) has one element.

3. For any x ∈ X, let π1 (X, x) denote Π(X, x, x). Then π1 (X, x) has a group structure
with above compositions i, m with unit x
e being the constant path in Π(X, x, x). We call
π1 (X, x) the fundamental group of X.

4. We say that X is simply connected if X is connected and π1 (X, x) is trivial.

5. We say that X is locally simply connected if for any point x ∈ X and any open neigh-
borhood U of x in X there is a neighborhood V of x included into U such that V is
simply connected. .

If X is connected then for any x, y connected by a path γ, we have an isomorphism

∼
π1 (X, x) −→ π1 (X, y) : g 7→ γgγ −1 .

Thus the simply connectedness is independent of x.

Example A.2.1. The Euclidean space R2 is simply connected while R2 − {(0, 0)} is not. Its
fundamental group is isomorphic to Z. But when n > 2, both Rn and Rn − {0} are simply
conected. For any n, any convex subset of Rn is simply connected, especially the unit balls
in Rn are simply connected.

Covering spaces
Definition A.2.4. Let X, Y be topological spaces.

1. A covering map is a continuous surjective map π : X → Y such that ∀y ∈ Y , there is

an open neighborhood U of y, such that π −1 (U ) is disjoint union of open sets V ⊂ X
∼
with V −→ U homeomorphisms. In this case the space X is called a covering space,
and Y the base space.

2. A covering π : X −→ Y of connected topological space is called a universal cover of Y

if X is simply connected.

We can see that the covering map is a local homeomorphism from the definition. The
following gives a functorial description of universal covers. We consider the pointed topop-
logical space: a pair (X, x) of manifold X with a point x ∈ X. A map of pointed topological
spaces f : (X, x) −→ (Y, y) means a map f : X −→ Y such that f (x) = y.

Theorem A.2.5 (Universal lifting). Let π : (X, x) −→ (Y, y) be a covering map of connected
pointed topological space. Let α : (Z, z) −→ (Y, y) be a continuous map with Z simply
connected. Then there is a unique continuous map αe : (Z, z) −→ (X, x) such that α = π ◦ α
e.

Sketch of proof. First we consider the special case (Z, z) = ([0, 1], 0). Then the map α defines
a path on (Y, y). We want to lift this path. Let [0, a) be the maximal interval which can
ea : [0, a) −→ X uniquely. Assume a ̸= 1. Let U be a neighborhood of
lifted to a path α

156
α(a) ∈ Y so that π −1 U is a disjoint union of Vi ’s. Let ϵ > 0 be a small number so that α−1 U
includes [a − ϵ, a + ϵ]. Then exactly one of them Vi contains of α e[a − ϵ, a]. We than extend
this to [1 − ϵ, a + ϵ] by homeomphism π : Vi −→ U . This is also unique. Thus we extend α e
to [0, a + ϵ]. This is a contradiction. Thus a = 1. The same argument above will extend the
map to the point a = 1.
Second, we consider the case Z = ([0, 1] × [0, 1], (0, 0). This can be done by applying
horizontal or vertical intervals [0, 1] × [0, 1].
For general case, for any p ∈ Z, let γ be a path connecting z and p. Then we can lift γ
uniquely γ e on (X, x). So we define α(p) = γ e(1). We claim that this definition does not depend
′
on the choice of γ. For different γ , they are connected by homotopy F : [0, 1] × [0, 1] −→ Z.
Then we consider the composition β = α ◦ F : [0, 1]2 −→ Y . Then apply the above map to
get a map βe : [0, 1]2 −→ X lifting β. Notice that β(1, s) = α(p). So β(1, e s) for each s is a
lifting of α(p). As all liftings of α(p) in Y are in a disjoint union of opens, these points must
be constant. Thus β(1,e s) is a constant. So

γ(1) = β(1, e 1) = γ ′ (1).

e 0) = β(1,

We will construct universal coverings for topological spaces that are connected and locally
simply connected.

Theorem A.2.6. Let (X, x) be a pointed topological space that is connected and locally
simply connected.

1. let (X,
e xe) the pointed set of homotopy classes of path [γ] on X starting from x:
a
Xe= Π(X, x, y), x
e = constant curve valued at x
y∈X

The map [γ] −→ γ(1) defines a map π : (X,

e xe) −→ (X, x). If X is simply connected,
then this is bijective.

