Introduction to Lie Algebras and Their Representations

Prof Ian Grojnowski (Michaelmas 2010)

Unofficial lecture notes - University of Cambridge
By Robert Laugwitz and Henning Seidler

Contents

1 Introduction 2
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Definition of Lie algebras and basic properties . . . . . . . . . . . . . . . . . . . . 4

2 Representations of sl2 8
2.1 Classification of sl2 representations . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Consequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 Structure and Classification of Simple Lie Algebras 18

3.1 Linear algebra preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.2 Structure of semisimple Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . 22

4 Structure Theory 25

5 Root Systems 33

6 Existence and Uniqueness 43

7 Representations of Semisimple Lie Algebras 46

7.1 Classification of finite-dimensional representations . . . . . . . . . . . . . . . . . 46
7.2 The PBW theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
7.3 The Weyl character formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
7.4 Principal sl2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

8 Crystals 63
8.1 Semi-standard Young tableaux . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
8.2 Littelmann paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

Reading to complement course material

Jacobson, N. (1979), Lie algebras, Dover Publications.

Kac, V. (1994), Infinite dimensional Lie algebras, 3 edn, Cambridge University Press.

Kashiwara, M. (1995), On cystal bases, in ‘Representations of groups (Banff, AB, 1994)’, CMS
Conf. Proc., 16, Amer. Math. Soc., pp. 155–197.

1
Lie Algebras and Their Representations

1 Introduction
1.1 Motivation
Definition 1.1: A linear algebraic group is a subgroup of the general linear group GLn of n×n
matrices where the matrix coefficients fulfill certain polynomial equations.

Example 1.2: The upper triangle matrices

1 a12 · · ·
 
a1n
.. 
0 . . . . . .

 .   .
.
 .. · · · . . .

an−1,n 
0 ··· 0 1

Other examples are

SLn = {A ∈ Matn | det A = 1},

SOn = {A ∈ SLn | AAT = I},
On = {A ∈ GLn | AAT = I},
 
T T 0 In
SP2n = {A ∈ GLn | M A M A = I}, M = .
−In 0

Remark 1.3: There is an intrinsic characterization of linear algebraic groups as affine algebraic
groups, i.e. groups which are affine algebraic varieties and where multiplication and inverse are
morphisms of algebraic varieties.

Consider G = SLn . If
   
1 0 a b
g= +ε + higher order terms ∈ SL2 , |ε|  1,
0 1 c d

then
  
1 + εa εb
1 = det g = det + higher order terms
εc 1 + εd
= (1 + εa)(1 + εd) − ε2 cb + higher order terms
= 1 + ε(a + d) + higher order terms.

Thus, det g = 1 if and only if a + d = 0.

We can make this notion of vanishing higher order terms more precise by defining the dual
numbers as
E = C[ε]/ε2 = {a + bε | a, b ∈ C}.
This omits the structure of a ring. Consider G(E) = {A ∈ Matn (E) | A ∈ G ⊂ GLn },
the matrices over the dual numbers which satisfy the polynomial equations defining the linear
algebraic group G. E.g.
  
α β
SL2 (E) = α, β, γ, δ ∈ E : αδ − βγ = 1 .
γ δ

2 Robert Laugwitz & Henning Seidler

 1 Introduction

By letting ε 7→ 0 we obtain a map E → C which extends to a map

π : G(E) → G, A + Bε 7→ A.

Definition 1.4: We define the Lie algebra of G, denoted Lie(G), as the preimage
g := π −1 (I) = {X ∈ Matn (C) | I + εX ∈ G(E)}.

a b


Example 1.5: For example sl2 = { c d ∈ Mat2 (C) | a + d = 0}.

Remark 1.6: I + Xε represents an ‘infinitesimal change’ at I in the direction X, i.e. the germ
of a curve Spec[[ε]] → G.

Exercise 1.7: Show G(E) = T G, the tangent bundle to G, and g = TI G, the tangent space to
G at I.

Example 1.8:
(i) Let G = GLn = A ∈ Matn | A−1 exists . Then

n o
G(E) = Ã ∈ Matn (E) | Ã−1 exists
= A + Bε | A, B ∈ Matn (C), A−1 exists ,


as (A + Bε)(A−1 − A−1 BA−1 ε) = I. So Lie(GLn ) = Matn (C).

(ii) Let G = SLn (C). Then
det(I + εX) = det((δij + εxij )i,j )
= (1 + εx11 ) · ... · (1 + εxnn )
= 1 + εtr(X).
From this we conclude that
sln = {X ∈ Matn | tr(X) = 0} .

(iii) Let G = On (C) = A | AAT = I ,



⇒ g = X ∈ Matn (C) | (I + εX)(I + εX)T = I



= X ∈ Matn (C) | I + ε(X + X T ) = I



= X ∈ Matn (C) | X + X T = 0 .


Notice, that as 2 6= 0, we have tr(X) = 0, so this is also the Lie algebra of SOn , denoted
by son .

Remark 1.9: This leads us to the question, what structure we have in g coming from G being
a group? Note that in E we have (I + Aε)(I + Bε) = I + (A + B)ε, which has nothing to do with
multiplication. Multiplication is a map G × G → G. Consider instead the map G × G → G given
by to commutator in a group, (P, Q) 7→ P QP −1 Q−1 . If we look at this infinitesimally, we obtain
a map TI G × TI G → TI G, write P = I + aε and Q = I + Bδ, where ε2 = δ 2 = 0 but εδ 6= 0.
Remember that (I + aε)−1 = I − Aε. Then we have P QP −1 Q−1 = I + (AB − BA)εδ, which
is the “shadow” of multiplication we will use. So for the Lie algebra of an algebraic groups, we
define [A, B] = AB − BA, the Lie bracket of g.

Robert Laugwitz & Henning Seidler 3

Lie Algebras and Their Representations

Exercise 1.10: Show that:

(i) Show that (P QP −1 Q−1 )−1 = QP Q−1 P −1 implies [A, B] = −[B, A], for all A, B ∈ g (skew
symmetry).

(ii) Multiplication in G is associative implies

0 = [[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] (Jacobi identity).

Solution:

(i) Take P, Q as in 1.9, then as before

[B, A] ↔ QP Q−1 P −1 = (P QP −1 Q−1 )−1 = (I + (AB − BA)εδ)−1

= I + (−(AB − BA))εδ ↔ −[A, B],

and so [A, B] = −[B, A].

Remark: Since we already have [A, B] = AB − BA, we have the much easier and more
obvious proof [A, B] = AB − BA = −(BA − AB) = −[B, A].

(ii) By simple calculation, we obtain

[[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = [XY − Y X, Z] + [Y Z − ZY, X] + [ZX − XZ, Y ]

= XY Z − Y XZ − ZXY + ZY X + Y ZX − ZY X
− XY Z + XZY + ZXY − XZY − Y ZX + Y XZ = 0.

1.2 Definition of Lie algebras and basic properties

Definition 1.11: Let k be a field, char k 6= 2, 3. A Lie algebra g is a k-vector space equipped
with a bilinear map [·, ·] : g × g → g, the so-called Lie bracket, such that

(i) [X, Y ] = −[Y, X], skew symmetry, and

(ii) [[X, Y ], Z] + [[Y, Z], X] + [[Z, X], Y ] = 0, the Jacobi identity.

A subspace h ⊆ g is a Lie subalgebra if [h, h] ⊆ h, i.e. for all x, y ∈ h, [x, y] ∈ h.

The previously defined Lie algebras of an algebraic groups satisfy these properties, as shown
in 1.10.

Example 1.12: We have the following examples of Lie algebras:

(i) For any vector space V , we can let [·, ·] be the zero-map, i.e. [u, v] = 0 for all u, v ∈ V .
This defines an abelian Lie algebra (named like this because for the matrix commutator,
we have [A, B] = 0 if and only if A and B commute).

(ii) gln = Mat(n × n), or, for V a vector space, glV = End(V ).

(iii) sln = {A ∈ gln | tr(A) = 0} (where tr denotes the trace). Observe that for X, Y ∈ sln
we have tr(XY ) 6= 0 in general, but since tr(XY ) = tr(Y X) we get tr([X, Y ]) = 0. So
[A, B] = AB − BA has [·, ·] : Λ2 sln → sln .

(iv) son = {A ∈ gln | A + AT = 0}.

4 Robert Laugwitz & Henning Seidler

 1 Introduction

(v) sp2n = {A ∈ gl2n | JAT J T + A = 0}, where

 
1
. 
..


 0 

1
 
J =  .
 
 −1 
.
 
..
 
 0 
−1

(vi) b, the upper triangular matrices in gln (b stands for Borel).

(vii) h, the strictly upper triangle matrices in gln .

Exercise 1.13:

(i) Check directly that gln is a Lie algebra.

(ii) Check that the examples (iii)-(vii) are Lie subalgebras of gln .
 
∗ ∗
Note that, for example, is not a subalgebra of gln .
∗ 0

Exercise 1.14:

(i) Find algebraic groups, whose Lie algebras are those above.

(ii) Classify all Lie algebras of dimension 3 (or 2) as vector spaces. Note that the 1-dimensional
Lie algebras are all abelian algebras.

Definition 1.15: A representation of a Lie algebra g on a vector space V is a homomorphism

of Lie algebras ϕ : g → glV , i.e. a map ϕ : g → End(V ), such that

ϕ([x, y]) = ϕ(x)ϕ(y) − ϕ(y)ϕ(x), ∀x, y ∈ g.

We say g acts on V .

Example 1.16: If g ⊆ glV , then g acts on V , so the Lie algebras from Example 1.12 act
faithfully on k n .

Definition 1.17: If x ∈ g, we define ad(x) : g → g by ad(x)(y) = [x, y], this defines ad : g →

End(g).

Lemma 1.18: ad is a representation, called the adjoint representation.

Proof: The identity ad[x, y] = ad x ad y − ad y ad x follows from skew symmetry and the Jacobi
identity.

Definition 1.19: The center of g is Z(g) := {x ∈ g | ∀y ∈ g : [x, y] = 0} = ker ad. So g has

trivial center if and only if g embeds via ad into glg .

Robert Laugwitz & Henning Seidler 5

Lie Algebras and Their Representations


0 ∗
Example 1.20: h = is abelian, so maps to 0 in glh via ad, but h ⊆ gl2 also admits a
0 0
faithful representation by definition.

Theorem 1.21 (Ado): Any finite-dimensional Lie algebra over some field k is a Lie subalgebra
of gln for some n (i.e. admits a faithful finite-dimensional representation).

Example
 1.22: 
a b
sl2 = a, b, c ∈ C has a basis
c −a
     
0 1 1 0 0 0
e= , h= , f= ,
0 0 0 −1 1 0

with relations [e, f ] = h, [h, e] = 2e, [h, f ] = −2f . So a representation of sl2 on k n is a triple
E, F, H of n × n-matrices, such that [E, F ] = H, [H, E] = 2E, [H, F ] = −2F . This leads to the
question how we can obtain such representations?

Definition 1.23: If G is an algebraic group, then an algebraic representation of G on a vector

space V is a homomorphism of groups ρ : G → GLV defined by polynomial equations in the
matrix coefficients of G ⊆ GLn .

Again, we can substitute E = k[ε]/ε2 for K. Thus, we get a homomorphism of groups

G(E) → GLV (E). As ρ(I) = I, we have ρ(I + Aε) = I + ε(some function of A). Call this
function d ρ, so ρ(I + Aε) = I + ε d ρ(A), which defines a map d ρ : g → glV . This gives a functor

AlgRepG → RepLie(G) , ρ 7→ d ρ.

Exercise 1.24:
(i) d ρ is the derivative of ρ, evaluated at I, i.e. d ρ : TI G → TI GLV .

(ii) ρ : G → GLV is a group homomorphism. Therefore, d ρ : g → glV is a Lie algebra homo-

morphism, i.e. V is a representation of g.

Example 1.25: Let G = SL2 and let L(n) be the set of homogeneous polynomial of degree n
in variables x and y. Then L(n) has the basis xn , xn−1 y, . . . , y n , so dim L(n) = n + 1. SL2 acts
on L(n) by

ρn : SL2 → Aut(L(n)) = GLn+1 , (ρn (g)f )(x, y) = f (ax + cy, bx + dy),

a b


if g = c d ∈ SL2 , f ∈ L(n). In particular, we have
ρ0 the trivial representation,

ρ1 the standard 2-dimensional representation on k 2 ,

ρ2 here ac db acts, w.r.t. the above basis of L(2), by the matrix
 2
b2

a ab
2ac ad + bc 2bd .
c2 cd d2

6 Robert Laugwitz & Henning Seidler

 1 Introduction

It is left as an exercise, to verify that SL2 acts on L(n) via ρn . Let us now compute represen-
tations of sl2 on L(n):

Remark 1.26: GL2 acts on P1 , and on O(n), hence on Γ(P1 , O(n)) = S n k 2 , and that is where
these representations come from.

If we take the basis element e = ( 00 10 ), then ρn (I +εe)·xi y j = xi (εx+y)j = xi y j +εjxi+1 y j−1 ,

which says d ρn (e) · xi y j = jxi+1 y j−1 if j ≥ 1 (d ρn = 0 if j = 0). This proves the first equation
in the following exercise:

Exercise 1.27:

(i) For d ρn , with ρn as in 1.25, we have:

(
i j jxi+1 y j−1 if j ≥ 1
e(x y ) =
0 if j = 0,
(
ixi−1 y j+1 if i ≥ 1
f (xi y j ) =
0 if i = 0,
h(xi y j ) = (i − j)xi y j .
∂ ∂ ∂ ∂
Hence d ρn (e) = x ∂y , d ρn (f ) = y ∂x , d ρn (h) = x ∂x − y ∂y .

(ii) Check directly that these formulas give representations of sl2 on L(n).

(iii) Check directly that L(2) is the adjoint representation.

∂ ∂ ∂ ∂
(iv) Show that the formulas e = x ∂y , f = y ∂x , h = x ∂x − y ∂y give an (infinite-dimensional!)
representation on k[x, y].

(v) Let char k = 0. Show that L(n) is an irreducible representation of sl2 , hence of SL2 .

Example 1.28: Let G = C∗ , then g = Lie(G) = C with [x, y] = 0. A representation of g = C

on V corresponds to A ∈ End(V ), as a linear map ρ : C → End V is determined by A = ρ(1).
W ⊆ V is a submodule if and only if AW ⊆ W , and ρ is isomorphic to ρ0 : g → End(V 0 )if and
only if A and A0 are conjugate as matrices. Hence, the classification of representations of g is
given by the Jordan normal forms of matrices.
As any linear transformation over C has an eigenvector, there is always a 1-dimensional
subrepresentation of V . Therefore, V is irreducible if and only if dim V = 1. Also, V is
completely decomposable (i.e. breaks up into a direct sum of irreducible representations) if and
only if A is diagonalizable.
01
 
0 1
.. ..
Let A = . . , then the associated representation to A is indecomposable, but not
 

..
. 1
0
irreducible. The invariant subspaces are he1 i, he1 , e2 i, . . . , he1 , e2 , . . . , en i, but their complements
are no subspaces.
What about algebraic representations of G = C∗ ? Here, the irreducible representations
are ρn : G → GL1 = Aut(C), z 7→ (x 7→ z n x), n ∈ Z. Moreover, every finite-dimensional
representation is a direct sum of these representations.

Robert Laugwitz & Henning Seidler 7

Lie Algebras and Their Representations

Exercise 1.29: The functor ρ 7→ d ρ takes ρn to multiplication by n in C, and this is an

irreducible representation of C, but there are other irreducible representations, as we have seen
before.

Notice that g = (C, ·) is also the Lie algebra of G = (C, +), so it is not surprising that its
representations are different from the representations of C∗ . What is surprising, is the following:

Theorem 1.30 (Lie): The functor ρ 7→ dρ is part of an equivalence of categories AlgRepG ∼ =

RepLie(G) if G is a simply connected simple algebraic group. (E.g. for G = SLn , SOn , SP2n ).

Remark 1.31: Note that for algebraic groups, there is a different definition of simplicity. An
algebraic groups is simple if it does not contain any proper nontrivial normal connected closed
subgroup. Note for example, that for G a simply connected and simple algebraic group, the
center does not have to be trivial, but it is finite, e.g. Z(SLn ) = Cn , the cyclic group with n
elements.

Exercise 1.32: If G is an algebraic group, and Z is a finite central subgroup of G, then

Lie(G/Z) = Lie(G). I.e. the tangent space does not change if we identify central elements of an
algebraic group.

We have now also seen that the map AlgGp → LieAlg, G 7→ Lie(G) is not injective.

Exercise 1.33:
(i) Let Gn = C∗ n C, where C∗ acts on C by t · λ = tn λ, i.e. (t, λ)(t0 , λ0 ) = (tt0 , (t0 )n λ + λ0 )).
Show that Gn ∼
= Gm if and only if n = ±m.
(ii) Show that Lie(Gn ) ∼
= Cx + Cy, [x, y] = y which is independently of n.

Moreover, the map AlgGp → LieAlg is not surjective, Lie algebras in its image are called
algebraic Lie algebras. This is really obvious in characteristic p. Take for example slp /Z(slp ).
This cannot be the image of an algebraic group. In general, algebraic groups have a Jordan
decomposition – every element can be written as a sum of a semisimple and a nilpotent element
– and therefore the algebraic Lie algebras should have a Jordan decomposition as well.

2 Representations of sl2
2.1 Classification of sl2 representations
From now on, all Lie algebras and representations are over C. For sl2 we have the following
basis:      
0 1 0 0 1 0
e= , f= , h= ,
0 0 1 0 0 −1
subject to the relations [e, f ] = h, [h, e] = 2e, [h, f ] = −2f .