2. for any y ∈ X, a simply connected neighborhood Uy of y, and a point ye ∈ X, e we

construct a neighborhood Uye homeomorphic to Uy as follows: let γ be a path connecting
x and y representing ye, then

Uye = {[γ] · [δ] : δ a path in Uy starting from y}.

Then X
e has a unique topological structure so that the maps

Uye −→ Uy

are homeomorphic.

157
 3. The Xe is simply connected, and every path γ on (X, x) has a unique lifting γ
e on (X,
e xe)
with end point [γ] ∈ X.
e

4. The covering π : (X, e) −→ (X, x) is a universal covering of X with an action of

e x
π1 (X, x) on X by:

[δ][γ] = [δ · γ] : [γ] ∈ X,
e δ ∈ π1 (X, x).

Moreover, π1 (X, x) is isomorphic to the group of homeomorphisms g : X −→ X such

that π(gy) = π(y) for all y ∈ X,
e and π : X e −→ X induces a bijection:
∼
e −→
π1 (X, x)\X X.

Proof. Everything is clear from construction except part 3. This follows from the universal
lifting theorem A.2.5 applying to the covering (X, e) −→ (X, x) and the path γ : [0, 1] −→ X.
e x
We will construct an explicit lifting γe : [0, 1] −→ Xe of γ as follows. For any s ∈ [0, 1] define
a path γs by
γs (t) = γ(st), t ∈ [0, 1].
Then the map s 7→ [γs ] define a path γ e on X e starting from point xe = [γ0 ] and ending at
[γ] = [γ1 ]. To prove simply connectedness, let [γ] ∈ π1 (X, x
e e) be an element represented by
a loop γ on (X, x
e e). Let δ be the image of γ on X: then γ is a unique lifting of δ. Thus its
end point x e = [δ]. This shows that [δ] is , so is [γ].
Example A.2.2. Consider Pn (R) of lines in Rn . Since each line is generated by exactly two
vectors with norm 1, we have have an isomorphism:

Pn (R) = S n / ± 1.
∼
If n = 1, then S 1 / ± 1 −→ S 1 . Thus π1 (P1 (R)) = Z. If n > 1, S n is simply connected. Thus
π1 (Pn (R)) = Z/2Z.

A.3 Manifolds
Differential manifolds
Definition A.3.1. Let M be a topological space. We say M is an n-manifold if For any
∼
p ∈ M , there exists an open neighborhood U of p and a homeomorphism ϕ : U −→ Ũ with
Ũ ⊂ Rn an open subset. We call (U, ϕ) a chart on M . If ϕ(p) = 0 we say the chart is
centered at p.

Definition A.3.2. Let U, V be open subsets of Euclidean spaces Rm and Rn . We say a

smooth map f : U → V is a diffeomorphism if f is bijective and has a smooth inverse map.

158
Definition A.3.3. Given M a topological n-manifold. We define an atlas for M to be a
collection of charts A = (Uα , ϕα ) such that M = α Uα . An atlas is smooth if any of two
S
charts are smoothly compatible: ∀α, β,
∼
ϕβ ◦ ϕ−1
α : ϕα (Uα ∩ Uβ ) −→ ϕβ (Uα ∩ Uβ )

is a diffeomorphism.
Two atlases A = (Uα ) and A ′ = (Uα′ , ϕα′ ) are called (smoothly) equivalent if for any
α, α′ ,
ϕα′ ◦ ϕ−1
α : ϕα (Uα ∩ Uα′ ) −→ ϕα′ (Uα ∩ Uα′ )

is a diffeomorphism.
Definition A.3.4. A smooth manifold is a topological manifold M with an equivalent calss
of the smooth atlas. If any smooth atlas that is compatible with atlas A is contained in A ,
we say A is maximal. A smooth structure on M is a maximal atlas on M .
Example A.3.1 (Euclidean spaces). The Euclidean space Rn is a smooth n-manifold with
smooth structure (Rn , Id). We call this the standard smooth structure. Any open subset
U ⊂ Rn is also a smooth manifold given by (U, Id).
Example A.3.2 (Finite dimensional vector spaces). Let V be any finite-dimensional vector
space over R. Then any norm on V determines a topology. Any basis (e1 , · · · , en ) of V
determines a homeomorphism
n
X
ϕ : Rn −→ V, (x1 , · · · , xn ) 7→ xi ei .
i=1