Theorem 2.1:
(i) For every n ≥ 0, there is a unique (up to isomorphism) irreducible representation sl2 of
dimension n + 1.

8 Robert Laugwitz & Henning Seidler

 2 Representations of sl2

(ii) Every finite dimensional representation of sl2 is a direct sum of irreducible representations
(i.e. the category of finite-dimensional representations of sl2 is semisimple, or every finite-
dimensional sl2 representation is completely reducible).

We now start proving part (i):

Let V be a representation of sl2 . Define the λ-weight space for V to be
Vλ = {v ∈ V | h · v = λv} ,
the eigenvectors of h with eigenvalue λ.
Example: L(n)λ = C · xi y j if i − j = λ.
Suppose v ∈ Vλ . Consider ev
h(ev) = (he − eh + eh)v = ([h, e] + eh)v
= 2ev + eλv = (2 + λ)ev,
so v ∈ Vλ if and only if ev ∈ Vλ+2 . Similarly, one shows that v ∈ Vλ if and only if f v ∈ Vλ−2 .

Definition 2.2: If v ∈ Vλ ∩ ker e, i.e. ev = 0 and hv = λv, we say that v is a highest weight
vector of weight λ.

Lemma 2.3: Let V be a representation of sl2 . If v ∈ V is a highest weight vector of weight λ,

then
W = hv, f v, f 2 v, · · · i
is an sl2 -invariant subspace of V , i.e. a subrepresentation.

Proof: We must show f W ⊂ W, hW ⊂ W, eW ⊂ W . Well, f W ⊂ W is obvious. We already

know that f k v ∈ Vλ−2k , so hW ⊂ W . Show eW ⊆ W by proving ef k v ∈ W, ∀k. Claim:
e · f n v = n(λ − n + 1)f n−1 v ∈ W. (1)
We prove this formula by induction on n:
The formula holds trivially for n = 0 since v ∈ ker e. Assume that the formula holds for n ≥ 0.
Then
e · f n+1 v = (ef − f e + f e)f n v
= hf n v + f ef n v
= (λ − 2n)f n v + n(λ − n + 1)f n v, by induction hypothesis
= (n + 1)(λ − n)f n v.

Lemma 2.4: Let V be a representation of sl2 and v a highest weight vector with weight λ. If
V is finite-dimensional, then λ ∈ N0 .

Proof: The vectors f i v all lie in different eigenspaces of h, and hence if non-zero are linearly
independent. But if V is finite-dimensional, then it must be f k v = 0 for some k, so f k+r v = 0,
for all f ≥ 0. Choose k minimal such that f k v = 0. So f k−1 v 6= 0, but then
(1)
0 = ef k v = |{z}
k (λ − k + 1) f k−1 v ,
| {z }
6=0 6=0

so λ = k − 1, i.e. λ ∈ N0 .

Robert Laugwitz & Henning Seidler 9

Lie Algebras and Their Representations

Proposition 2.5: If V is a finite-dimensional representation of sl2 , then there exist a highest

weight vector.

Proof: Let v ∈ V be some eigenvector for h with eigenvalue λ (exists as C is algebraically

closed). As before, v, ev, e2 v, ... are all eigenvalues for h, with respect to the distinct eigenvalues
λ, λ − 2, .... Hence, v, ev, e2 v, ... are linearly independent unless they are zero. But V is finite-
dimensional, so there exists a k s.t. ek v 6= 0, but then ek+1 v = ek+r v = 0, ∀r ≥ 1. Hence, ek v
is a highest weight vector with weight λ + 2k.

Corollary 2.6: If V is irreducible, then dim V = n + 1, for some n ≥ 0. We have seen that we
can find a basis v0 , v1 , ..., vn with

hvi = (n − 2i)vi ,
(
vi+1 , if i ≤ n
f vi =
0, if i = n
evi = i(n − i + 1)vi−1 ,

i.e. there is precisely one irreducible representation of sl2 of dimension n + 1. In particular, this
representation is given by L(n).

This finishes the proof of part (i) of Theorem 2.1. We now prove part (ii). Notice, that the
statement implies, in particular, that h acts diagonalizable on every finite-dimensional represen-
tation. First, another exercise:

L
Exercise 2.7: We have seen that C[x, y] = n≥1 L(n) is a representation of sl2 , a direct sum
of irreducible representations L(n), show that xµ y λ ∈ C[x/y, y/x] is a representation of sl2 for
all λ, µ ∈ C (using the given formulas) and describe its submodule structure.

Definition 2.8: Let V be a finite-dimensional representation of sl2 . Define

1
Ω := ef + f e + h2 ∈ End(V ). (2)
2

Ω is called the Casimir of sl2 .

Lemma 2.9: The Casimir Ω is central, i.e. eΩ = Ωe, f Ω = Ωf , hΩ = Ωh as elements of

End(V ).

10 Robert Laugwitz & Henning Seidler

 2 Representations of sl2

Proof: For example,

1
eΩ = e(ef + f e + h2 )
2
1
= eh + 2ef e + eh2
2
1 1
= eh + 2ef e + (eh − he)h + heh
2 2
1
= 2ef e + heh
2
1
= 2ef e − he + he + heh
2
1 1
= 2ef e − (ef − f e)e + h(he − eh) + heh
2 2
1 2
= ef e + f ee + h e = Ωe,
2

and similar calculations show that Ω also commutes with f and h.

Corollary 2.10: If V is an irreducible finite-dimensional representation of sl2 , then Ω acts on

it by a scalar.

Proof: Since Ω is central, ρΩ : V → V, v 7→ Ωv defines an sl2 -linear map, then Schur’s Lemma
gives that ρΩ = λIdV , for some λ ∈ C.

Lemma 2.11: Let L(n) denote the irreducible representation with highest weight vector v, of
weight n, then Ω acts on L(n) by 12 n(n + 2).

Proof: We have ev = 0, hv = nv, Ω = ( 12 h2 + h) + 2f e, so Ωv = 21 n(n + 2)v by Schur’s

Lemma. Notice, that Schur’s Lemma is actually not needed here. One can simply apply that
Ωf i v = f i Ωv, and {f i v | i ∈ N} span L(n).

Observe, if L(n) and L(m) are two irreducible finite-dimensional representations of sl2 , and
Ω acts on them by the same scalar, then n = m.

Proof: Ω acts by 12 n2 + 2 = 12 m2 + m, but f (x) = 21 x2 + x is a strictly increasing function for

x > −1.

Let V be any finite-dimensional representation of sl2 , set

λ
V λ := hv ∈ V | (Ω − λ)dim V v = 0i,

the generalized eigenspace of Ω with eigenvalue λ. Using Jordan decomposition, we can decom-

Robert Laugwitz & Henning Seidler 11

Lie Algebras and Their Representations

λ,
L
pose V = λV and write Ω in Jordan normal form
 
λ1
 .. 

 . 


 λ1 


 λ2 1 

 .. .. 
Ω=
 . .  ,

..
 
.
 
 1 
 

 λ2 

 .. 
 . 
λn

for some generalized eigenvalues λ1 , ...λn ∈ C. The Jordan blocks correspond to generalized
eigenspaces V λ .

Claim 2.12: Each V λ is a subrepresentation of sl2 .

Proof: Let x ∈ sl2 , v ∈ V λ , then

(Ω − λ)dim V xv = x(Ω − λ)dim V v

= x0 = 0,

so xv ∈ V λ also.

If V λ 6= 0, then λ = 12 n2 + n for a unique n ≥ 0, and we can show, that each V λ is ‘glued

together’ from copies of the representation L(n). This can be formulated more precisely using
composition series.

Definition 2.13: Let W be a finite-dimensional g-module, for a Lie algebra g. A composition

series for W is a sequence of submodules

0 = W0 < W1 < ... < Wr = W,

such that each quotient Wi /Wi−1 is an irreducible module.

Example 2.14:
0 1

.. ..
. . 
(i) If g = C, W = Cr , where 1 ∈ C acts as the matrix  ..
, then there is a unique
. 1
0
composition series
0 < he1 i < he1 , e2 i < ... < he1 , e2 , ..., er i,
and the subquotients are all C (the trivial module).

(ii) If g = C, W = Cr , 1 ∈ C acts as 0, then any chain

0 < W1 < ... < Wr = W,

with dim Wi = i, is a composition series and again, the subquotients are C.

12 Robert Laugwitz & Henning Seidler

 2 Representations of sl2

Claim 2.15: Composition series exist for any finite-dimensional g-module W .

Proof: By induction on dim W . Take any irreducible submodule W1 < W , then W/W1 is of
smaller dimension, so has a composition series

0 < W 2 < W 3 < ... < W r = W/W1 ,

by induction, then

0 < W1 < W1 + W 2 < W1 + W 3 < ... < W1 + W r−1 < Wr = W,

is a composition series of W .

Lemma 2.16: Let V λ 6= 0, then λ = 12 n2 + n for a unique n, and V λ has a composition series,
such that all quotients are isomorphic to L(n).

Proof: Let W be an irreducible submodule of V λ , Ω still acts on W by λ, but W is L(n), for

some unique n ≥ 0, and so λ = 21 n2 + n. Now consider V λ /W . If this space is non-zero, Ω still
acts on V λ /W with only one generalized eigenvalue λ. We can repeat this procedure as long as
the quotient is non-zero. This shows that V λ has a composition series with L(n) as the only
module which appears as a quotient, i.e. Wi /Wi−1 = L(n), for all i.

Corollary 2.17: h acts on V λ with (generalized) eigenvalues in {n, n − 2, ..., 2 − n, −n}.

Proof: If h acts on W , W 0 ≤ W a subspace s.t. hW 0 ≤ W 0 , then

{gen. eigenvalues of h on W } = {gen. ev. of h on W 0 } ∪ {gen. ev. of h on W/W 0 }. (3)

If we apply this to V λ , we obtain

{generalized eigenvalues of h on V λ } = {eigenvalues of h on L(n)}.

Lemma 2.16 says that the action of h on V λ has the following form:

∗ ∗ ∗
 
L(n)
 0 L(n) ∗ ∗ 
.
 
 . . . .
 . . ∗ 
0 L(n)

Using Jordan normal form, L(n) has diagonal form with eigenvalues {n, n − 2, ..., −n + 2, −n}
on the diagonal. So these are the only generalized eigenvalues of V λ , i.e. (V λ )m = 0 if m ∈ /
λ λ
{n, n − 2, ..., −n + 2, −n}. Further, h acts on ker(e : V → V ) with only one generalized
λ
eigenvalue, namely n, i.e. if x ∈ ker e, then (h − n)dim V · x = 0. To see this, apply (3) to the
composition series given by W i = Wi ∩ ker e, if the Wi come from the composition series of V λ ,
developed in Lemma 2.16.

Robert Laugwitz & Henning Seidler 13

Lie Algebras and Their Representations

Lemma 2.18: For the endomorphisms given by elements of sl2 acting on V λ , the following
identities hold:
(i) hf n = f n (h − 2n)

(ii) ef n+1 = f n+1 e + (n + 1)f n (h − n)

Proof:
(i) By induction on n:
For n = 1, we have hf = hf − f h + f h = −2f + f h = f (h − 2).
Assume, the formula holds for n ≥ 1. Then

hf n+1 = f n (h − 2n)f = f n hf − 2nf n+1

= −2f n+1 + f n+1 h − 2nf n+1
= f n+1 (h − 2(n + 1))

(ii) By induction on n:
For n = 0, observe ef = ef − f e + f e = h + f e.
Assume, that the formula holds for n ≥ 0, then

ef n+2 = (f n+1 e + (n + 1)f n (h − n))f

= f n+1 ef + (n + 1)f n hf − n(n + 1)f n+1
= f n+1 h + f n+2 e − 2(n + 1)f n+1 + (n + 1)f n+1 h − n(n + 1)f n+1
= f n+2 e + (n + 2)f n+1 (h − (n + 1))

Proposition 2.19: h acts diagonalizable on ker(e : V λ → V λ ), i.e.

ker e = (V λ )n = {x ∈ V λ | hx = nx}.

Proof: “⊇”: If hx = nx, then ex ∈ (V λ )n+2 = 0, so x ∈ ker e.

“⊆”: Let x ∈ ker e. We showed in Corollary 2.17 that in this case
λ
(h − n)dim V x = 0. (4)

Now, by part (i) of Lemma 2.18:

λ
(h − n + 2k)dim V f k x = 0,

i.e. f k x lies in the generalized eigenspace for h with eigenvalue n − 2k (∗). On the other hand,
if y ∈ ker e, and y 6= 0, then f k y 6= 0 (∗∗). To prove this, let

0 = W0 < W1 < ... < Wr = V λ

be a composition series for V λ . There exists an i s.t. y ∈ / Wi , but y ∈ Wi−1 , put y = y + Wi−1 .
∼
Note that y 6= 0 ∈ Wi /Wi−1 = L(n). Then y is a highest weight vector for L(n), so f n y 6= 0
in L(n), so f n y 6= 0 in V λ . Now, f n+1 y lies in the generalized eigenspace for h with eigenvalue
−n − 2, by (∗), but this is the zero space. Hence, f n+1 y = 0. Now, by applying Lemma 2.18,
we can conclude
0 = ef n+1 y = (n + 1)f n (h − n)y + f n+1 ey ,
|{z}
=0
so f n (h − n)y = hf n y = 0, but if (h − n)y 6= 0, this would contradict (∗∗), so hy = ny.

14 Robert Laugwitz & Henning Seidler

 2 Representations of sl2

Now we can finish our proof of Theorem 2.1. We can now choose a basis w1 , ..., wk of
ker(e : V λ → V λ ) s.t. hwi = nwi and ewi = 0, by Proposition 2.19 (i.e. ker e = (ker e)n . This
gives a direct sum composition of V λ , using the basis w1 , f w1 , ..., f n w1 , ..., wk , f wk , ..., f n wk . h
acts diagonalizable on V λ with respect to this basis, and hence on the whole of V (using Jordan
decomposition). To convince ourselves, that this is true, consider

hf k wi = f k (h − 2k)wi
= f k (n − 2k)wi
= (n − 2k)f k wi ,

i.e. f k wi ∈ (V λ )n−2k , and h acts diagonalizable on the whole of V λ . This concludes the proof
of Theorem 2.1.

Exercise 2.20 (fun!): Show that, if chark = p, then

(i) irreducible highest weight representations of sl2 (Fp ) are parametrized by n ∈ N , and

(ii) arbitrary finite-dimensional representations of sl2 (Fp ) do not need to break up into a direct
sum of irreducibles.

2.2 Consequences
Let V, W be representations of a Lie algebra g.

Claim 2.21: The map g → End(V ⊗ W ) = End(V ) ⊗ End(W ) given by x 7→ x ⊗ 1 + 1 ⊗ x is

a homomorphism of Lie algebras.

Proof: This map is obviously linear. To see that it is a Lie algebra morphism, consider

[x ⊗ 1 + 1 ⊗ x, y ⊗ 1 + 1 ⊗ y] = (x ⊗ 1 + 1 ⊗ x)(y ⊗ 1 + 1 ⊗ y) − (y ⊗ 1 + 1 ⊗ y)(x ⊗ 1 + 1 ⊗ x)
= xy ⊗ 1 + x ⊗ y + y ⊗ x + 1 ⊗ xy − (yx ⊗ 1 + y ⊗ x + x ⊗ y + 1 ⊗ yx)
= (xy − yx) ⊗ 1 + 1 ⊗ (xy − yx)

Remark 2.22: This comes from the group homomorphism G → G × G, g 7→ (g, g) by differ-
entiating.

Corollary 2.23: If V, W are representations of g, so is V ⊗ W .

Remember that if A is an algebra, V, W representations of A, then V ⊗ W is a representation

of A ⊗ A. To make it a representation of A, we need an algebra homomorphism A → A ⊗ A
(such a map is called coproduct of a Hopf algebra).
Now, take g = sl2 . This gives rise to the question how L(n) ⊗ L(m) breaks up into a direct
sum of irreducibles L(i) (using Theorem 2.1). One method to answer this question is to find all
the highest weight vectors.

Exercise 2.24: Find all highest weight vectors in L(1) ⊗ L(m), ..., L(n) ⊗ L(m).
Easy start: L(n) ⊗ L(m). Write vn for the highest weight vector in L(n), we claim that vn ⊗ vm

Robert Laugwitz & Henning Seidler 15

Lie Algebras and Their Representations

is a highest weight vector in L(n) ⊗ L(m). To prove this, consider

h · (vn ⊗ vm ) = (hvn ) ⊗ vm + vn ⊗ (hvm )

= (n + m)vn ⊗ vm , and
e · (vn ⊗ vm ) = (evn ) ⊗ vm + vn ⊗ (evm ) = 0.

From this, we can conclude that L(n) ⊗ L(m) = L(n + m) + X, but since

(n + 1)(m + 1) = dim L(n) ⊗ L(m)

= dim L(n + m) + dim X
= n + m + (nm + 1),

there is still “lots of stuff” remaining, if we quotient out by the submodule L(n + m). One
strategy to find this “other stuff” is to write down explicit formulas for all the other highest
weight vectors. These are complicated, but mildly interesting.
However, to determine the summands of L(n) ⊗ L(m) we do not have to do this.

Definition 2.25: Let V be a finite-dimensional representation of sl2 . The character of V is

defined as
X
ch V = dim Vn z n ∈ N[z, z −1 ].
n∈Z

Lemma 2.26: Let V, W be sl2 -representations, then

(i) ch V |z=1 = dim V ,

z n+1 −z −(n+1)
(ii) ch L(n) = z n + z n−2 + ... + z −n+2 + z −n = z−z −1
, sometimes denoted as [n + 1]z ,

(iii) ch V = ch W ⇐⇒ V ∼
= W,

(iv) ch V ⊗ W = chV · chW .