So the atlas (V, ϕ−1 )defines a smooth structure independent of the choice of basis
Example A.3.3 (General linear groups). The group GLn (C) which is an open subset of
2 ∼ 2 2
Cn −→ R2n is a smooth manifold. The group GLn (R) which is an open subset of Rn
is a smooth manifold
Example A.3.4 (Projective spaces). RPn is a topological n-manifold under the quotient topol-
ogy determined by the natural map
π : Rn+1 − {0} −→ Pn
sending x to the line Rx through x and 0.
• Let Ũi ⊂ Rn+1 − {0} be the set where xi ̸= 0. Then for any x ∈ Ũi , π −1 (π(x)) ⊂ Ũi .
Let Ui = π(Ũi ), we have a cover {Ui } for RPn .
• Define
x1 xi−1 xi+1 xn+1
ϕi : Ui −→ Rn , R(x1 , · · · , xn+1 ) 7→ ( ,··· , , ,··· , ).
xi xi xi xi
This map is well defined and continuous had has inverse map given by
ϕ−1
i (y1 , · · · , yn ) = (y1 , · · · , yi−1 , 1, yi+1 , · · · yn ).

159
So we have atlas (Ui , ϕi ) and they are all smoothly compatible: suppose that I > j, compute
y1 yj−1 yj+1 yi−1 1 yi+1 yn
ϕj ◦ ϕ−1
i (y1 , · · · , yn ) = ( ,··· , , ,··· , , , , · · · , ),
yj yj yj yj yj yj yj

Which is a diffeomorphism

Definition A.3.5. Let M be a smooth manifold. A function f : M −→ R is said to be

smooth if for every smooth chart (U, ϕ) on M , the composition map f ◦ ϕ−1 is smooth on
the open subset ϕ(U ) ⊂ Rn . Let C ∞ (M ) denote the space of smooth functions from M to R.
This is an R-algebra.

Definition A.3.6. Let M, N be smooth manifolds. A map F : M → N is said to be smooth

if for any smooth charts (U, ϕ) on M and (V, ψ) on N , the composition map

ψ ◦ F ◦ ϕ−1 : ϕ(U ∩ F −1 (V )) → ψ(V )

is smooth.

Exercise A.3.7. Let F : M → N be a map between smooth manifolds. Suppose that each
point has a neighborhood U such that F |U is smooth. Show F is smooth.
Exercise A.3.8. Any composition of smooth maps between manifolds is smooth.
Exercise A.3.9. The quotient map π : Rn+1 − {0} −→ RPn is smooth.
Exercise A.3.10. Let F : M −→ N be a map between two smooth manifolds. Then F is
smooth if and only if for any smooth function ϕ ∈ C ∞ (N ), the function F ∗ (ϕ) := ϕ ◦ F on
M is smooth. Thus we obtain an homomorphism of R-algebra:

F ∗ : C ∞ (N ) −→ C ∞ (M ).

Tangent spaces
When studying calculus in Euclidean spaces, a function f (x) on R can be approximated by
its tangent line. We want to make a similar approximation on a smooth manifold, so one has
to define the derivation and tangent spaces for manifolds. We construct the tangent space
at a point first; then, the tangent bundle is just the union of tangent spaces.
The algebraic way to define the derivation of smooth functions on manifolds is very
similar to the definition of the derivation of functions on Rn :

Definition A.3.11. Let M be a smooth n-dimensional manifold and p ∈ M . A deriva-

tion(also called tangent vector) at p is a linear map D : C ∞ (M ) −→ R satisfying

D(f g) = f (p)D(g) + g(p)D(f ), f, g ∈ C ∞ (M ).

The set of all derivations of C ∞ (M ) at p, denoted Tp M is a vector space, called the tangent
space of M at p.

160
Theorem A.3.12. Let M = R, and a ∈ R. Then for any f ∈ C ∞ (M ), the tangent space
of M at a is just the space : R{ dtd |t=a }.
Proof. It is clear that c dtd |t=a is a derivation at p for any c ∈ R. Now we want to show that
every derivation has this form. Let D be a derivation at a. Assume that D(x − a) = c. We
want to show that D = c dtd . Let f be any smooth function on R. Using taylor expansion we
now that near a, f has an expression

f (x) = f (a) + f ′ (a)(x − a) + (x − a)2 g(x)

where g is a smooth function. Now we apply D to this equation to get

Df = D(f (a)) + f ′ (a)D(x − a) + D((x − a)2 g(x)).

It is clear that by Lebnitz rule, D(f (a)) = f (a)D(1) = 0, and D((x − a)2 g(x)) = 0. Thus
d
Df = f ′ (a)D(x − a) = cf ′ (a) = c .
dt

Exercise A.3.13. Consider the Euclidean space Rn and point a ∈ R. Show that the tangent
space Ta Rn at a can be identified with the set Rn by the isomorphism
X
D : Rn −→ Ta Rn , Dv = ai ∂ x i , ∀v = (a1 , · · · , an ) ∈ Rn .
i

Exercise A.3.14. Let M be a smooth n-manifold and p ∈ M be a point. Show that Tp M is

of dimension n.
Definition A.3.15. Given two smooth manifolds M, N and a smooth map F : M → N , the
differential of F at p ∈ M is

dF : Tp M → TF (p) N, D 7→ (f 7→ D(f ◦ F )),

where f ∈ C ∞ (N ).
Exercise A.3.16. Let f : M → R be a smooth function and p ∈ M . Show that that the
differential map dfp : Tp M → Tf (p) R = R · dtd is given by

d
dfp (D) = D(f ) .
dt
Definition A.3.17. Let M be a smooth n-dimensional manifold and I an interval of R. A
smooth curve of M parametrized by I means a smooth map γ : I −→ M .
For a smooth curve γ : I −→ M , and any t ∈ I, we define a tangent vector γ ′ (t) ∈
Tγ(t) (M ) by
∂(f ◦ γ)
γ ′ (t)f = .
∂t
161
Exercise A.3.18. Let M be a smooth manifold , and p ∈ M a point, and v ∈ Tp M a
tangent vector. Show that there is an open interval I of R containing 0, and a smooth curve
γ : I −→ M such that γ(0) = p and γ ′ (0) = v.
Definition A.3.19. Let F : M −→ N be a smooth morphism of smooth manifold. Let
p ∈ M and dFp : Tp M −→ Tf (p) M the induced differential map on tangent spaces.
1. F is called a diffeomorphism at p if dfp is bijective;
2. F is called an immersion at p, if dFp is injective;
3. F is called a submersion at p if dFp is surjective.
Theorem A.3.5 (Inverse function theorem). Let F : M −→ N be a smooth morphism of
smooth manifold of dimension m, n. Let p ∈ M and dFp : Tp M −→ TF (p) M the induced
differential map on tangent spaces. The following statements hold:
1. if F is a diffeomorphism at p, then there exists open neighborhoods U of p and V of
∼
F (p) such that F induces a diffeomorphism U −→ V ;
2. if F is an immersion at p, then there exists open neighborhoods U of p, and V of F (p),
∼
and an diffeomorphism G : V −→ U × W with W a neighborhood 0 in Rn−m such that
the composition G ◦ F is given by x 7→ (x, 0);
3. if F is a submersion at p, then there exists open neighborhoods U of p, and V of F (p),
∼
and an diffeomorphism G : V × W −→ U with W a neighborhood 0 in Rm−n such that
the composition F ◦ G is given by (x, y) 7→ x.
Sketch of proof. Using local charts, it reduces to the case that M and N are open neigh-
borhoods of 0 in Rm and Rn respectively with p = 0 and F (p) = 0. Then we can write
F = (f1 (x1 , · · · , xm ), · · · , fn (x1 , · · · , xm )). The differential map dF is given by linear map
Rm −→ Rn given by Jacobian matrix J = (∂fi /∂xj ). Then part one is the classical inverse
function theorem.
∼
For part two, we take a decomposition T0 N = dF (T0 Rm ) ⊕ L with L −→ Rn−m a linear
subspace. Then we define a new map
Fe : M × L −→ N, Fe(x, y) = F (x) + y.

Then dFe is bijective at 0. Thus we can apply part one to get that Fe is diffeomorphism in a
small neighborhood of U × W of M × L.
∼ ∼
For part three, we take an isomorphism Rm −→ T0 M −→ T0 N × L with L −→ Rm−n so
that dF is the composition of this isomorphism with the first projection. Then we define a
new map Fe : M −→ N × L by x e = (F (x), p2 (x)). Then dFe is bijective at 0. Then we can
apply part one to get that Fe is an diffeomorphism in a small neighborhood:
∼
Fe : U −→ V × W.