Proof:
L
(i) h acts diagonalizable with all its eigenvalues integers, i.e. V = n∈Z Vn by Theorem 2.1.

(ii) Follows from Theorem 2.1.

(iii) The characters ch L(0) = 1, ch L(1) = z + z −1 , ch L(2) = z 2 + 1 + z −2 , ... form a basis of

Z[z, z −1 ]Z/2 , the symmetric Laurent polynomials. Clearly, they are linearly independent
and span this space (by inspection). On the other hand, by part (ii) of Theorem 2.1
(complete reducibility), we have

V ∼ an L(n), W ∼
M M
= = bn L(n),
n≥0 n≥0

and V ∼
= W if and only if an = bn for all n ∈ N. But now, as {ch L(n)}n forms a basis,
P
ch V = n≥0 an ch L(n) determines an .

16 Robert Laugwitz & Henning Seidler

 2 Representations of sl2

(iv) Since for vn ∈ Vn , vm ∈ Vm , we have h(vn ⊗ vm ) = (n + m)vn ⊗ vm , we see that Vn ⊗ Vm ⊆

(V ⊗ W )n+m , so
X
(V ⊗ W )p = Vn ⊗ Vm ,
n,m
n+m=p
X
⇒ dim(V ⊗ W )p = (dim Vn )(dim Vm )
n,m
n+m=p
X X
⇒ ch V ⊗ W = (dim Vn )(dim Vm )z p
p∈Z n,m
n+m=p

= (ch V )(ch W ),

since this is how we multiply polynomials.

Example 2.27: Decompose L(1) ⊗ L(3).

ch L(1) ⊗ L(3) = (z + z −1 )(z 3 + z + z −1 + z −3 )

= (z 4 + z 2 + 1 + z −2 + z −4 ) + (z 2 + 1 + z −2 )

hence L(1) ⊗ L(3) ∼

= L(4) ⊕ L(2).

We can use the Clebsch-Gordon-rule

n+m
M
L(n) ⊗ L(m) = L(k). (5)
k=|n−m|
k≡n−m (mod 2)

Without giving a formal prove, the formula can be verified by drawing diagrams:

zn z n−2 ··· z −n+2 z −n

L(n)
L(kl )
zm
.
..

z m−2

..
.

z −m+2
.
..
L(k1 )
z −m
L(m)

Here, L(k1 ), ..., L(kl ) are the components of the direct sum composition of L(n) ⊗ L(m), and
k1 , ..., kl refer to the respective length of the lines in the inner square of the diagram.

Robert Laugwitz & Henning Seidler 17

Lie Algebras and Their Representations

Example 2.28: Compute L(3)⊗L(4), ch L(3)⊗L(4) = (z 3 +z +z −1 +z −3 )(z 4 +z 2 +z −2 +z −4 ).

Since z 7 appears as the highest coefficient in the product of the characters, L(7) appears in the
decomposition. Subtracting ch L(7), the highest coefficient is z 5 and thus L(5) appears as an
summand. Continuing in a similar manner, we conclude
L(3) ⊗ L(4) = L(7) ⊕ L(5) ⊕ L(3) ⊕ L(1).
This can also be seen by considering the diagram
z3 z z −1 z −1

z4

z2

z −2

z −4
L(1) L(3) L(5) L(7)

Here L(7) refers to the line of length 7, L(5) refers to the line of length 5,...

In the following chapters we will look at other Lie algebras including sln , son , sp2n . We will
• see that the categories of representations are semi-simple,
• parametrize irreducible representations,
• compute the character of the irreducibles and their dimensions,
• see how to decompose ⊗ using picture crystals.
In order to do this, we need
• linear algebra characterizations of such Lie algebras, and
• the structure theory of roots and weights.

3 Structure and Classification of Simple Lie Algebras

3.1 Linear algebra preliminaries
Definition 3.1:
(i) A Lie algebra g is simple if the only ideals of g are 0 and g and dim g > 1 (⇐⇒ g is
non-abelian). g is semi-simple if it is a direct sum of simple Lie algebras.
(ii) [g, g] is the span of {[X, Y ] : X, Y ∈ g}, the derived algebra of g.
(iii) The central series of g is defined by g0 = g and gn = [gn−1 , g], i.e
g ⊇ [g, g] ⊇ [[g, g], g] ⊇ ...
The derived series is: g < (0) = g and g(n) = [g(n−1) , g(n−1) ], i.e.
g ⊇ [g, g] ⊇ [[g, g], [g, g]] ⊇ ...

18 Robert Laugwitz & Henning Seidler

 3 Structure and Classification of Simple Lie Algebras

(iv) g is nilpotent if gn = 0 for some n > 0, solvable if g(n) = 0 for some n > 0.

Remark 3.2: g nilpotent implies g solvable since we always have g(n) ⊆ gn .

Exercise 3.3: Show [g, g] is an ideal, and g/[g, g] is abelian.

Solution: [g, g] is an ideal since [g, [a, b]] ∈ [g, g], for all g, a, b ∈ g. To see that g/[g, g] is abelian,
consider

[a + [g, g], b + [g, g]] = [a, b] + [g, g]

= [g, g].

Example 3.4: The Lie algebra h of strictly upper triangular matrices is nilpotent, the Lie
algebra b of upper triangular matrices is solvable.

Exercise 3.5:

(i) Compute the derived and the central series of h and b and check the above claim.

(ii) Compute the center of h and b.

Let W be a symplectic vector space, i.e. a vector space with an inner product h , i (that is
an non-degenerate antisymmetric bilinear form). For example, take L to be any vector space,
set W = L + L∗ , and define hL, Li = hL∗ , L∗ i = 0, hv, v ∗ i = v ∗ (v) = −hv ∗ , vi, ∀v ∈ L, v ∗ ∈ L∗ .

Exercise 3.6: Define the Heisenberg Lie algebra HW := W ⊕ Cc as a vector space, [w, w0 ] =
hw, w0 ic, for w, w0 ∈ W , and [c, w] = 0. Show that HW is a Lie algebra that is nilpotent.

Solution. [ , ] is bilinear and skew-symmetric since h , i is. Note that [ , ] ⊆ Cc. Thus, the
Jacobi identity follows immediately from [c, w] = 0. Because of the same property, we have that
HW2 = [[HW , HW ], HW ] = [hW, W ic, HW ] = 0. This proves that HW is nilpotent, and thus
also solvable.

Example 3.7: Let L = C, then HW = Cp + Cq + Cc , [p, q] = c, [c, p] = [c, q] = 0. Show that

∂
this has a representation on C[x] by q 7→ x, p 7→ ∂x , c 7→ 1.

Solution: We need to check that the defined map preserves the relations [p, q] = c, [c, p] =
0, [c, q] = 0:
   
∂ n ∂ ∂
,x x = x−x xn
∂x ∂x ∂x
= (n + 1)xn − nxn = 1 · xn ,
   
∂ n ∂ ∂
,1 x = 1−1 xn = 0,
∂x ∂x ∂x
[x, 1]xn = (x − x)xn = 0.

This shows that the defined map is a representation of HW .

Robert Laugwitz & Henning Seidler 19

Lie Algebras and Their Representations

Proposition 3.8:

(i) Subalgebras and quotient algebras of solvable (resp. nilpotent) Lie algebras are solvable
(resp. nilpotent).

(ii) Let g be a Lie algebra, h an ideal. Then we have g solvable ⇐⇒ h and g/h are solvable.
(So solvable Lie algebras are built out of abelian Lie algebras, it exists a refinement of
derived series s.t. the subquotients are 1-dimensional).

(iii) g is nilpotent if and only if the center Z(g) 6= 0 and g/Z(g) is nilpotent. (Indeed: if g
nilpotent g ⊃ g1 ⊃ . . . ⊃ gn−1 ⊃ gn = 0. But 0 = gn = [gn−1 , g] ⇒ gn−1 lies in the center
of g.)

(iv) In particular, g is nilpotent ⇐⇒ ad(g) ⊆ gl(G) is nilpotent (as 0 → Z(g) ,→ g  ad(g) =

g/Z(g) → 0 is an exact sequence).

Theorem 3.9 (Lie’s Theorem): Let g ⊆ glV be a solvable Lie algebra over an algebraic
closed field k with char k = 0. Then there exists a basis v1 , . . . , vn of V such that w.r.t. this
basis the matrices of all elements of g are upper triangular, i.e. g ⊆ bV .
Equivalently, there exists a λ : g → k linear and v ∈ V st. xv = λ(x)v for all x ∈ g (that is
v is a common eigenvector for g, i.e. a one-dimensional subrepresentation of V ).

Exercise 3.10:

(i) Show these are equivalent.

(ii) Show it is necessary that K = K and char K = 0. For example, take g = HW =

hp, q, ci, char K = p and show K[x]/K is an irreducible representation of g, contradicting
Lie.

Solution:

(i) Assume first statement of Lie’s Theorem. Fix basis v1 , ..., vn such that all elements of g act
as an upper triangle matrix. Note that for any h ∈ g we have hv1 = λh for some λh ∈ K.
Define common eigenvector by h 7→ λh .
Conversely, if we have a one-dimensional subrepresentation V1 of V , take 0 6= v1 ∈ V1 as
first basis vector. Assume, we have found v1 , ..., vk basis vectors such that every h ∈ g
acts as an upper triangular matrix on Wk := hv1 , ..., vk i. Then V /Wk is still solvable,
and we again find a one-dimensional subrepresentation Vk+1 . Take 0 6= vk+1 ∈ Vk+1 .
Then hhvk+1 i ∩ Wk = 0 and therefore h acts as an upper triangular matrix on Wk+1 :=
hWk , vk+1 i. Thus, the claim follows by induction.

Corollary 3.11: Let char k = 0, g a solvable finite-dimensional Lie algebra, then [g, g] is
nilpotent.

Proof: Apply Lie’s theorem to the adjoint representation ad : g → End(g). Then – w.r.t. some
basis – ad(g) ⊆ b, but [b, b] ⊆ h, so [ad g, ad g] is nilpotent. Moreover, [ad g, ad g] = ad[g, g], so
[g, g] has to be nilpotent (by property (iv) of Proposition 3.8).

Exercise 3.12: Find a counterexample to the previous corollary for char k = p.

20 Robert Laugwitz & Henning Seidler

 3 Structure and Classification of Simple Lie Algebras

We call a endomorphism φ : V → V nilpotent if all its eigenvalues are zero. Or, equivalently,
ifφn= 0 for some n ≥ 0.

Theorem 3.13 (Engel’s Theorem): Let k be an arbitrary field. g is a nilpotent Lie algebra
if and only if ad(g) consists of nilpotent endomorphisms of g. Or, equivalently, if (V, π) is a
finite-dimensional representation of g such that π(x) : V → V is a nilpotent endomorphism for
all x ∈ g, then there exists 0 6= v ∈ V st. π(x)v = 0 for all x ∈ g (i.e. V has a trivial
subrepresentation). This again, implies the existence of a basis such that all matrices π(x) are
strictly upper triangular.

Exercise 3.14: Show that the two formulations of Engel’s Theorem are equivalent.

Solution:

⇐: Claim: For V exists a series

0 = V0 ⊂ V1 ⊂ ... ⊂ Vn = V

s.t. dim Vi = i and gVi ⊂ Vi−1 . Note that from the claim it follows that we can find a basis
for V s.t. g acts as an strictly upper triangular matrix, for all g ∈ g. Setting V = ad(g)
gives that ad g is nilpotent and therefore g is nilpotent.
Proof of the claim by induction on n = dim V : If dim V = 1, we have that V = kv is
abelian. In this case, set V1 = V . Now, let dim V > 1, then we find an one-dimensional
subrepresentation V1 of V . Consider V 0 = V /V1 with canonical projection π, then dim V 0 <
dim V and we can apply the induction hypothesis to obtain a series

0 = V00 < V10 < ... < Vn−1

0
=V0

with the claimed properties. Now, the series defined by Vi = π −1 (Vi−1

0 ), i = 1, ..., n − 1
−1 0 −1 0
and Vn = V gives the claim as g(Vi ) = gπ (Vi−1 ) ⊂ π (Vi−2 ) = Vi−1 .

⇒: First, show that if (V, π) consists of nilpotent endomorphism, then also ad does. Assume
π(x)n = 0, x ∈ g.

Definition 3.15: A symmetric bilinear form (·, ·) : g×g → k is invariant if ([x, y], z) = (x, [y, z])
for all x, y, z ∈ g.

Exercise 3.16: If a ⊆ g is an ideal, (·, ·) an invariant form on g, then a⊥ = {x ∈ g : (x, a) = 0}

is an ideal.

Solution: To check: for a ∈ a⊥ , [a, g] ⊆ a⊥ , i.e. ([a, g], a) = 0. But since (·, ·) is invariant, we
have ([a, g], a) = (a, [g, a]) = (a, a) = 0, where we use that a is an ideal.

Definition 3.17: Let V be a representation of g via ρ : g → gl(V ), then we define

(x, y)V := tr(ρ(x)ρ(y) : V → V ),

the trace form of V .

Robert Laugwitz & Henning Seidler 21

Lie Algebras and Their Representations

Exercise 3.18: Check that ρ is a representation implies that (·, ·)V is symmetric, bilinear and
invariant.

Solution: The trace form is symmetric as tr(AB) = tr(BA). Bilinearity follows from linearity
of ρ and tr. Check that the trace form is invariant:

tr(ρ[x, y]ρ(z)) = tr(ρ(x)ρ(y)ρ(z) − ρ(y)ρ(x)ρ(z))

= tr(ρ(x)ρ(y)ρ(z)) − tr(ρ(x)ρ(z)ρ(y))
= tr(ρ(x)ρ[y, z]).

Example 3.19: Define (·, ·)ad the killing form, to be the trace form attached to the adjoint
representation, i.e. (x, y)ad = tr(ad x · ad y : g → g).

Theorem 3.20 (Cartan’s Criterion): Let g ⊆ glV , char k = 0, then g is solvable if and only
if for all x ∈ g and y ∈ [g, g] we have (x, y)V = 0, i.e. [g, g] ⊆ g⊥ .

Exercise 3.21: Observe that Lie’s theorem implies Cartan’s criterion immediately. If g is
solvable and non-abelian, then all trace forms are degenerate.

Solution: If g is solvable, use Lie’s theorem to find a basis of a representation (V, ρ) such that
ρg ⊆ b. Then [ρg, ρg] ⊆ h, the set of upper triangular matrices with only zeros on the diagonal.
Now it is clear, that tr(ρ[x, y]ρz) is 0.

Corollary 3.22: A Lie algebra g is solvable if and only if (g, [g, g])ad = 0.

Proof: ⇒: is Lie’s Theorem.

⇐: Cartan’s criterion gives that ad(g) = g/Z(g) is solvable. But the center is abelian and so it
is always solvable. Therefore g is solvable, too.

Warning: Not every invariant form is a trace form.

Exercise 3.23: Let H˜ = Chp, q, c, di with [c, H˜ ] = 0, [p, q] = c, [d, p] = p, [d, q] = −q. Con-
struct a non-degenerate invariant form on H˜ . Show that H˜ is solvable. Extend the represen-
tation of Chc, p, qi on k[x] (given in 3.7) to a representation of H˜ .

3.2 Structure of semisimple Lie algebras

Definition 3.24: Let R(g) denote the maximal solvable ideal in g, the radical of g.

Exercise 3.25:

(i) Show that the sum of solvable ideals is solvable, i.e. R(g) is the sum of all solvable ideals.

(ii) Show R(g/R(g)) = 0.

Definition 3.26: A derivation is a linear map D : g → g satisfying D[x, y] = [Dx, y] +

[x, Dy] (e.g. ad(x) is a derivation (follows from Jacobi identity and skew symmetry of [·, ·])). A
derivation of the form ad(x) is called inner.

22 Robert Laugwitz & Henning Seidler

 3 Structure and Classification of Simple Lie Algebras

Theorem 3.27: Let char k = 0, then the following are equivalent:

(i) g is semisimple

(ii) R(g) = 0

(iii) The Killing form (·, ·)ad is non-degenerate (killing criterion).

Moreover, if g is semisimple, then every derivation D : g → g is inner. (But not conversely, i.e.
this does not characterize semisimple Lie algebras.)