162
 2
Exercise A.3.20. Let M = Rn consider as the space of n × n-matrices. Consider the map
f = det : M −→ R as determinant map.
1. Show that det is submission for any point A ∈ M with det A ̸= 0.

2. Show that for any nonzero number λ ∈ R, the set

{A ∈ M, det A = λ}

has a smooth manifold structure.

Tangent bundles and Vector fields

Definition A.3.21. Let M be`a smooth n-dimensional manifold. As a set, The tangent
bundle T M of M is the union p∈M Tp M with the projection map

π : T M −→ M, (p, δ) 7→ p.

The tangent bundle T M is a 2n-dimensional manifold. The charts on T M are defined as

follows:
∼
Let ϕ : U −→ Ũ be a char on M . Then
∼
a a
π −1 (U ) = T U = Tp U −→ Tp Ũ = T Ũ = Ũ × Rn
p∈U p∈Ũ

where the isomorphism is given by (p, δ) 7→ (ϕ(p), dϕp (δ)). Then the projection π is smooth.
Definition A.3.22. Let M be a smooth manifold. A vector field on M is a smooth map
V : M → T M such that π ◦ V = IdM . Thus Vp := V (p) ∈ Tp M for any p ∈ M . So we also
call V a section of T M , denote the space of vector fields on M by Γ(T M ).
Remark A.3.1. Let V, V ′ be vector fields on M and p ∈ M . Since we have (V + V ′ )(p) =
V (p) + V ′ (p), and there is a zero lement 0 : p 7→ 0 ∈ Tp M , Γ(T M ) is a R-vector space.
In geometric way, Let ϕ : U → Ũ ⊂ Rn be a chart of M , one can view a point x ∈ U as
a point in Rn . Then x has a local coordinate (x1 , · · · , xn ) on U . The value Vp for p ∈ U can
be written
∂
Vp = V i (p) i |p ,
∂x
for some numbers V (p). This defines n functions V i : U → R called the component functions
i

of V with respect to the given chart.

∼ ∼
Example A.3.6. Let M ⊂ Rn , then there is an isomorphism M × Rn −→ T M −→ given by
(p, v) 7→ (p, ∂v ).
Here coms to the definition of maps between vector fields:
Definition A.3.23. Let F : M → N be a smooth map between smooth manifolds. The dif-
ferential of F , denoted dF is the smooth map T M −→ T N given by dF (p, D) = (F (p), dFp ).

163
Definition A.3.24. Let M be a smooth manifold. An map D : C ∞ (M ) → C ∞ (M ) is a
derivation if D(f g) = D(f )g + f D(g), ∀f, g ∈ C ∞ (M ). Denote the set of derivations on M
by Der(C ∞ (M )). Note that Der(C ∞ (M )) is also an R-vector space.
Exercise A.3.25. Define an operation in Der(C ∞ (M )) by

[D1 , D2 ] = D1 D2 − D2 D1 , ∀D1 , D2 ∈ Der(C ∞ (M )).

Show that [D1 , D2 ] ∈ Der(C ∞ (M ))

Proposition A.3.26. Γ(T M ) is isomorphic to Der(C ∞ (M )) given by the isomorphism

φ : Γ(T M ) / Der(C ∞ (M )) .
 /
V (DV : f 7→ (p 7→ dfp (Vp )))

Proof. Let ϕ : U → Ũ be a local chart on M . If the condition for Ũ = M is true, then

compose with ϕ−1 it is also true for U . Since M is smooth we can cover M with smooth atlas,
∼
then reduce to the case when M is an open subset of Rn . Then Γ(T M ) −→ C ∞ (U, Rn ).
Define
ψ : Der(C ∞ (U )) / C ∞ (U, Rn ) ,
D / (p 7→ (D(e1 )(p), · · · , D(en )(p))
where (e1 , · · · en ) is a basis for Rn . And show that ψ ◦ φ = Id.
Corollary A.3.27. One can define an operation [, ] on Γ(T M ) by [V, V ′ ] := [DV , DV ′ ].

Integral curves
Let M be a smooth manifold and V ∈ Γ(T M ).
Definition A.3.28. An integral curve of V is a smooth curve γ : J → M defined on an
open subset J ,→ R such that

γ ′ (t) = Vγ(t) , ∀t ∈ J.
∂
Example A.3.7. Let M = R2 . Let V = x ∂x be the first coordinate vector field on M . Then
the curve γ(t) = (x(t), y(t)) for J is an integral curve of V if and only if
∂ ∂ ∂
x(t) |γ(t) = γ ′ (t) = x′ (t) |γ(t) + y ′ (t) |γ(t) .
∂x ∂x ∂y
This is equivalent to the system of two equations

x′ (t) = x(t), y ′ (t) = 0.

Thus we have x(t) = aet , y = b for some constants a, b. In the other words, the integral
curves all have the forms γa,b (t) = (aet , b) for real numbers a, b.

164
Example A.3.8. Let M be a smooth n-manifold embedded in Rn . Let γ : J → M be any
smooth curve. Writing γ(t) = (f1 (t), · · · , fn (t)), the condition γ ′ (t) = Vγ(t) such that γ is
integral curve of vector field V can be written in local coordinates V = (V i (x1 , · · · , xn ) ∂x∂ i )
on open subset U ⊂ M , where V i are smooth functions U → Rn
∂ ∂
fi′ (t) i
|γ(t) = V i (γ(t)) i |γ(t) ,
∂x ∂x
which induces to the system of ordinary differential equitions

(A.3.1) f1′ (t) = V 1 (f1 (t), · · · , fn (t)),

..
(A.3.2) .
′ n
(A.3.3) fn (t) = V (f1 (t), · · · , fn (t)).

The fact is that this system has a unique solution at least for t in a small time interval
(−ϵ, ϵ), satisfying any initial condition of the form (f1 (0), · · · , fn (0)) = (a1 , · · · , an ) ∈ U .

Theorem A.3.9. Let M be a smooth manifold, V ∈ Γ(T M ). Then

1 There is an open inteval J with 0 ∈ J and integral curve γ : J → M for V such that
γ(0) = p.

2 The integral curve described in 1 is unique up to identification as germs of curves at p.

That shows the integral curves exist and are well-defined.

Exercise A.3.29. Let M = R2 and V ∈ Γ(T M ). If V(x,y) = (y, −x). Construct all integral
curves for this vector field.

Definition A.3.30. We say that V ∈ Γ(T M ) is complete if ∀p ∈ M there is an integral

curve γ : R → M for V with starting point p.

Theorem A.3.31 (Frobenius theorem). Let M be a connected smooth manifold and E a

vector subbundle of T M . Assume that the space of global sections Γ(M, E) is closed under
Lie bracket, i.e., Γ(M, E) is a Lie subalgebra of Γ(M, T M ) = Der(M ). Then for any point
p, there is a unique closed submanifold N (p) of M in a neighborhood of p in M such that
for any q ∈ Np ,
Tq N (p) = Ep .

A.4 Measures and Hilbert spaces

Measure via distributions
Let X be a topological space. By our convention, we always assume that X is Hausdörff,
second countable, and locally compact. Thus X is the union of relatively compact open

165
subsets. Let C(X, R) be the space of continuous functions on X, and Cc (X) the subspace
of C(X) consisting of compactly supported functions. Then Cc (X, R) has a norm
∥f ∥ := sup |f (x)|.
x∈X

More precisely, for K a compact subset of X, define CK (X, R) to be the space of contin-
uous functions with support in K, then CK (X, R) is a complete with respect to norm ∥f ∥,
and Cc (X, R) is the union of CK (X, R).
Definition A.4.1. By a distribution on X, we mean a linear map
µ : Cc (X, R) −→ R
such that for any compact subset K of X, there is a constant cK such that
|µ(f )| ≤ cK ∥f ∥, ∀f ∈ CK (X, R).
Let D(X, R) denote the space of distributions on X. We write
Z
µ(f ) = f dµ.
X

Definition A.4.2. A distribution µ is called a (Rondon) measure, if µ(f ) ≥ 0 if f ≥ 0. Let

D+ (X, R) denote the subset of positive distributions.