Proof: First notice that R(g) = 0 ⇔ g has no non-zero abelian ideal. “⇒” clear since abelian
ideals are always solvable; “⇐”: if some ideal p ⊆ g is solvable, then the last term of its derivated
series is abelian.
Therefore, (i)⇒(ii) clear (if g is semisimple, it does not have non-zero abelian ideals).
(iii)⇒(ii): We will show: If a is an abelian ideal, then a ⊆ g⊥ = {x ∈ g : (x, a) = 0} where
(·, ·) = (·, ·)ad . Write g = a + h, h a vector space complement to a. If x ∈ a, then ad(a) has
matrix  
a 0 ∗
h 0 0
as a abelian and an ideal. If x ∈ g, then ad(x) has matrix
 
∗ ∗
0 ∗

as a is an ideal, so  
0 ∗
tr(ad a, ad x) = tr =0
0 0
so (a, g)ad = 0.
(ii)⇒(iii): Let i = g⊥ , which is an ideal. Suppose i 6= 0, then ad : i → gl(g) has (x, y)ad = 0 for
all x, y ∈ i. Now, by Cartan’s criterion i/Z(i) is solvable, so i is solvable.
(ii),(iii)⇒(i): Let (·, ·)ad be non-degenerate. Let a ⊆ g be a minimal ideal.
Claim: (·, ·)ad |a is either 0 or non-degenerate.
Proof: {x ∈ a : (x, a) = 0} = a ∩ a⊥ is an ideal. But a is minimal, so a ∩ a⊥ = 0 or a.
But Cartan implies a is solvable if (·, ·)ad |a is zero. But R(g) = 0, so it must be (·, ·)ad |a non-
degenerate. Hence g = a ⊕ a⊥ , as (·, ·)ad |a and (·, ·)ad are non-degenerate, with a simple. As this
is a direct sum of Lie algebras, any ideal of a⊥ is an ideal of g. Inductively repeating this with
a⊥ instead of g gives g =
L
ai where ai are simple Lie algebras (minimal and ideals).
(i)⇒(ii): Claim: If g is semisimple, then g is a direct sum of its minimal ideals in a unique
manner. To prove this, note first that all the components of the direct sum are ideals in g.
L
Write g = ai . Assume that r ⊆ g is an minimal ideal. Consider r ∩ ai . These are either 0 or
ai , since the ai are minimal. Hence, find j s.t. aj = r.
Now, by Cartan’s criterion, we have a is solvable if and only if (·, ·)ad |a is zero. But that would
contradict the direct sum composition into minimal ideals (since then a ⊂ a⊥ ). Hence, R(a) = 0.
Finally, let D : g → g be a derivation, g semisimple. Consider the linear function l : g → K
with x 7→ trg (D(ad x)). As g is semisimple, (·, ·)ad is non-degenerate, so there exists y ∈ g, st.
l(x) = (y, x)ad for all x ∈ g (this follows from x 7→ ad(x) being an isomorphism g → g∗ (as a
linear map with trivial kernel)). So we will show E = D − ad y is zero, i.e. D = ad y. (Note E
is a derivation). So to prove Ea = 0 for all a ∈ g, it is enough to show (Ex, z)ad = 0 for all x, z.

Robert Laugwitz & Henning Seidler 23

Lie Algebras and Their Representations

Assume (·, ·)ad is non-degenerate. Observe that ad(Ex) = E · ad x − ad x · E = [E, ad x] : g → g

(as ad(Ex)(y) = [Ex, y] = E[x, y] − [x, Ey] as E is a derivation), so

(Ex, z)ad = trg (ad(Ex) · ad z) = trg ([E, ad x] · ad z)

= trg (E[ad x, ad z]) = trg (E · ad[x, z]) = 0,

as by the definition of E: trg (E · ad(a)) = trg (D · ad(a)) − l(a) = 0.

Exercise 3.28: Show that [R(g), R(g)] ⊆ g⊥ ⊆ R(g).

Remark 3.29: If g is any Lie algebra, then

0→ R(g) → g → g/R(g) → 0
| {z } | {z }
solvable ideal semisimple

is an exact sequence with maximal semisimple quotient.

Theorem 3.30 (Levi’s theorem): If char k = 0, this exact sequence splits, i.e. there exists a
subalgebra s ⊆ g isomorphic to g/R(g) (this algebra is not canonical), so we have g = s n R(g)
(semidirect product). This is false in characteristic p.

Exercise 3.31:


(i) Let g = slp Fp . Show that R(g) = Fp I, but there is no complement.

(ii) A nilpotent Lie algebra always has non-inner derivations.

(iii) Let g = ha, bi with [a, b] = b. Show that g has only inner derivations. Note that for this
example (·, ·)ad = 0, so this is an example showing that the condition that all derivations
are inner does not imply that the Lie algebra is semisimple.

(iv) Let g be a simple Lie Algebra above field a k, (·, ·)1 and (·, ·)2 two non-degenerate invariant
bilinear forms. Show that there exists a λ ∈ k ∗ st. (·, ·)1 = λ(·, ·)2

(v) Let g = sln (C) (assume this is simple). Define (A, B) = tr(AB), so (A, B) = λ(A, B)ad .
Compute λ.

24 Robert Laugwitz & Henning Seidler

 4 Structure Theory

4 Structure Theory
In this section, we consider finite-dimensional Lie algebras.

Definition 4.1: A torus t ⊆ g is an abelian subalgebra s.t. for all t ∈ t, ad t = [t, ·] : g → g

is a diagonalizable (i.e. semisimple) linear map. A maximal torus is a torus not contained in a
bigger torus. A maximal torus is also called a Cartan subalgebra.

Example 4.2: Let T = (S 1 )r ,→ G a compact Lie group (or T = (C∗ )r ,→ G a reductive

algebraic group). Then t = Lie T ⊆ Lie G is a torus, and maximal if T is.

Exercise 4.3:

(i) g ⊂ sln or gln , t be the set of diagonal matrices (or the matrices of trace 0 if in sln ), then
t is a maximal torus.

(ii) ( 00 0∗ ) ⊆ sl2 is not a torus.

Proof:

(i) Case t ⊆ gln first: Clearly, t is an abelian Lie subalgebra of gln . Moreover, if we choose
the basis {Eij , Ell − Ekk | i 6= j, l < k} of sln , we notice that for D = Diag(λ1 , . . . , λn ) we
have ad D(Eij ) = (λi − λj )Eij and therefore ad D is represented by a diagonal matrix. t
is maximal since if t contains any other matrix (w.l.o.g. take Eij , i 6= j) then [t, Eij ] =
(ti − tj )Eij 6= 0, for a suitable choice of t. Hence, t is not abelian.
 
  0 0 0
0 1
(ii) ad is represented by the matrix 0 0 2 which is not diagonalisable as the
0 0
1 0 0
eigenspace of the only eigenvector 0 is 2-dimensional.

Let t1 , ..., tr : V → V be pairwise commuting (ti tj = tj ti ) diagonalizable linear maps. Let

λ = (λ1 , ..., λr ) ∈ Cr . Set Vλ = {v ∈ V | ti v = λi v, ∀i = 1, ..., r} simultaneous eigenspaces of all
ti .
L
Lemma 4.4: V = λ∈(Cr )∗ Vλ , i.e. V breaks up into a direct sum of simultaneous eigenspaces.

Proof: Induction on r. If r = 1, this is clear by requiring that ti is diagonalizable for all i.

L
If r > 1 consider t1 , ..., tr−1 , V = Vλ1 ,...,λr−1 by induction hypothesis. Now decompose
Vλ1 ,...,λr−1 into eigenspaces for tr (possible since tr diagonalizable).

Set t to be the r-dimensional abelian Lie algebra with basis t1 , ..., tr . Then V is a semisim-
L
ple (that is completely reducible) representation of t, by Lemma 4.4, and V = Vλ is its
decomposition into isotypical (i.e. direct sums of isomorphic summands) representations.

Exercise 4.5: Show that every irreducible representation of t is one dimensional.

Solution: If a subrepresentation W is not one-dimensional (as a vector space) then take 0 6=

v ∈ W and hvi is a t-subrepresentation of W .

Robert Laugwitz & Henning Seidler 25

Lie Algebras and Their Representations

Set Cλ to be the one-dimensional representation of t, where ti · w = λi w, for all i. Then

Vλ is a sum of dim Vλ copies of Cλ (a direct sum), and λ 6= µ implies Cλ  Cµ . Really, λ is
a linear map t → C, i.e. λ ∈ t, where λ(ti ) =: λi . So, one-dimensional representations of t
correspond to irreducible representations of t which are in 1-1 correspondence to elements of
t∗ = HomVect (t, C), and V = λ∈t∗ Vλ , Vλ = {v ∈ V | t · v = λ(t)v, ∀t ∈ t} is called the weight
L

space decomposition of V . Now, let g be a Lie algebra, t a maximal torus. The weight space
decomposition of g is M
g = g0 + gλ , (6)
λ∈t∗
λ6=0

where g0 = {x ∈ g | [t, x] = 0}, gλ = {x ∈ g | [t, x] = λ(t)x ∀t ∈ t}.

Definition 4.6: R = {λ ∈ t∗ | gλ 6= 0, λ 6= 0} are the roots of g.

Example 4.7 (Essential): g = sln , t the diagonal matrices in sln . If

 
t1 0
t=
 ..  ,

Eij = (δi,k δj,l )k,l ,
.
0 tn

then [t, Eij ] = (ti − tj )Eij . Define εi (t) := ti , so εi : t → C, i.e. εi ∈ t∗ and ε1 , ..., εn span t∗ , but
ε1 + ... + εn = 0 (as t ⊆ sln ). So [t, Eij ] = (εi − εj )(t)Eij and so

R = {εi − εj | i 6= j}, g0 = t.

R are the roots of t. (This shows also that t is a maximal torus), and gεi −εj = CEij , i 6= j, is
one-dimensional. So M
sln = t ⊕ gεi −εj
εi −εj ∈R

is the root space decomposition of sln .

Exercise 4.8 (Exam!): Compute the root space decomposition for g = so2n , so2n+1 , sp2n ,
where t = {diagonal matrices} ∩ g, and
 
0 1
son = {A ∈ gln | JA + AT J = 0}, J =  . . . ,
 

1 0
 
0 1
.. .
..
 

 . 

0 1
 
sp2n = {A ∈ gl2n | M A + AT M = 0}, M =  .
 
 −1 0 
. ..
 
.. .
 
 
−1 0

In particular, show that t is maximal torus and the root spaces are one-dimensional.

(i) Show A ∈ son (C) ⇐⇒ A is skew-symmetric w.r.t. side diagonal.

26 Robert Laugwitz & Henning Seidler

 4 Structure Theory

 
A1 A2
(ii) Show A = A 3 A4
∈ sp2n (C) ⇐⇒ A2 , A3 symmetric w.r.t. side diagonal, and A4 is
−A1 transposed with side diagonal.

(iii) h = {Diag(a1 , ..., an , 0, −an , ..., −a1 ) | ai ∈ K} is a Cartan subalgebra in both cases.

(iv) son : C(Eij −En−j+1,n−i+1 ), i+j < n+1, i 6= j are the root spaces for son . If n = 2l, roots
are R = {±εi ± εj | 1 ≤ i, j ≤ l, i 6= j}. If n = 2l + 1, the roots are R = {±εi ± εj , ±εi |
1 ≤ i, j ≤ l, i 6= j}.

(v) sp2n (C): root spaces are C(Eij − En−j+1,n−i+1 ), 1 ≤ i, j ≤ l, C(Eij + En−j+1,n−i+1 ), l <
i ≤ 2l, j ≤ l or i ≤ l, l < j ≤ 2l and roots are R = {±εi ± εj , ±2εi | 1 ≤ i, j ≤ l, i 6= j}

(vi) Show sp2l (l ≥ 2) and son are simple (n > 4, n = 3).

(vii) Show so4 ∼ = so3 ⊕ so3 , so3 ∼ = sp2 ∼

= sl2 , and so2 ∼ = C. Further, we have the isomorphisms
∼ ∼
so5 = sp4 , so6 = sl4 .

Remark: all these root spaces are one-dimensional.

Solution:

(i) Consider the ij-th element:

n
X
(JA + AT J)ij = Jik Akj + Aki Jkj
k=1
= An−i+1,j + An−j+1,i = 0,

i.e. A is skew symmetric w.r.t. the side diagonal.

 
A1 A2
(ii) M A + AT M = 0 ⇐⇒ M AM T = −AT . If A = A3 A4 this means that

−AT
   
JA4 J −JA3 J 1 −A3 T
−JA2 J JA1 J = −AT AT
,
2 4

looking at the four squares gives the claimed identities.

(iii) Note that JDiag(a1 , ..., al , 0, −al , ..., −a1 )T J = Diag(−a1 , ..., −al , 0, al , ..., a1 ). Thus, t ⊆
so2l+1 . For n = 2l consider the diagonal matrices of the form Diag(a1 , ..., al , −al , ..., −a1 ).
Also M Diag(a1 , ..., al , −al , ..., −a1 )T M T = Diag(−a1 , ..., −al , al , ..., a1 ) and hence t ⊆ sp2l .
Clearly, h is abelian (diagonal matrices commute). Further, for any diagonal matrix t =
(t1 , ..., tn ), [t, Eij ] = (ti − tj )Eij . Hence, ad t is diagonal for all h ∈ h. It remains to show
that h is maximal; this follows from the fact that the diagonal matrices form a maximal
torus in gln .

(iv) Consider son . We have a basis {Ei,j − En−j+1,n−i+1 : i + j ≤ n} for son . First, consider

Robert Laugwitz & Henning Seidler 27

Lie Algebras and Their Representations

so2l , then

[t, Ei,j − En−j+1,n−i+1 ] = [t, Ei,j ] − [t, En−j+1,n−i+1 ]

= (ti − tj )Ei,j − (tn−j+1 − tn−i+1 )En−j+1,n−i+1

(ai − aj )Ei,j − (−aj + ai )En−j+1,n−i+1 if i, j ≤ l


= (−ai − aj )Ei,j − (−aj − ai )En−j+1,n−i+1 if j ≤ l,i > l

(a + a )E − (a + a )E

if i ≤ l,j > l
i j i,j j i n−j+1,n−i+1

(ai − aj )(Ei,j − En−j+1,n−i+1 ) if i, j ≤ l


= (−ai − aj )(Ei,j − En−j+1,n−i+1 ) if j ≤ l,i > l .

(a + a )(E − E

if i ≤ l,j > l
i j i,j n−j+1,n−i+1 )

This gives root spaces of the claimed form, with roots {±εi ± εj : i + j < n, i < j}. If we
consider so2l+1 , we have the additional equations

[t, Ei,l+1 − En−l,n−i+1 ] = ±(ai ± 0)(Ei,l+1 − En−l,n−i+1 ),

giving the roots {εi , i = 1, ..., l}.

(v) For sp2l we have the following basis:

Eij − En−j+1,n−i+1 , 1 ≤ i, j ≤ l,
Eij + En−j+1,n−i+1 , i ≤ l, j > l or i > l, j ≤ l, and i + j ≤ 2l,
Ei,n−i+1 , i = 1, . . . , n.

Then a similar calculation as in (iv) shows that

[t, Ei,j − En−j+1,n−i+1 ] = (ai − aj )(Ei,j − En−j+1,n−i+1 ),

giving the root spaces C(Ei,j − En−j+1,n−i+1 ), (w.r.t. root ±εi ± εj ), 1 ≤ i, j ≤ l, i 6= j.

Further,

[t, Eij + En−j+1,n−i+1 ] = (ti + tj )Eij + (tn−j+1 + tn−i+1 )En−j+1,n−i+1

= (ti + tj )(Eij + En−j+1,n−i+1 ),

this gives the root spaces C(Ei,j + En−j+1,n−i+1 ) (w.r.t. root εi + εj ), for i ≤ l, j >
l, i + j ≤ 2l, and for i > l, j ≤ l, i + j ≤ 2l we obtain the roots −(εi + εj ). Finally,
[t, Ei,n−i+1 ] = 2ti Ei,n−i+1 = εi (t)Ei,n−i+1 .

Proposition 4.9: The Lie algebra sln is simple, for n ≥ 2.

L
Proof: Recall sln = t ⊕ α∈R gα , R = {εi − εj : i 6= j}, gεi −εj = CEij . Suppose a ⊆ sln is a
P
non-zero ideal. Choose r ∈ a, r 6= 0, s.t. if we write r = t + α∈R eα with eα ∈ gα , then the
number of non-zero terms is minimal.
Now suppose t 6= 0. Choose t0 ∈ t, st. α(t0 ) 6= 0 for all α ∈ R (i.e. t0 has distinct
P
eigenvalues). Consider [t0 , r] = α∈R α(t0 ) · eα ∈ a, as a is an ideal. This, if non-zero, has
fewer terms than r, contradicting our choice of r, hence must be zero. Therefore eα = 0 for all
α ∈ R and so r = t ∈ t, t 6= 0. Now this implies that there exists an α ∈ R with α = εi − εj
s.t. α(t) =: c 6= 0. Hence cEij = [t, Eij ] ∈ a, as a is an ideal with c 6= 0, so Eij ∈ a. But now

28 Robert Laugwitz & Henning Seidler

 4 Structure Theory

[Eij , Ejk ] = Eik if i 6= k and [Esi , Eij ] = Esj if j 6= s, so Eij ∈ a implies Eab ∈ a, for all a 6= b.
But now Eii − Ei+1,i+1 = [Ei,i+1 , Ei+1,i ] ∈ a also, but {Eab , Eii − Ei+1,i+1 } forms a basis for
sln , so a = sln .
P
If t = 0, write r = α∈R eα , and if there is only one non-zero term, then r = cEij , c 6= 0,
P
argue as before, to get a = sln . So r = cEα + dEβ + γ∈R\{α,β} eγ with α, β distinct. Choose
t0 ∈ t s.t. α(t0 ) 6= β(t0 ). Then a suitable linear combination of [t0 , r] and r has fewer terms
than r, contradicting our choice.

Proposition 4.10: Let g be a semisimple Lie algebra. Then maximal tori exist, i.e if t is a
maximal torus, then t 6= 0. Moreover g0 = {x ∈ g | [t, x] = 0} = t.

Proof: omitted

This means that the root space decomposition of a semisimple Lie algebra g is
M
g=t+ gα .
α∈R

Theorem 4.11 (Structure theorem for semisimple Lie algebras): Let g be a semisimple
L
Lie algebra, t ⊆ g maximal torus, write g = t + α∈R gα . Then:
(i) CR = t∗ , i.e. the roots span t∗ ,

(ii) dim gα = 1,

(iii) If α, β ∈ R and α + β ∈ R, then [gα , gβ ] = gα+β . If α + β ∈

/ R, and α 6= −β, then
[gα , gβ ] = 0.

(iv) [gα , g−α ] ⊂ t and is one-dimensional, and gα ⊕[gα , g−α ]⊕g−α is a Lie subalgebra, isomorphic
to sl2 .
Proof:
(i) If not, there is some t ∈ t, t 6= 0 with α(t) = 0 for all α ∈ R. But then for x ∈ gα , we have
[t, x] = 0, i.e. [t, gα ] = 0 for all α ∈ R. But [t, t] = 0, as t is abelian. So t is in the center
of g. But g is semisimple, so it has no abelian ideals and therefore no center.
Next we will prove several properties, which lead to the proof of the theorem, but will not be
directly assigned to its statements.
(a) [gλ , gµ ] ⊆ gλ+µ for all λ, µ ∈ t∗

Proof. By the Jacobi identity for all t ∈ t, x ∈ gλ , y ∈ gµ we have

[t, [x, y]] = [[t, x], y] + [x, [t, y]] = λ(t)[x, y] + µ(t)[x, y] = (λ + µ)(t)[x, y]

(b) (gλ , gµ )ad = 0 if λ 6= −µ. Moreover, (·, ·)ad |gλ +g−λ is non-degenerate.

Proof. Let x ∈ gλ , y ∈ gµ . By (a) (ad(x) ad(y))N gα ⊆ gα+N (λ+µ) , but λ + µ 6= 0 and g

is finite-dimensional. So for N sufficiently large, we have α + N (λ + µ) ∈ / R and then
gα+N (λ+µ) = 0. So ad(x) ad(y) is nilpotent and therefore

trg (ad x ad y) = (x, y)ad = 0.

Robert Laugwitz & Henning Seidler 29

Lie Algebras and Their Representations

On the other hand, (·, ·)ad is non-degenerate (by (3.27) as g is semisimple) and g =
L
λ∈t∗ gλ , so it must be that (·, ·)ad |gλ +gµ is non-degenerate, so in particular (·, ·)ad |gλ +g−λ
is non-degenerate.

(c) In particular, (·, ·)ad |t is non-degenerate (t = g0 ) (warning: this is not (·, ·)ad t , which is 0),
∼
so it defines an isomorphism ν : t → t∗ , ν(t)(t0 ) = (t, t0 )ad and equippes t∗ with the induced
inner product, i.e. (ν(t), ν(t0 ))ad := (t, t0 )ad .

(d) If α ∈ R, then −α ∈ R

Proof. (·, ·)ad is non-degenerate on gα + g−α . But by (b) (gα , gα )ad = 0 if α 6= 0. This
implies g−α is non-zero (and isomorphic to (gα )∗ ).

(e) If x ∈ gα , y ∈ g−α , then

[x, y] = (x, y)ad · ν −1 (α) ∈ t. (7)

Proof. ν([x, y]) ∈ t∗ , so it is determined by ([x, y], t)ad (7) follows from

(∗)
ν([x, y])(t) = ([x, y], t)ad = (t, [x, y])ad = (x, y)ad · α(t),

(∗) as (·, ·)ad is an invariant form.

(f) Let eα ∈ gα be non-zero, and pick e−α ∈ g−α s.t. (eα , e−α )ad 6= 0 (possible as (·, ·)ad |gα +g−α
is non-degenerate), so [eα , e−α ] = (eα , e−α )ad ν −1 (α), by (7). This implies

[ν −1 (α), e±α ] = ±α(ν −1 (α))e±α = ±(ν −1 (α), ν −1 (α))eα = ±(α, α)eα .

Claim: (α, α) 6= 0.

Proof. Suppose (α, α) = 0. Put m := heα , e−α , ν −1 (α)i. Then [m, m] = Cν −1 (α), and so
m is solvable. But then Lie’s Theorem implies that ad[m, m] acts by nilpotent operators
on g, i.e. ad ν −1 (α) is nilpotent (⇐⇒ all eigenvalues are 0). But ν −1 (α) ∈ t, so acts
diagonalizable, by definition. Hence ν −1 (α) = 0, i.e. α = 0, contradiction.

−1
Therefore, we can define hα = 2ν(α,α)
(α) 2
∈ t and rescale eα so that (eα , e−α )ad = (α,α) .
Exercise: Check that the linear map m → sl2 defined by eα 7→ e, e−α 7→ f, hα 7→ h is an
isomorphism of Lie algebras.

(g) dim g−α = 1, ∀α ∈ R.

Proof. Pick mα = heα , hα , e−α i as above, so mα ∼

= sl2 , and suppose dim g−α > 1, then

g−α → Cν −1 (α), x 7→ [eα , x]

must have a non-trivial kernel, i.e. there exists v ∈ g−α s.t. ad(eα )v = 0, i.e. v is a highest
weight vector with weight -2 as ad(hα )v = −α(hα )v = −2v (by definition α(hα ) = 2), but
dim g < ∞, so contradiction (highest weights of finite-dimensional sl2 representations are
in N).

30 Robert Laugwitz & Henning Seidler

 4 Structure Theory

Note that the proof of part (iii) is still incomplete. It will be proven in the following theorem,
which continues the structure theorem.

Exercise 4.12: Check Theorem 4.11 for the classical Lie algebras sln , so2n , so2n+1 , sp2n .

Theorem 4.13 (Structure Theorem continued):

2(α,β)
(v) If α, β ∈ R, then (α,α) ∈ Z.

(vi) If α ∈ R and kα ∈ R, then k = ±1.

L
(vii) k∈Z gβ+kα is an irreducible module for (sl2 )α = heα , hα , e−α i. In particular, the set

{kα + β | kα + β ∈ R, k ∈ Z} ∪ {0}
2(α,β)
is of the form β − pα, β − (p − 1)α, ..., β + qα, where p − q = (α,α) , the α-string through β.

Proof:

(v) Let q = max{k | β + kα ∈ R}, v ∈ gβ+qα , v 6= 0. Then [eα , v] ∈ gβ+(q+1)α = 0, and

[hα , v] = (β + qα)(hα )v. But

2ν −1 (α) 2(β, α)
(β + qα) = + 2q =: N,
(α, α) (α, α)
| {z }
hα

2(α,β)
so v is a highest weight vector with weight N ∈ N, as q ∈ N, this implies (α,α) ∈ Z.

(vii) Structure of sl2 -modules tells us that (ad e−α )r 6= 0, if 0 ≤ r ≤ 2(α,β)

(α,α) + 2q = N , and
N +1
(ad e−α )  that {β +
v = 0. It follows  qα − kα : 0 ≤ k ≤ N } ∪ {0} are all roots. So
2(α,β)
β + qα, β + (q − 1)α, ..., β − q + (α,α) α are roots. So, we show that there are no more
roots.
Let
 p = max{k  | β − kα ∈ R}, and w ∈ gβ−pα , w 6= 0, then [e−α , w] = 0, [hα , w] =
2(α,β)
(α,α) − 2p · w is the lowest weight vector of an (sl2 )α -module, so we get that

 
2(α, β)
β − pα, β − (p − 1)α, ..., β + p − α
(α, α)
2(α,β) 2(α,β)
are all roots. Put p0 = q + (α,α) , so p0 ≤ p by definition of p, i.e. q + (α,α) ≤ p, and by
2(α,β)
definition of q, p − (α,α) ≤ q. Hence equality.

2(α,kα)
(vi) If kα is a root, then as is (v) (kα,kα) = k2 ∈ Z, and 2(α,kα)
(α,α) = 2k ∈ Z, so it is enough to
show that α ∈ R implies 2α ∈ / R. If not, let v ∈ g−2α , v 6= 0. Then ad eα v ∈ g−α , but this
implies ad(eα )v = 0, as (·, ·)ad is non-degenerate on gα + g−α = Ceα + Ce−α , so v ∈ g−2α
is a highest weight vector of weight -4, a contradiction.

(iii) Finally, we prove that [gα , gβ ] = gα+β if α, β, α + β ∈ R. We have just shown that
L
k∈Z gβ+kα is irreducible mα -module, i.e. ad(eα ) : gβ+kα → gβ+(k+1)α is an isomorphism
if k < q. But gα+β 6= 0 implies q ≥ 1, so ad(eα )gβ = gβ+α

Robert Laugwitz & Henning Seidler 31

Lie Algebras and Their Representations

Definition 4.14: For α ∈ t∗ , define the reflection at α as

2(v, α)
sα : t∗ → t∗ , sα (v) = v − α.
(α, α)

Claim 4.15: Property (vii) of Theorem 4.13 says sα (β) ∈ R, for all α, β ∈ R.

Proof: Put r = 2(α,β)

(α,α) ∈ Z. If r ≥ 0, p = q + r ≥ r. If r ≤ 0, q = p − r ≥ −r. In both cases, we
get β − rα in the α-string through β.

Proposition 4.16: Recall that R spans t∗ .

(i) If α, β ∈ R then (α, β) ∈ Q.

(ii) If we pick a basis β1 , . . . , βl of t∗ with βi ∈ R and β ∈ R, then β =

P
ci βi with ci ∈ Q, i.e.
dimQ R = dimC t.

(iii) (·, ·) is positive definite on QR.

Proof:
(i) Since 2(α,β) 0
(β,β) ∈ Z, it is enough to show (β, β) ∈ Q if β ∈ R. Now let h, h ∈ t. Then by
structure theorem X
(h, h0 )ad = trg (ad h ad h0 ) = α(h)α(h0 )
α∈R
So if λ, µ ∈ t∗ , we have
X X
(λ, µ) = (ν −1 (λ), ν −1 (µ))ad = α(ν −1 (λ))α(ν −1 (µ)) = (λ, α)(µ, α)
α∈R α∈R
2. Dividing by 41 (β, β)2 we get
P
so (β, β) = α∈R (α, β)

4 X  2(α, β) 2
= ∈ Z ⇒ (β, β) ∈ Q.
(β, β) (β, β)
α∈R

(ii) Let β1 , . . . , βl be a basis of t∗ consisting of roots and let B = ((βi , βj )ij be the matrix of
P
the bilinear form. Since (·, ·) is non-degenerate, det B 6= 0. Now if β = ci βi ∈ R, we
P
have (β, βi ) = j cj (βj , βi ) but
   
c1 (β, β1 )
 ..  −1  .. l
 . =B · ∈Q

.
cl (β, βl )
P
(iii) If λ ∈ QR, then λ = ci βi with ci ∈ Q by (ii), so (λ, α) ∈ Q for all α ∈ R, by (i). But
then X
(λ, λ) = (λ, α)2 ≥ 0.
α∈R
And if (λ, λ) = 0, then (λ, α) = 0 for all α ∈ R, hence λ = 0 as R spans t∗ and (·, ·) is
non-degenerate.

Exercise 4.17: Let (·, ·) be a non-degenerate, bilinear, symmetric form and let B defined as
in 4.16. Show that det B 6= 0.

32 Robert Laugwitz & Henning Seidler

 5 Root Systems

5 Root Systems
Definition 5.1: Let V be a vector space over R. Let (·, ·) be an inner product (here it is a
positive definite, bilinear, symmetric form). If α ∈ V , α 6= 0, write α∨ := (α,α)
2α
. Note that
∨
(α, α ) = 2.
Define sα : V → V by sα (v) = v − (v, α∨ )α (which is a linear map).

Lemma 5.2: The linear map sα is the reflection in the hyperplane orthogonal to α. In particu-
lar, all of its eigenvectors are 1, except of one which is −1. So s2α = 1 (⇐⇒ (sα + 1)(sα −1) = 0),
and sα = O(V, (·, ·)) the orthogonal group defined by (·, ·).

Proof: V = Rα ⊕ α⊥ where α⊥ = {v ∈ V : (α, v) = 0}. Furthermore, we have

sα (α) = α − (α, α∨ ) · α = α − 2α = −α.

And for each v ∈ α⊥ we have

2·α
sα (v) = v − (v, α∨ ) · α = v − · (v, α) = v.
(α, α) | {z }
=0

Definition 5.3: A root system R in V is a finite set R ⊆ V s.t.

(i) 0 ∈
/ R and RR = V (i.e. V = span(R)),

(ii) for all α, β ∈ R we have (α, β ∨ ) ∈ Z,

(iii) sα R ⊆ R for all α ∈ R.

A root system is called reduced if

(iv) α, kα ∈ R =⇒ k = ±1.

Note that (iii) implies that sα (α) = −α ∈ R.

L
Example 5.4: If g is a Lie algebra and g = t + α∈R gα is its weight space decomposition,
then (R, RR) is a reduced root system.

Definition 5.5: Let W := h{sα : α ∈ R}i ⊆ GL(V ). The group W is called the Weyl group of
R.

Lemma 5.6: The Weyl group W is finite.

Proof: Since the sα are invertible, and sα R ⊆ R by 5.3(iii), each sα permutes the elements of
R which is finite. So there exists an embedding W ,→ Sym(|R|). But by 5.3(i) this map is an
injection since RR = V and therefore, if sα , sβ act equally on R, they coincide on the whole of
V.

Definition 5.7: The rank of a root system R ⊂ V is defined as dimR V as a vector space.
An isomorphism of root systems (R, V ) → (R0 , V 0 ) is a bijective linear map ϕ : V → V 0 s.t.
ϕ(R) = R0 . Note that ϕ is not required to be an isometry (i.e. does not have to preserve the
inner product).

Robert Laugwitz & Henning Seidler 33

Lie Algebras and Their Representations

If (R, V ) and (R0 , V 0 ) are root systems, so is (R R0 , V ⊕ V 0 ). A root system which is not
`

isomorphic to a direct sum like this is called irreducible.

Example 5.8:

rk 1: A1 : The only rank 1 root system is V = R with inner product (x, y) = xy and roots
R = {α, −α}, α 6= 0. Its Weyl group is given by W = Z/2. We call this root system A1 .
This is the root system of sl2 .

−α α

rk 2: A1 × A1 : Take V = R2 with the usual inner product. Then R = {e1 , −e1 , e2 , −e2 } with
the standard basis vectors is a root system. Note that this is A1 × A1 and therefore not
irreducible. Here W = Z/2 × Z/2.
A2 : Let α = α∨ , β = β ∨ , (α, β) = −1. Then W = S3 . We call this root system A2 , it
appears as the root system of sl3 .
B2 : Let α = e1 , (α, α) = 1, β = e2 − e1 , (β, β) = 2, α, α + β are short roots, β, 2α + β long
roots. Then W is the symmetry group of the square, i.e. W = D8 , the dihedral group of
order 8. This is the root system of sp4 and so5 .
G2 : Also D12 appears as Weyl group of a root system, called G2 .

e2

β α+β
e1
α

(a) A1 × A1 (b) A2

3α + 2β
β α + β 2α + β β 3α + β
α
α

(c) B2 (d) G2

Figure 1: rank 2 root systems

Exercise 5.9: Check that all the above examples are root systems and that A2 , B2 , G2 are the
only irreducible roots systems of rank 2.

Lemma 5.10: Let R be a root system. Then R∨ = {α∨ | α ∈ R} is an root system.

/ R∨ . Also, since α∨ has the same direction as α, we get that RR = V .

Proof: Clearly, 0 ∈

34 Robert Laugwitz & Henning Seidler

 5 Root Systems

Notice (α∨ )∨ = α. Thus, for α, β ∈ R, (α∨ , (β ∨ )∨ ) = (β, α∨ ) ∈ Z. Lastly,

sα∨ (v) = v − (v, α)α∨

2α
= v − (v, α)
(α, α)
2α
= v − (v, )α
(α, α)
= sα (v).

Hence, sα∨ R = sα R ⊂ R.

Definition 5.11: R is simply laced if all the roots are of the same length (e.g. A1 , A1 × A1 , A2 ,
not B2 , G2 ).

Exercise 5.12: If R is simply laced, then (R, V ) is isomorphic to a root system (R0 , V 0 ) with
(α, α) = 2, for all α ∈ R0 (i.e. α = α∨ ).
√
Solution: Say |(α, α)| = (α, α) = λ, for all α ∈ R as R is simply laced. Now define α := λ2 α
p

(if this is an root system, then it is isomorphic to R via multiplication by a scalar). Then
(α, α) = 2, and α∨ = α. R := {α | α ∈ R} is also a root system. Clearly, 0 ∈ / R, RR = V .
Further,
√ !2 2
∨ 2 λ
(α, β ) = (α, β ∨ ) = (α, β ∨ ) ∈ Z,
λ 2
where we apply that R is simply laced. Lastly, notice that sα = sα .

Definition 5.13: A lattice L is a finitely generated free abelian group (∼ = Zl ) with bilinear
form (·, ·) : L × L → Z s.t (L ⊗Z R, (·, ·)) is an inner product space. A root of L is an α ∈ L
with (α, α) = 2. Write

RL = {l ∈ L | (l, l) = 2} = {l ∈ L | l∨ = l}

for the set of roots of L. Note that α ∈ RL implies sα (L) ⊆ L.

Lemma 5.14: The set of roots RL is a root system in RRL . Moreover, it is simply laced.

Proof: Everything is obvious, except: RL is finite. But RL is the intersection of a compact set,
the sphere {v ∈ RL | (v, v) = 2}, with the discrete set L, so it is finite.

We say L is generated by roots if ZRL = L. Note, this implies that L is an even lattice, i.e.
(l, l) ∈ 2Z for all l ∈ L.

Example 5.15:
(i) Let L = Zα, (α, α) = λ. If λ = 2 and RL = {±α}, then L is generated by roots. If
k2 λ
2 6= 1, for all k ∈ Z, then RL = ∅.

(ii) An : Consider Zn+1 = n+1

L
i=1 Ze1 and (ei , ej ) = δij as a square lattice. Define
(n+1 )
ai = 0 ∼
X X
n+1
= Zn .