Distributions via differential forms

Let X be a smooth oriented manifold; then for any ω ∈ C(X, Ωn ), we want to define a
distribution via integral Z
f ω.
X
If ω is non-negative, this integral defines a measure on X. We can modify the integral
without orientation using |ω| as it is defined over every chat.

Measures via Borel sets

Definition A.4.3 (Borel subset and measures). Let X be a topological space.
1. The Borel σ-algerba B(X) of X is a set of subsets of X generated by open subsets
of X under following operations: countable unions, countable intersections, and com-
plements. The elements of B(X) are called Borel subsets(or Borel measurable sets of
X).
2. A Borel measure on X is a function
µ : B(X) → [0, +∞)
P
such that µ(∅) = 0 and µ(∪n∈Z An ) = n∈Z µ(An ) for any pairwise disjoint An ∈
B(X).

166
 3. A Borel measure µ is called regular if

(a) For any compact subset K ⊂ X, µ(K) < ∞.

(b) For any A ∈ B(X), then
open
µ(A) = inf{µ(U )| A ⊂ U ,→ X} = sup{µ(K)| K ⊂ U compact}.

The first equality is known as inner regularity, the second as outer regularity.

Let M (X) denote the set of regular Borel measures on X. For each µ ∈ M (X), we can
define theP
integrals for some functions on X by approximation by simple functions of the
form f = ni=1 ai 1Ui where we define
X
µ(f ) = ai µ(Ui ).
i

We may identify Rondon measures D+ (X, R) with regular Borel measures M (X). For
any µ ∈ M (X), we can define a distribution on X as follows:

1. if f ≥ 0, then we define

µ(f ) = inf{µ(ϕ) : ϕ is simple, ϕ ≤ f }.

2. for general f , write f ± = max(±f, 0). Then define

µ(f ) = µ(f + ) − µ(f − ).

Conversely, for any positive distribution µ ∈ D+ (X, R), we define a Borel measure as
follows:

1. if U is a open subset of X, we define

µ(U ) = sup{µ(f ) ∈ Cc (U ) : f ≤ 1}.

2. if A ∈ B(X), we define

µ(A) = inf{µ(U ) : U is open and U ⊃ A}.

Remark A.4.1 (Lebesgue measure). On Rn , the standard differential form ω = dx1 · · · dxn
defines usual Lesbesgue measure λ on Borel sets of Rn . We can extend this set to the
Lesbesque σ-algebra consisting of subset A of Rn such that for any subset B of

λ(A) = λ(B ∩ A) + λ(A \ B).

167
Hilbert spaces
Hilbert spaces and bounded operators
Definition A.4.4 (Hilbert space). 1. A Hilbert space H is a hermtian space which is
complete in the sense that any Cauchy sequence has a unique limit.

2. A linear map φ : H1 −→ H2 is called an isomorphism if it is bijective and preserves

the hermitian forms.
Example A.4.1. The space
( ∞
)
X
ℓ2 = (· · · , xi , · · · ) | ∥xi ∥2 < ∞, xi ∈ C
i=0

is a Hilbert space over C. And for x = (· · · , xi , · · · ), y = (· · · , yi , · · · ), we can define the

inner product as
X ∞
⟨x, y⟩ = xi ȳi .
i=0

In fact any Hermitian space with an infinite countable dense subset is isomorphic to ℓ2 .
Definition A.4.5 (Hilbert basis). A Hilbert basis (or orthonormal basis) of a Hilbert space
H is a family {ei } satisfying ⟨ei , ej ⟩ = δij such that ⊕Cei is dense in H .
Remark A.4.2.
P Let HPbe a hilbert space with a Hilbert basis {ei } and x, y ∈ H . Suppose
that x = i ai ei , y = i bi ei , then we have
X X X
⟨x, y⟩ = ⟨ ai e i , bi e i ⟩ = ai b¯i .
i i i

Example A.4.2. The main examples for representations is as follows: Let (X, µ) be a topo-
logical space with a Borel measure dµ. Let Cc (X) denote the space of continuous functions
with compact support. Define a hermitian structure on Cc (X) by
Z
⟨f, g⟩ = f (x)g(x)dµ(x).
X

Then for any f , ⟨f, f ⟩ ≥ 0. Let Cc (X)0 be the subspace of Cc (X) such that ⟨f, f ⟩ = 0. Then
we have a positive definite hermtian form on the quotient

Cc (X)/Cc (X)0 .