L= l∈Z : (l, e1 + . . . + en+1 ) = 0 = ai ei : ai ∈ Z,
i=0

Robert Laugwitz & Henning Seidler 35

Lie Algebras and Their Representations

Now, RL = {ei − ej : i 6= j} and so |RL | = n(n + 1), ZRL = L. If α = ei − ej then

n+1
! n+1 n+1
! n+1
!!
X X X X
sα xk e k = xk e k − ei , xk ek − ej , xk e k (ei − ej )
k=1 k=1 k=1 k=1
n+1
X
= xk ek − (xi − xj )(ei − ej )
k=1
= x1 e1 + ... + xj ei + ... + xi ej + ... + xn+1 en+1 ,

i.e. sei −ej swaps ith and j th coordinate. Hence

W = hsei −ej : i, j = 1, . . . , ni = Sn+1 ,

the symmetric group of n + 1 letters. Call (RL , RL) root system of the type An , where n
is the rank of the root system. Note that An is irreducible.
Exercise: Check these statements, then draw L ⊆ Zn+1 and RL for n = 1, 2, check that
these agree with A1 , A2 as defined before. E.g. the roots system A1 is:

x2

α x1
α
L

Moreover, show that the root system of sln+1 is of type An .

(iii) Dn : Consider the square lattice Zn = ni=1 Ze1 , with (ei , ej ) = δij . Then RZn = {±ei ±
L

ej | i 6= j}. Set
n
( )
X X
L = ZRZn = l = ai ei ai ∈ Z, ai ∈ 2Z (i.e. even) ,
i=1

then sei −ej swaps the i-th and j-th component as before, and

sei +ej (x1 e1 + ... + xn en ) = x1 x1 + ... − xj ei ... − xi ej ... + xn en ,

i.e. sei +ej swaps the i-th and j-th component and changes signs of these components. If
L has this form, we say (RL , ZRL ) is of type Dn . In this case, |Dn | = 2n(n + 1) and

W = (Z/2Z)n−1 n Sn ,

where (Z/2)n−1 is the subgroup with even number of sign changes.

Exercise: As before, check all these statements. But Dn is only irreducible if n ≥ 3. We
`
have the identities RD3 = RA3 , RD2 = RA1 RA1 . These are the root systems of so2n .

(iv) E8 : Let
X
Γn := {(k1 , ..., kn ) | ki ∈ 2Z and either all ki in Z or all in Z + 21 }.

36 Robert Laugwitz & Henning Seidler

 5 Root Systems

Consider α = ( 12 , ..., 12 ), then (α, α) = n4 , using the usual inner product. If α ∈ Γn and Γn
is an even lattice, then 8 divides n.
Exercise:

(a) Γ8n is a lattice.

(b) If n > 1, the roots of Γ8n are a root system of type D8n .
(c) RΓ8 = {±ei ± ej , i < j, 12 (±e1 ± ... ± e8 ), with even number of minus signs} is a root
system, the root system of type E8 . Note, |RΓ8 | = 8·7 2 4 + 128 = 240.
(d) Can you compute |W |? (It is 214 · 35 · 52 · 7).

Remark: A Lie algebra with root system RΓ8 should have dimension 8 + 240 = 248,
as dim t = dimR RR, and every root space has dimension 1. The smallest non-trivial
representation of such a Lie algebra would also have dimension 248 (adjoint representation).

Exercise 5.16: If R is a root system, α ∈ R, then α⊥ ∩ R is a root system.

Definition 5.17: We can apply this to RΓ8 . Take α = 21 (1, ..., 1), β = e7 + e8 :

(i) α⊥ ∩ RΓ8 is a root system, the root system of type E7 .

(ii) α⊥ ∩ β ⊥ ∩ RΓ8 = hα, βi⊥ ∩ RΓ8 is a root system, the root system of type E6 .

Exercise 5.18: Show |RE7 | = 126, |RE6 | = 72 and describe the corresponding lattices.

Theorem 5.19:

(i) “ADE” classification: The complete list of irreducible simply laced root systems is

An , n ≥ 1, Dn , n ≥ 4, E6 , E7 , E8 ,

and no two root systems in this list are isomorphic.

(ii) The remaining irreducible (reduced) root systems are denoted by

B2 = C2 , Bn , Cn , n ≥ 3, F4 , G2 ,

where

RBn = {±ei , ±ei ± ej , i > j} ⊆ Zn (root system of so2n+1 ),

RCn = {±2ei , ±ei ± ej , i > j} ⊆ Zn (root system of sp2n ),
∨
RC n
= RBn ,
WBn = WCn = (Z/2Z)n n Sn .

F4 : Put
1
Qn = {(k1 , ..., kn ) | ∀i, ki ∈ Z or ∀i, ki ∈ Z + }, and define
2
RF4 = {α ∈ Qn | (α, α) = 2 or (α, α) = 1}
1
= {±ei , ±ei ± ej , i > j, (±e1 ± e2 ± e3 ± e4 )}.
2

Robert Laugwitz & Henning Seidler 37

Lie Algebras and Their Representations

G2 : Consider the lattice L = {(x, y, z ∈ Z3 | x + y + z = 0)}, and define

RG2 ={α ∈ L | (α, α) = 2 or 6}

={±(e1 − e2 ), ±(e1 − e3 ), ±(e2 − e3 ),
± (2e1 − e2 − e3 ), ±(−e1 + 2e2 − e3 ), ±(−e1 − e2 + 2e3 )}.

Exercise 5.20: Check that F4 as defined in Theorem 5.19 is a root system.

We want to choose a “good” basis for V . Assume, we have f : V → R linear, s.t. f (α) 6= 0,
for all α ∈ R. Define α ∈ R positive if f (α) ≥ 0, and negative if f (α) < 0. Denote

R+ := {α ∈ R | f (α) > 0}, and R− := −R+ .

Definition 5.21: A root α ∈ R+ is simple if it is not the sum of two positive roots, i.e.
α 6= β + γ, for all β, γ ∈ R+ . Write Π = {α1 , ..., αl } for the set of simple roots. Note that using
a different function f may give other simple roots.

Example 5.22:
An : Here, R = {ei − ej | i 6= j}. Choose f (e1 ) = n + 1, f (e2 ) = n, ..., f (en+1 ) = 1, so
R+ = {ei − ej | i < j}. f (R+ ) ⊂ N, so if f (α) = 1, α must be simple, thus Π =
{e1 − e2 , e2 − e3 , ..., en − en+1 }.

Bn : R = {±ei , ±ei ± ej | i < j}. Put f (e1 ) = n, ..., f (en ) = 1, then R+ = {ei , ei ± ej | i < j}
and Π = {e1 − e2 , ..., en−1 − en , en }.

Cn : R = {±2ei , ±ei ± ej | i < j}. Using the same f as for Bn , we obtain R+ = {2ei , ei ± ej |
i < j} and Π = {e1 − e2 , ..., en−1 − en , 2en }.

Dn : R = {±ei ± ej | i < j}. Using the same f as for Bn , Cn , we obtain R+ = {ei ± ej | i < j}
and Π = {e1 − e2 , ..., en−1 − en , en−1 + en }.

E8 : Consider E8 with f (e1 ) = 28, f (ei ) = 9 − i, i = 2, ..., 8 (note 28 = 1 + 2 + 3 + 4 + 5 + 6 + 7),

then
1
R+ = {ei ± ej (i < j), (e1 ± e2 ± ... ± e8 ) (with even number of minus signs)},
2
1
Π = {e2 − e3 , ..., e7 − e8 ; (e1 + e8 − e2 − ... − e7 ); e7 + e8 }.
| {z } |2 | {z }
f =1 f =3
{z }
f =2

Exercise 5.23: Check all theres examples, pick nice functions f and also do E6 , E7 , F4 , G2 .

Proposition 5.24 (Dynkin):

(i) If α, β ∈ Π, then α − β ∈
/ R.

(ii) If α, β ∈ Π, α 6= β, then (α, β ∨ ) ≤ 0.

(iii) Every α ∈ R+ can be written as α =

P
ki αi , with αi ∈ Π and ki ∈ Z≥0 .

(iv) Simple roots are linearly independent (i.e. the sum in (iii) is unique). Remark: This shows
that Π is the desired “nice” basis for V .

38 Robert Laugwitz & Henning Seidler

 5 Root Systems

(v) If α ∈ R+ \ Π, then there exist β ∈ Π, s.t. α − β ∈ R+ .

`
(vi) R irreducible ⇐⇒ Π is indecomposable, i.e. Π 6= Π1 Π2 with (Π1 , Π2 ) = 0.

Proof: Exercise. Either case by case checking, or finding an uniform proof from the axioms of
a root system (see e.g. (Kac, 2010, Thm 17.1)).

Definition 5.25: Let Π = {α1 , ..., αl }. Define aij = (αi , αj∨ ), A = (aij )1≤i,j≤l is the Cartan
matrix.

Proposition 5.26: The Cartan matrix A = (aij ) of a root system satisfies the following
properties:

(i) aij ∈ Z, for all i, j, aii = 2, aij ≤ 0 if i 6= j;

(ii) aij = 0 ⇐⇒ aji = 0;

(iii) det A > 0;

(iv) all principal subdeterminants of A have positive determinant.

Proof: (i), (ii) have been proven before.

(iii):  2 
(α1 ,α1 ) 0

A= .. 
 ((αi , αj )) ,
 . 
2
(αl ,αl )

where det((αi , αj )) > 0 as it is the Gram matrix of a positive definite bilinear form. For (iv)
notice that the principal subdeterminants are matrices of exactly the same form, thus also
have positive determinant (or argue that the restriction of the bilinear form to hα1 , ...αl−k i,
k = 0, ..., l − 1, is also positive definite).

We can draw A as a graph using so-called Dynkin diagrams. In these diagrams, vertices are
simple roots, and edges are given by aij aji lines joining simple roots αi and αj . Note that for
irreducible root systems on the following values appear:

1, if simply laced,


aij aji = 2, appears in Bn , Cn , F4


3, appears in G2 .

If aij aji = 2 or 3 put an arrow in the direction of the short root. The Dynkin diagrams of all
the root systems (classification in Theorem 5.19) are shown in Figure 2.

Exercise 5.27: Show that the Dynkin diagrams are as claimed in Figure 2.

Exercise 5.28: If (R, V ) is an irreducible root system with positive roots R+ and simple roots
Π, then there exists an unique positive root θ ∈ R+ , s.t. for all αi ∈ Π θ + αi ∈
/ R. θ is called
the highest root. Note, as sα θ ∈ R, (αi , θ) < 0, ∀i.

Solution: Examine the roots systems one by one (later, we will give a uniform proof of this
statement). E.g. for An , take θ := e1 − en+1 .

Robert Laugwitz & Henning Seidler 39

Lie Algebras and Their Representations

An

Bn

Cn

Dn

E6

E7

E8

F4

G2

Figure 2: Dynkin diagrams

Define the extended Cartan matrix à by setting α0 = −θ, à = (aij )0≤i,j≤l , where aij =
2(αi ,αj )
(αj ,αj )= (αi , αj∨ ).

Example 5.29:
2 −2


A1 : A = (2), take θ = α, as this is the only positive root, then à = −2 2 .
 2 −1 −1

−1 2 −1
 . 
An : The extended Cartan matrix is Ãn =  −1 . . , if n > 1. The Dynkin diagram
..
 
. −1
−1 −1 2
of Ãn is

Corollary 5.30: Notice that à satisfies:

(i) aij ∈ Z, for all i, j, aij ≤ 0 if i 6= j;

(ii) aij = 0 ⇐⇒ aji = 0;

(iii) det à = 0, and all principal subdeterminants A of à have det A > 0.

40 Robert Laugwitz & Henning Seidler

 5 Root Systems

Proof: (i) and (ii) follow directly from the properties of A. For (iii) notice that Π ∪ {θ} is not
linearly independent.

B̃n , α0 = −e1 − e2
C̃n , α0 = −2e1

D̃n , α0 = −e1 − e2

Ẽ6

Ẽ7

Ẽ8
F̃4
G̃2

Figure 3: Extended Dynkin diagrams Ã

Exercise 5.31: Write down the highest root θ and the extended Dynkin diagram à for all
types of root systems.

Solution: See Figure 3.

Exercise 5.32:
(i) Show the corresponding Dynkin matrix to

(2)
also has determinant 0. We call this matrix twisted Ãn , denoted by Ãn .
(ii) The Dynkin diagram of AT is the Dynkin diagram of A with the arrows reversed.

Theorem 5.33: An irreducible (i.e. connected) Dynkin diagram, and hence an indecomposable
Cartan matrix is one of An , Bn , Cn , Dn , E6 , E7 , E8 , F4 , G2 .

Proof:
(i) Classify the rank 2 Dynkin diagrams. These have a Cartan matrix of the form
 
2 −a
A= =⇒ det A = 4 − ab > 0
−b 2
ab = (0, 0), (1, 1), (2, 1), (1, 2), (3, 1), (1, 3)
| {z } | {z } | {z }| {z }
A1 ×A1 A2 B2 G2 ,

Robert Laugwitz & Henning Seidler 41

Lie Algebras and Their Representations

are the only possibilities, as aij aji ∈ {0, 1, 2, 3}.

(ii) Observe that any subdiagram of a Dynkin diagram is a Dynkin diagram (follows from the
fact that the principal subminors have determinant > 0).

(iii) Dynkin diagrams contain no cycles. To prove this, let α1 , ..., αn be distinct simple roots
and consider
n
X αi
α= p ∈ V, then
i=1
(αi , αi )
X 2(αi , αj )
0 < (α, α) = n + p
i<j
(αi , αi )(αj , αj )
X√
=n− aij aji ,
i<j

by definition of aij = (αi , αj∨ ) and the fact that (αi , αj∨ ) ≤ 0 if i 6= j.
P √
So i<j aij aji < n. But now if there is a cycle on α1 , ...αn , we must have n or more
P √
edges, i.e. i<j aij aji ≥ n, a contradiction.

(iv) The Dynkin diagram does not contain any extended Dynkin diagrams.

(v) If the diagram is simply laced (i.e. no 2 or 3 bonds), then it is of type A, D or E. To

prove this, suppose such a diagram is not of type A, D or E. As D̃4 is not contained in
any Dynkin diagram, we only have triple branch points. Denote by Tp,q,r the diagram

q
p
Tp,q,r

having 3 branches with p, q, and r edges (e.g. E8 = T5,3,2 ). Exercise: finish the proof by

(a) arguing that, as a Dynkin diagram does not contain Ẽ6 , Ẽ7 , Ẽ8 , we are left with
Dn (n ≥ 4) or En (n = 6, 7, 8),
(b) showing that det Tp,q,r = pq + pr + qr − pqr by induction on p + q + r, and hence argue
that Dn , E6 , E7 , E8 are the only possibilities for this. E.g. det E8 = 15+10+6−30 = 1.

(vi) Consider the case if the diagram is not simply laced.

Exercise: If G2 is a subdiagram, then the diagram is G2 . Hint: We have seen

does not appear, show

also do not appear.

(2)
Finally, if a 2 bond occurs, only one of such as C̃n and Ãn are not contained in a Dynkin
diagram, and then no branches occur as B̃n is not contained. If the double bond is in the
middle, the diagram has to be F4 (otherwise, it contains F̃4 ). If the double bound is not
in the middle, we have Bn or Cn .

42 Robert Laugwitz & Henning Seidler

 6 Existence and Uniqueness

Exercise 5.34: Compute the determinant of all the Cartan matrices. For example:

A1 : A1 = (2) has det A1 = 2,

2 −1


A2 : A2 = −1 2 has det A2 = 3, etc.

Remark 5.35: Notice that SLn+1 = {X | det X = 1} has center isomorphic to the cyclic
group of order n + 1. The order of this is the determinant of the Cartan matrix. In general, the
order of the center of the simply connected group with Lie algebra g whose Cartan matrix is A
is det A.

6 Existence and Uniqueness

Setting: For a semisimple Lie algebra g, we chose a maximal torus t and obtained a direct
sum composition M
g=t⊕ gα .
α∈R

Further, we chose f : RR → R, giving us R+ , and thus the simple roots Π = {α1 , . . . , αl }. From
this we defined the Cartan matrix A.

(A) Independence of choices

Theorem 6.1: Let char k = 0 and k = k. All maximal tori are conjugate, i.e. if t and
t0 are maximal tori of g, then there exists some g ∈ (Aut g)◦ = {g ∈ GL(g) | g : g →
g is a Lie alg. homomorphism}◦ , such that gt = t0 . Note that Aut(g) is an algebraic group
(with Lie(Aut(g)) = g). Aut(g)◦ is defined as the connected component which contains
the 1 ∈ Aut(g).

Theorem 6.2: All choices of positive roots R+ are conjugate. Let (R, V ) be a root
system. For f1 , f2 : V → R (s.t. fi (α) 6= 0 ∀α ∈ R) denote the corresponding sets of
positive roots by R1+ , R2+ . Then there exists a unique w ∈ W (the Weyl group), such that
wR1+ = R2+ . Hence wΠ1 = Π2 and thus they have the same Cartan matrix.

Corollary 6.3: g determines the Cartan matrix, regardless of the choices of the maximal
torus and the function f .

(B) Uniqueness

Theorem 6.4: Let gi for i = 1, 2 be semisimple Lie algebras with respective ti , Ri , Ri+ , Πi , Ai .
Assume that after reordering of indices, we have A1 = A2 . Then there exists some isomor-
∼
phism ϕ : g1 → g2 , such that ϕ(t1 ) = t2 , ϕ(R1 ) = R2 , etc.

(C) Existence

Theorem 6.5: Let A be a Cartan matrix. Then there exists a semisimple Lie algebra
with A as its Cartan matrix.

Robert Laugwitz & Henning Seidler 43

Lie Algebras and Their Representations

Remark 6.6: We already know this, except for G2 , F4 , E6 , E7 , E8 .

Let g be a semisimple Lie algebra. Choose some non-zero Ei ∈ gαi and Fi ∈ g−αi , such that
(Ei , Fi )ad = (αi2,αi ) (this is possible as (·, ·)|gα is non-degenerate (4.11) and (gα , gα ) = 0) and let

2ν −1 (αi ) 2(αi , αj )
Hi = ∈ t, aij = (αi∨ , αj ) = .
(αi , αi ) (αi , αi )

We have [Hi , Hj ] = 0 (since the Hi are in the torus, thus commuting), [Ei , Fi ] = Hi , [Ei , Fj ] = 0
if i 6= j (since [Ei , Fi ] ∈ gαi −αj , which is no root for i 6= j, making gαi −αj = 0) and

αj (2ν −1 (αi )) 2(αj , αi )

[Hi , Ej ] = αj (Hi )Ej = · Ej = · Ej = aij Ej and so [Hi , Fj ] = −aij Fj
(αi , αi ) (αi , αi )

Let n+ = α∈R+ gα and n− = α∈R− gα . So g = n+ ⊕ t ⊕ n− .

L L

Lemma 6.7: The Ei generate n+ and the Fi generate n− . Hence, {Ei , Fi } generates g (as a
Lie algebra).

ki αi ∈ R+ , so ki ≥ 0. Define the height of α as ht(α) =

P P
Proof: Let α = ki ≥ 0. Induct
on ht(α) that gα is spanned by linear combinations of the Ei . If ht(α) = 1, then α = αi ∈ Π
for some i, so gαi = CEi . If ht(α) > 1 we know that there exists some αi ∈ Π, such that
β = α − αi ∈ R+ (by 5.24(v)). But we know [gαi , gβ ] = gα by structure theorem, as αi , α, β
are all roots. So by induction hypothesis, gαi , gβ are generated by some Ei , and thus also gα .
Inductively, this proves that the Ei generate n+ as a Lie algebra. The argument for n− = hFi i is
similar. Finally [Ei , Fi ] = Hi . But t = hHi : i = 1, . . . , li. So Ei , Fi generate g = n+ + t + n− .

Now let A be a generalized Cartan matrix, i.e. aii = 2, aij = 0 ⇔ aji = 0, aij ∈ −N if i 6= j.

Definition 6.8:

(i) Let e
g denote the Lie algebra with generators Ei , Fi , Hi , where i = 1, . . . , l, and the relations

[Hi , Hj ] = 0,
[Hi , Ej ] = aij Ej ,
[Hi , Fj ] = −aij Fj ,
[Ei , Fj ] = 0, if i 6= j,
[Ei , Fi ] = Hi ,

as above. (Remark: so e
g is basically a “bunch of sl2 glued together.”)

(ii) Let g be the quotient of e

g by the additional relations

(ad Ei )1−aij Ej = 0 and (ad Fj )1−aij Fj = 0, if i 6= j,

the so-called Serre relations (though discovered by Harish-Chandra, Chevalley). (Note

that if aij = 0, then these relations become [Ei , Ej ] = 0; if aij = −1, [Ei , [Ei , Ej ]] = 0.)

Exercise 6.9: Check that the Serre relations hold for the classical groups sln , so2n , so2n+1 , sp2n .

44 Robert Laugwitz & Henning Seidler

 6 Existence and Uniqueness

Theorem 6.10:

(i) If A is indecomposable, then e g has a unique maximal ideal and g is its quotient, i.e. g is
simple (not necessarily finite-dimensional).

(ii) Hence, if g is a finite-dimensional semisimple Lie algebra with Cartan matrix A, then the
g → g (Ei 7→ Ei , Fi 7→ Fi ) factors through g, is surjective and gives an isomorphism
map e
∼
g → g.

Remark: (i)⇒(ii) follows from Lemma 6.7, (ii) implies uniqueness as stated in (6.4).

The above theorem shows that existence is equivalent to the following theorem:

Theorem 6.11: g is finite dimensional if and only if A is a Cartan matrix.

Definition 6.12: In general, g is called a Kac-Moody algebra.

Theorem 6.13 (Presentation of W ): Write ri = sαi , then

W = hr1 , . . . , rl | ri2 = 1, (ri rj )mij = 1i

is a presentation of W , where
aji aij 0 1 2 3
mij 2 3 4 6

Example 6.14: Consider the simple cases

i j
ri rj = rj ri
i j
ri rj ri = rj ri rj

For An , we have the relations ri ri+1 ri = ri+1 ri ri+1 and ri rj = rj ri if j 6= i + 1, visualized as

= .

Exercise 6.15: Check for each root system that the relations claimed do hold. (Hint: it is
enough to show this for all rank 2 root systems).

Robert Laugwitz & Henning Seidler 45

Lie Algebras and Their Representations

7 Representations of Semisimple Lie Algebras

7.1 Classification of finite-dimensional representations
From now on, let g be a semisimple Lie algebra, so
M
g=t⊕ gα = t ⊕ n + ⊕ n − .
α∈R

Furthermore, let V be a finite-dimensional representation of g.

Proposition 7.1: In the above setting, we have:

L
(i) V = λ∈t∗ Vλ , where Vλ = {v ∈ V : tv = λ(t)v ∀v ∈ V }, the weight space decomposition
w.r.t. t (i.e. t acts semisimply on V ).

(ii) If Vλ 6= 0, then λ(hα ) ∈ Z for all α ∈ R. (Recall that we had (sl2 )α = heα , hα , e−α i,
hα = ν −1 (α∨ ), for all α ∈ R).

Proof: As V is a finite-dimensional representation, it is also a finite dimensional representation

for (sl2 )α , so hα acts diagonalizable on V by the sl2 -theory and λ(hα ) ∈ Z. As the hα span t,
(i) follows immediately.

Definition 7.2: Let R be a root system with simple roots α1 , . . . , αl .

(i) Set Q = ZR = li=1 Zαi , the lattice of roots of R.

L

(ii) Set P = {γ ∈ QR | ∀α ∈ R : (γ, α∨ ) ∈ Z} = {γ ∈ QR | ∀i : (γ, αi∨ ) ∈ Z}, the lattice of

weights of R.

Remark 7.3: Note, if β, α ∈ R, then (β, α∨ ) ∈ Z, so Q ⊆ P . Notice also (γ, α∨ ) = γ(hα ), so

if V is a finite dimensional-representation of g and Vλ 6= 0, then λ ∈ P by Proposition (7.1ii).

Exercise 7.4:

(i) Show |P/Q| < ∞, in fact |P/Q| = det A, where A is the Cartan matrix of g.

(ii) Show that the Weyl group W acts on t∗ , and W · P ⊆ P , hence W acts on P .

Example 7.5: Consider sl2 with R = {±α}. Then Q = Zα. Since (α, α) = 2, this means
P = Z α2 . So here we have |P/Q| = 2 = det(2) = det A.

Definition 7.6: If V is a finite-dimensional representation of g, define the character of V

X
ch V = dim Vλ eλ ∈ Z[P ]
λ∈P

where eλ is a formal symbol, basis for Z[P ], with eλ · eµ = eλ+µ .

α
Example 7.7: For sl2 , we have P = Z α2 . Write z = e 2 , then

z n+1 + z −(n+1)
ch L(n) = z n + . . . + z −n = .
z + z −1

46 Robert Laugwitz & Henning Seidler

 7 Representations of Semisimple Lie Algebras

Now look at the adjoint representation of V = g for sl3 . Put w = eα1 and z = eα2 . Then

ch V = 2 + z + w + z −1 + w−1 + zw + z −1 w−1 ,

For the root lattice we have the following picture:

α2 α1 + α2
1 1

0 α1
1 2 1

1 1

where the numbers next to the roots indicate the dimensions of the root spaces. Note that this
picture is S3 invariant.

Proposition 7.8: Let V be a finite-dimensional representation of g, then dim Vλ = dim Vwλ

for any w ∈ W , i.e. ch V is W -invariant (ch V ∈ Z[P ]W ).

Proof Sketch 1. If G is an algebraic group with g = Lie(G) and T is the subgroup with
t = Lie(T ) (e.g. g = sp2n , G = SP2n , T are the diagonal matrices in SP2n ), then W = N (T )/T
(we do not prove this result; for example in sln , T is the set of diagonal matrices, N (T ) are
the basis matrices Eij (monomial matrices) and N (T )/T = Sn ), so for any w ∈ W there exists
ẇ ∈ N (T ), such that ẇT = w. Now if G acts on V (always if G is simply connected), then
ẇ(Vλ ) = Vwλ as tẇ(v) = ẇw−1 tw · v = ẇ(λ(w−1 tw)v) = λ(w−1 tw)ẇv.
   
0 1 2 −1 0
Example 7.9: ṡ = , ṡ = ∈ T \{1} shows that you cannot embed W ,→ G
−1 0 0 −1
in general and W itself does not act on V . Instead we have a small 2-group (∼
= (Z/2)l at worst),
that intervenes. So W is the normalizer of the maximal torus modulo the torus.

Proof Sketch 2. Mimic this in g. How to see

 

0 −1
 
1 0

1 −1

1 0
 X xn
= = exp(f ) exp(−e) exp(f )? where exp(x) = 
1 0 1 1 0 1 1 1 n!
n≥0

This is only easy for nilpotent matrices. So for each not α, define

ṡα = exp(fα ) exp(−eα ) exp(fα ).

The following steps (exercise) finish the proof:

Robert Laugwitz & Henning Seidler 47

Lie Algebras and Their Representations

(i) If V is a finite-dimensional representation of g, then eα , fα obviously act nilpotently on

V , so ṡα : V → V is a well-defined finite sum.

(ii) ṡ2α = εα : Vλ → Vλ , where ε2α = 1. εα is multiplication by a scalar. Determine it explicitly

in terms of λ (denote the scalar by sα ).

(iii) Then ṡα Vλ = Vsα λ .

Remark 7.10: We do not need V to be finite-dimensional for this, just need that each eα , fα
acts locally nilpotent (some x : V → V acts locally nilpotent if for all v ∈ V there exists N ∈ N
s.t. xN v = 0).

Exercise 7.11: Show that this is equivalent to V splitting up – as an (sl2 )α -module – into a
direct sum (possibly infinite) of finite-dimensional (sl2 )α -modules, for all α ∈ Π. Such a V is
called integrable.

In the following, all the statements for Lie algebras and their proofs also work in the case
of Kac-Moody algebras if whenever the assumption of finite dimension of V is made, this is
replaced by the condition of V being integrable.

Proof Sketch 3. The statement is actually obvious from the sl2 -theory: Consider V as a rep-
resentation of (sl2 )α × t, then V breaks up into a direct sum of strings, each of which is of the
form
λ, λ − α, . . . , λ − mα,
where m = λ(hα ). Such a string is obviously sα invariant.

For µ, λ ∈ t∗ write µ ≤ λ to mean λ − µ =

P
Definition 7.12: ki αi , ki ∈ N. Graphically,
this means that Q≤λ = {µ ∈ P | µ ≤ λ} is the set of lattice points in an obtuse cone.

Q≤λ

Definition 7.13: Let V be a representation of g, we say

(i) The weight of a vector 0 6= v ∈ V is defined as λ if v ∈ Vλ , write wt(v) = λ in this case.

(ii) λ ∈ P is a highest weight if Vλ 6= 0 (i.e. λ is a weight) and if Vµ 6= 0, then µ ≤ λ.

(iii) Say v ∈ Vγ is a singular vector if v 6= 0 and eα v = 0 for all α ∈ R+ . Note that wt(eα v) =
α + β > β, if eα v 6= 0. (This follows from gα Vλ ⊂ Vλ+α , as for x ∈ g we have hβ xv =
([hβ , x]+xhβ )v = (α(hβ )+λ(hβ ))v). So if γ is a highest weight, all 0 6= v ∈ Vγ are singular
vectors.

(iv) A weight µ is an extremal weight if wµ is a highest weight for some w ∈ W .

48 Robert Laugwitz & Henning Seidler

 7 Representations of Semisimple Lie Algebras

(v) Set

P + = {λ ∈ P | (λ, α∨ ) ≥ 0, ∀α ∈ R+ }
= {λ ∈ P | (λ, αi∨ ) ≥ 0, ∀αi ∈ Π}

and call P + the cone of dominant weights.

In the picture for the sl3 root lattice, all points of the outside hexagon are extremal weights,
and α1 + α2 is highest weight. Note that if V is finitely-dimensional, then highest weights exist,
what implies that singular vectors exist.

Theorem 7.14: Let g be a semisimple Lie algebra over C.

(A) (Complete reducibility) If V is a finite-dimensional representation of g, then V is a

direct sum of irreducibles.

We have P + ∼
= {irreducible f.-d. representations of g} via λ 7→ L(λ). More precisely:

(B) Let V be a finite-dimensional irreducible representation of g, v ∈ Vλ a singular vector,

then:

(i) Vλ = C · v, i.e. dim Vλ = 1.

(ii) If Vµ 6= 0, then λ ≤ µ, so v is a highest weight vector (we say V has highest weight
λ).
(iii) λ(hi ) ∈ N for all i = 1, . . . , l, i.e. λ ∈ P + .

Moreover, if U is another irreducible finite-dimensional representation of g with highest

weight λ, and u ∈ Uλ , then there exists a unique isomorphism V → W sending v 7→ w.

(C) Given λ ∈ P + , there exists a finite-dimensional irreducible representation with highest

weight λ, denoted by L(λ)

(D) We will later give a closed formula for ch L(λ), the so-called Weyl character formula.

Corollary 7.15: ch L(λ) = eλ + µ<λ aµ eµ ∈ Z[P ], and hence {ch L(λ) | λ ∈ P + } are linearly
P

independent. Write ch L(λ) = mλ + µ<λ ãµλ mµ ∈ Z[P ], where mµ = γ∈W eγµ , the so-called
P P

monomial symmetric functions. As the mµ clearly form a basis of Z[P ]W , this shows that
ch{L(λ)} is a basis of Z[P ]W .

Corollary 7.16: If V, W are finite-dimensional, then V ∼

= W if and only if ch V = ch W .

Proof: Apply complete reducibility and the previous corollary.

Remark 7.17: Define wi ∈ P to be the dual basis to the simple coroots hαi , i.e. (wi , αj∨ ) = δij ,
for i = 1, . . . , l. These wi are called the fundamental weights. Using this notion, we can write

l
( l )
M X
+
P = Z≥0 wi = n i wi ni ≤ 0 .
i=1 i=1

Robert Laugwitz & Henning Seidler 49

Lie Algebras and Their Representations

Exercise 7.18:

(i) Compute P + for sln , so2n , so2n+1 , sp2n , . . . and draw the picture for A2 , B2 , G2 .

(ii) If λ ∈ P , then λ = i λ(hi )wi , where hi = ν −1 (αi∨ ) as always.

P

Example 7.19:

(i) For any Lie algebra g over C, we have the trivial representation C = L(0).
L
(ii) For g we have the adjoint representation g as a representation of itself, g = t + α∈R gα .
Here, a highest weight λ is a root s.t. λ + αi ∈ / R, for all i. Then λ = θ is the highest root
+
in R . Now Theorem 7.14 implies that θ is unique as promised (as g is simple if and only
if ad g is irreducible).
Take e.g. An−1 as an concrete example. Then θ = ε1 − εn , hi = Eii − E − i + 1, i + 1,
αi = εi − εi+1 , so θ(h1 ) = 1, θ(h2 ) = 0, . . . , θ(hn−2 ) = 0, θ(hn−1 ) = 1.
Exercise: Compute θ(hi ) for all simple Lie algebras.

(iii) Examples of representations of sln . P = Zn /Z ∼ = Zn−1 . Take Cn as standard rep-

P
resentation, with basis v1 , . . . , vn and weights e1 , . . . , en , ei = 0. Then the highest
weight is e1 (as e1 > e2 > · · · > en since e1 = (e1 − e2 ) + e2 , etc.) Then L(w1 ) = Cn ,
ch Cn = ee1 + . . . + een . If we write zi = eei , this becomes ch Cn = z1 + . . . + zn , and
Z[P ] = Z[z1±1 , . . . , zn±1 ]/(z1 · . . . · zn = 1).

Recall that if V, W are representations of g, then so is V ⊗W , where x ∈ g acts by x⊗1+1⊗x.

Hence, V ⊗ V is a representation, but σ : V ⊗ V → V ⊗ V, a ⊗ b 7→ b ⊗ a, commutes with the
g-action, so its eigenspaces are g-modules. σ 2 = 1, so the only eigenvalues are ±1, i.e. S 2 V (the
symmetric algebra V ⊗V /hv⊗w−w⊗v | v, w ∈ V i with product vw := 21 (v⊗w+w⊗v)) and Λ2 V
(the exterior algebra V ⊗ V /hv ⊗ w + w ⊗ v | v, w ∈ V i, with product v ∧ w := 12 (v ⊗ w − w ⊗ v))
are g-modules. In general, these must not be irreducible, but for sln they are.

Example 7.20: Let V = Cn as sln -module as above.

(i) Consider Λs V , s ≤ n − 1, this space has a basis {vi1 ∧ . . . ∧ vis | i1 < · · · < is } (if
{v1 , . . . , vn } is a basis of V ). Further, Λs V has weights ei1 +. . .+eis (as x(wi1 ∧. . .∧wis ) =
xwi1 ∧ wi2 . . . ∧ wis + . . . + wi1 ∧ . . . ∧ wis−1 ∧ xwis ), and check that Ei · (vi1 ∧ . . . ∧ vis ) = 0,
for all i, if and only if vi1 ∧ . . . ∧ vis = v1 ∧ v2 ∧ . . . ∧ vs , i.e. this is the only singular
vector. So Λs Cn is an irreducible sln -module with highest weight ws = e1 + . . . + es (the
s-th fundamental weight), as (ws , ei − ei+1 ) = δis . Thus, Λs Cn = L(ws ). For example,
Λn−1 Cn ∼ = (Cn )∗ = L(wn−1 ).