After completing this space, we get L2 (X). If X is a manifold, then L2 (X) is isomorphic to
ℓ2 . In particular, L2 (X) always has a Hilbert base consisting of many vectors.
The main point of the theory of Hilbert space is about the set of bounded operators
between Hilbert spaces.

168
Definition A.4.6. Let T : H1 → H2 be a linear operator.

1. T is said to be bounded if there is a constant C > 0 such that

∥T v∥
≤ C, ∀v ∈ H1 .
∥v∥

2. We write B(H1 , H2 ) to denote the space of all bounded operators from H1 to H2 , and
define the norm of the operator:

∥T ∥ := sup ∥T (x)∥.
x∈H1 ,∥x∥=1

3. We define the dual space of a Hilbert space H to be H ∗ = B(H , C).

The B(H1 , H2 ) equipped with norm ∥·∥ is a topological space. This space is in fact
complete in the sense that for any Cauchy sequence φi in B(H1 , H2 ) covergent to a limit
φ∞ ∈ B(H1 , H2 ).

Theorem A.4.7 (Riesz’s representation theorem). Let H be a Hilbert space. Then the map

H → H ∗, x 7→ (y 7→ ⟨y, x⟩)

is a R-linear isomorphism, and an isometry. In other words, we have an isomorphism

∼
H¯ −→ H ∗ .

In particular, H ∗ is a Hilbert space.

Adjoint operators and unitary operators

Using theorem, we can define an adjoint bijection:

∗ : B(H1 , H2 ) −→ B(H2 , H1 ) : φ 7→ φ∗

by
⟨φ(x), y⟩H2 = ⟨x, φ∗ y⟩H1 , ∀x ∈ H1 , y ∈ H2 .

Definition A.4.8. Let H be a Hilbert space and T : H −→ H be a bounded operator.

1. T is called a self-adjoint operator if T ∗ = T ; in other words, for any x, y ∈ H ,

⟨T x, y⟩ = ⟨x, T y⟩.

Let Herm(H ) denote the set of self-adjoint operators.

169
 2. T is called a unitary operator if T ∗ T = T T ∗ = idH ; i.e., for any x, y ∈ H ,
⟨T x, T y⟩ = ⟨x, y⟩.
Let U (H ) denote the group of all unitary operators in B(H , H ).
The U (H ) is a subset of B(H , H ). Thus it is a topological group.
Exercise A.4.9. Let H = ℓ2 = L2 (N) with othornormal basis ei , i ∈ N. Then each linear
endomorphism φ : H −→ H is represented by an infinite matrix Aij ∈ MN (C) such that
X
φ(ei ) = Aij ej
ij

1. Describe the C-algebra of B(H , H ) of bounded endomorphisms in terms of matrices;

2. Describe the group U (H ) of unitary operators in terms of matrices;
3. Describe the space of self-adjoint operators in terms of matrices.

Compact operators
Definition A.4.10. We say an operator T ∈ B(H1 , H2 ) is compact if for any bounded
sequence {xi } in H1 , the sequence {T xi } has a convergent subsequence. We denote the set
of compact operators in B(H1 , H2 ) by K(H1 , H2 ).
Example A.4.3. Let ξ = i vi ⊗ wi ∈ H1 × H2 . Define an operator φξ ∈ B(H1 , H2 ) by
P
X
φξ (v) = < vi , v > wi , v ∈ H1 .
i

So φξ is compact if Im(φξ ) ,→< w1 , · · · , wn > is finite dimensional.

Remark A.4.3. Any compact operators in B(H1 , H2 ) can be approximated by elements in
H1 ⊗ H2 :
H\ ∼
1 ⊗ H2 = K(H1 , H2 ).

The following spectral theorem for compact self-adjoint operators in Hilbert space will
play an important role in proving the Peter-Weyl theorem.
Theorem A.4.4 (Spectral theorem). Let H be a Hilbert space, and T ∈ B(H ) a self-
adjoint and compact operator. Then we have
1. Spec(T ) ⊂ R.
2. For any distinct λ, µ ∈ Spec(T ), Vλ ⊥ Vµ .
3. For any 0 ̸= λ ∈ Spec(T ), dim Vλ < ∞.
4. Spec(T ) is at most countable, only possible limit point is 0.

λ∈Spec(T ) Vλ is dense in H .
L
5.

170