(ii) Consider S m Cn , the m-th symmetric power of Cn with basis {vi1 · . . . · vim | i1 ≤ · · · ≤ im }.
These are weight vectors with weights ei1 + . . . + eim .

Ei · (vi1 · . . . · vim ) = 0 ∀i ⇐⇒ vi1 · . . . · vim = v1 · . . . · v1 = v1m ,

so S m Cn is irreducible and isomorphic to L(mw1 ).

Exercise 7.21:

(i) Check all the statements in the above example, compute ch Λs Cn , and ch S m Cn .

50 Robert Laugwitz & Henning Seidler

 7 Representations of Semisimple Lie Algebras

(ii) Find closed formulas for

X
ch S m Cn · q m , and
m≥0
X
ch Λm Cn · q m .
m≥0

Exercise 7.22:
∼
(i) Let V be finite-dimensional g-module, then V ∗ ⊗ W → Hom(V, W ) as g-modules via
v ∗ ⊗ w 7→ (u 7→ w · v ∗ (u)) and Hom(V, V ) 6= 0 as it contains IdV . Note that V ∗ is a
g-representation by defining (g · f )(v) = −f (g · v) for all f ∈ V ∗ . (This comes from
differentiating the group action (gf )(v) = f (g −1 v).)
(ii) Show if V = Cn , g = sln , then V ⊗ V ∗ ∼
= sln ⊕ C (the sum of the adjoint and the trivial
∼
representation). In contrast, V ⊗ V = S 2 V ⊕ Λ2 V (in general).

Exercise 7.23: Let g = son or sp2l , 2l = n, V = Cn as g-representation in the obvious manner.

(i) Compute the highest weights of V .
(ii) V ∼
= V ∗ via the form defining g, so V ⊗ V has at least three summands (since it must have
the trivial subrepresentation). Show that it has exactly three summands, describe them
and find their highest weights.

In the following, we will prove Theorem 7.14. Let g be any Lie algebra with a non-degenerate
bilinear form (·, ·) (for example, g semisimple with the killing form). Let x1 , . . . , xN be a basis
of g, with x1 , . . . , xN dual basis, i.e. (xi , xj ) = δij .
xi xi , the Casimir of g.
P
Define Ω =

Lemma 7.24: If x ∈ g, then [Ω, x] = 0.

Proof: We will give the proof in two different ways: First,

X
xi xi , x

[Ω, x] =
i
X X
= xi [xi , x] + [xi , x]xi ,
i i

[xi , x] xj ,
P P
as [·, x] is a derivation. Now write = aij [xi , x] = bij xj . But then
aij = ([xi , x], xj ) = ([xj , xi ], x)
bij = ([xi , x], xj ) = ([xj , xi ], x) = −aji ,
using that (·, ·) is an invariant form. So [Ω, x] = xi xj aij + xj xi bij = 0.
P P
∼
We can also prove this without coordinates: We have maps of g-modules C ,→ End(g) →
g ⊗ g∗ via λ 7→ λ Id (i.e. 1 7→ xi ⊗ xi ), and the isomorphism g → g∗ is implied by the
P

non-degenerate form (·, ·). Further, the g-action on V implies a map of g-modules g → End(V )
which gives a g-module map
multiplication
g ⊗ g → End(V ) ⊗ End(V ) −−−−−−−−→ End(V ).
So we have a map of g-modules C → End(V ), which is the statement of the lemma as it maps
1 7→ Ω (i.e. Ω generates the trivial submodule of End(V )).

Robert Laugwitz & Henning Seidler 51

Lie Algebras and Their Representations

L
Now, let g be semisimple. Then g = t ⊕ α∈R gα , and let (·, ·) = (·, ·)ad be the killing form.
Choose a basis u1 , . . . , ul of t, and 0 6= xα ∈ gα . Denote the dual basis of t by u1 , . . . , ul and x−α
of g−α , i.e. (xα , x−α ) = 1. Normalize xα so that (xα , x−α ) = 1, then x−α = x−α and [xα , x−α ] =
ν −1 (α) (note that in general we had that x ∈ gα , y ∈ g−α implies [x, y] = (x, y)ν −1 (α)). Hence
X X
Ω= ui ui + (xα x−α + x−α xα )
α∈R
X X X
= i
ui u + 2 xα x−α + ν −1 (α).
α∈R+ α∈R+

Define
1 X
ρ= α,
2 +
α∈R

then we obtain X X
Ω= ui ui + 2ν −1 (ρ) + 2 xα x−α . (8)
α∈R+

Note that this is (up to normalisation) the same Casimir as defined before, in the case of sl2 .

Lemma 7.25: Let V be a g-module, v ∈ V a singular vector with weight λ (i.e. n+ v = 0,

tv = λ(t)v). Then Ωv = (|λ + ρ|2 − |ρ|2 ) · v.

Proof: Apply (8) to v, xα v = 0 for all α ∈ R+ , so

l
!
X
Ω·v = λ(ui )λ(ui ) + λ(2ν −1 (ρ)) ·v
i=1
= ((λ, λ) + 2(λ, ρ)) · v.

Hence, if V is irreducible, Ω acts on V by (λ, λ) + 2(λ, ρ) by Schur’s lemma.

7.2 The PBW theorem

Let g be any Lie algebra over k, where k is a field.

Definition 7.26: The universal enveloping algebra of g, U g is the associative algebra over k
generated by g and relations xy − yx = [x, y] for all x, y ∈ g.

More formally, if V is a vector space over k, then

M
TV = k + V + V ⊗ V + V ⊗ V ⊗ V + ... = V ⊗n
n≥0

is the tensor algebra over V , the free associative algebra generated by V . Multiplication V ⊗n ⊗
V ⊗m → V ⊗(n+m) is defined in the obvious way. Let J be the two-sided ideal in T g generated
by x ⊗ y − y ⊗ x − [x, y] for x, y ∈ g, then U g = T g/J.

Exercise 7.27: An enveloping algebra for g is a linear map ι : g → A, where A is an associative

algebra and ι a k-linear map s.t.

ι(x)ι(y) − ι(y)ι(x) = ι[x, y].

52 Robert Laugwitz & Henning Seidler

 7 Representations of Semisimple Lie Algebras

For example, for V a representation of g, A = End(V ), the action map ι is an enveloping algebra.
Show that U g is initial in the category of enveloping algebras, i.e. the diagram

Ug

>
∃! (9)
∨
g >A
ι

commutes.

Note that the Casimir Ω ∈ U g. Indeed, Ω ∈ Z(U g). Observe that T g is a graded algebra,
but the relations
x ⊗ y − y ⊗ x − [x, y]
| {z } | {z } | {z }
deg 2 deg 2 deg 1

are not homogeneous, so U g is filtered : Define (U g)n to be the span of elements of degree ≤ n
of g. Then (U g)n (U g)m ⊆ (U g)n+m . In particular, k ⊆ (U g)o , k + g ⊆ (U g)1 , . . ..

Exercise 7.28: Show that the above statements hold, and that if x ∈ (U g)n , y ∈ (U g)m , then
xy − yx = [x, y] ∈ (U g)n+m−1 .

Definition 7.29: For a filtration F0 ⊆ F1 ⊆ F2 ⊆ . . . we set

M
gr F = Fi /Fi−1

and call gr F the associated graded algebra.

Theorem 7.30 (PBW Theorem, Poincaré-Birkhoff-Witt):

L ∼
(i) gr U g = (U g)n /(U g)n−1 ←− Sg.

(ii) Equivalently, if x1 , . . . , xN is a basis of g, then {xa11 · . . . · xaNN | ai ∈ N} is basis of U g. In

particular, g ,→ U g.

Exercise 7.31:

(i) Show that the previous exercise (x ∈ (U g)n , y ∈ (U g)m , then xy−yx = [x, y] ∈ (U g)n+m−1 )
implies that we have a well-defined map Sg → gr U g extending the map g → g.

(ii) This map is surjective, i.e. the monomials above span U g. The content of the PBW
Theorem is then to show that this map also injects. We omit the proof of this.

Exercise 7.32: If V is a representation of g, v ∈ V , then the g-submodule of V generated by

v is just U g · v (the image of the map U g ⊗ Cv → V ).

Definition 7.33: A g-module V is a highest weight module for g if there exists a singular vector
v ∈ V (i.e. n+ v = 0, t · v = λ(t)v, for all t ∈ t and some λ ∈ t∗ ) such that V = U g · v.

Lemma 7.34: Observe that it follows that U n− · v = V .

Robert Laugwitz & Henning Seidler 53

 Lie Algebras and Their Representations

Proof: The PBW Theorem 7.30 implies that if x1 , . . . , xN is a basis of g then xa11 . . . xaNN spans
U g. Taking a basis x1 , . . . , xr for n− , xr+1 , . . . , xr+l of t, and xr+l+1 , . . . , xN of n− we see that
U g = U n− ⊗ U t ⊗ U n+ as a vector space. But U n+ · v = Cv (as n+ · v = 0) and U t · v = Cv,
thus U g = U n− .

Remark 7.35: If V is irreducible and finite-dimensional, then it is a highest weight module.

Proposition 7.36: Let V be a highest weight module for g (no necessarily finite-dimensional),
and let vΛ be a highest weight vector with highest weight Λ ∈ t∗ , then:
L
(i) t acts diagonalizable on V , and V = λ∈D(Λ) Vλ , where
X
D(Λ) = {Λ − ki αi | ki ∈ Z≥0 } = {µ ∈ t∗ | µ ≤ Λ}.

D(Λ) is called the descent of Λ.

(ii) VΛ = CvΛ , and all other weight spaces are finite-dimensional.

(iii) V is irreducible if and only if all singular vectors are in VΛ .

(iv) Ω acts on V as |Λ + ρ|2 − |ρ|2 .

(v) If vλ is any singular vector in V , then |λ + ρ| = |Λ + ρ|

(vi) There exist only finitely many λ such that Vλ contains a singular vector.

(vii) V contains a unique maximal proper submodule I, I is graded by t (i.e. I = ⊕λ∈t∗ (I ∩Vλ )),
and I is the sum of all proper submodules of V .
Proof:
(i),(ii) As V = U n− · vΛ , expressions of the form e−β1 e−β2 . . . e−βr vΛ span V , where βi ∈ R+ and
e−βi ∈ g−βi . But the weight of such an expression is Λ − β1 − β2 − . . . − βr (Exercise:
proof this, note tev = [t, e]v + etv = (−β(t) + Λ(t))ev). Whence (i) and (ii) hold as there
are only finitely many β ∈ R+ which sum up to a given weight λ ∈ Z≥0 R+ .

(iii) If vλ ∈ Vλ is a singular vector, then N = U g · vλ = U n− · vλ is a submodule of V , whose

weights, are in D(λ), by (i). But λ 6= Λ implies D(λ) ( D(Λ), so N is a proper submodule
as vΛ ∈/ N , i.e. V is not irreducible.
Conversely, if N ( V is a proper submodule, then, as tN ⊆ N , N is graded by t, and its
P P
weights are in D(Λ). Let λ = Λ − ki αi be a weight of N (αi ∈ Π) and ki minimal.
P
Then ki > 0 as otherwise Λ = λ and N = V . Now, if 0 6= v ∈ Nλ , then v is singular as
for α ∈ R+ , eα · v ∈ Nλ+α , but Nλ+α = 0 by minimality of
P
ki .

(vii) Any proper submodule of V is t-graded and does not contain vΛ . Therefore, the sum of all
proper submodules still does not intersect VΛ and is t-graded, so it is the maximal proper
submodule.

(iv) We know from 7.25 that for any singular vector vΛ we have

ΩvΛ = (|Λ + ρ|2 − |ρ|2 )vΛ .

Moreover, Ω is central, so Ωe−β1 . . . e−βr vΛ = e−β1 . . . e−βr ΩvΛ and these elements span V .
Therefore, we see that Ω acts by the same constant on all of V .

54 Robert Laugwitz & Henning Seidler

 7 Representations of Semisimple Lie Algebras

(v) Follows immediately from (iv) and 7.25 by applying Ω to a singular vector with weight λ.

(vi) If Vλ contains a singular vector, then |λ + ρ|2 = |Λ + ρ|2 . This equation defines a sphere
in RR (center ρ, radius |Λ + ρ|) — a compact set. On the other hand, D(Λ) is discrete,
and the intersection of a compact and a discrete set is finite.

Definition 7.37: Let Λ ∈ t∗ . A Verma module with highest weight Λ and highest weight
vector vΛ , M (Λ), is a universal module with highest weight Λ, i.e. if V is any highest weight
module with highest weight vector v (also of weight Λ), there exists a unique map

M (Λ) → V, vΛ 7→ v.

Proposition 7.38: Let Λ ∈ t∗ , then:

(i) There exists a unique Verma module M (Λ).

(ii) There exists a unique irreducible highest weight module of weight Λ, we denote it by L(Λ).

Proof:

(i) Uniqueness of M (Λ) is clear by the universal property. For existence, define

M (Λ) = U g ⊗U b CΛ ,

where b = n+ + t and CΛ is the b-module on which n+ · v = 0 and t · v = Λ(t)v, i.e.

M (Λ) = U g/J(Λ),

where J(Λ) is the left ideal generated by u − Λ(u) for all u ∈ U b. Here we extend Λ to
the character U b → C. In other words, M (Λ) is the module generated by g acting on 1,
with relations n+ · 1 = 0, t · 1 = Λ(t)1, for all t ∈ t and only the relations these imply. So if
V is an arbitrary highest weight module with weight Λ, it is clear that V = U g/J, where
J is some ideal containing J(Λ), i.e. M (Λ)  V .

(ii) From the proof of (i) follows in particular, that an irreducible highest-weight module must
be of the form M (Λ)/I(Λ) where I(Λ) is a maximal proper submodule of M (Λ). But we
have just shown that there is an unique maximal proper submodule, so L(Λ) is unique.

Proposition 7.39: Let R+ = {β1 , . . . , βr }, then ek−β

1
1
· · · ek−β
r
v is a basis of M (Λ).
r Λ

Proof: The “hard” part of the PBW Theorem 7.30 implies this immediately.

Corollary 7.40: Any irreducible finite-dimensional g-module is of the form L(Λ) for some
Λ ∈ P +.

Proof: We know that it is of the form L(Λ), some Λ, and we have seen that the highest weight
must be in P + by results of the sl2 theory.

Example 7.41: Let g = sl2 . The Verma module M (Λ) is an infinite string of the following
shape:

Robert Laugwitz & Henning Seidler 55

Lie Algebras and Their Representations

vλ

F vλ

F 2 vλ

F 3 vλ
..
.

Exercise 7.42 (Essential): Let first g = sl2 .

(i) Show that M (λ) = L(λ) (i.e. M (λ) is irreducible) if and only if λ ∈
/ Z≥0 .

(ii) Show that if λ ∈ Z≥0 , then M (λ) contains a unique proper submodule, the span of F λ+1 vλ ,
F λ+2 vλ , . . .. This submodule is itself a Verma module.
Λ(Hi )+1
(iii) Let now g be an arbitrary simple Lie algebra, and Λ(Hi ) ∈ Z≥0 . Show that Fi · vΛ
is a singular vector of M (Λ) (NB: there will also be other singular vectors).

(iv) Very important: compute ch M (Λ).

Proposition 7.43: Let Λ ∈ P + , then L(Λ) is integrable, i.e. Ei and Fi act locally nilpotently
(that is, all v ∈ L(Λ) are contained in a finite-dimensional subspace on which Ei acts nilpotently,
i.e. Ein v = 0 for some n > 0, and ditto for Fi ).

Proof: If V is any highest-weight module, then Ei acts locally nilpotently (as Ei Vλ ⊆ Vλ+αi ,
P
but weights of V are in the cone D(Λ) = {Λ − ki αi | ki ≥ 0}). We must show that Fi
Λ(H )+1
acts locally nilpotently. We know that Fi i · vΛ is a singular vector, by Exercise 7.42 (iii).
Λ(H )+1
But L(Λ) is irreducible, so it has no singular vectors other than vΛ , so Fi i vΛ = 0 by the
following exercise, which finishes the proof.

Exercise 7.44:

(i) ak b = ki=0 k
((ad a)i b)ak−i
P

i

(ii) Using (i) and the Serre relations (ad eα )4 eβ = 0 (for all α, β ∈ R), show FiN e−β1 . . . e−βr vΛ =
0 for N  0 by induction on r.

Note that we need the power 4 in the Serre relations in the worst case, for G2 where we have
a string α, α + β, α + 2β, α + 3β. Notice that this is true for a generalized Kac-Moody algebra
as well, but then the 4 is replaced by the maximal −aij + 1 in the Cartan matrix.

Corollary 7.45: We have dim L(Λ)µ = dim L(Λ)wµ for all w ∈ W .

Proof: We have seen that Ei , Fi act locally nilpotently implies this statement is true for w = sαi
— a simple reflection (see proof sketch 2 of 7.8). But W is generated by sα1 , . . . sαl , so this even
holds for all w ∈ W .

56 Robert Laugwitz & Henning Seidler

 7 Representations of Semisimple Lie Algebras

Theorem 7.46 (Cartan’s theorem): If g is finite-dimensional and Λ ∈ P + , then L(Λ) is

finite-dimensional.

Proof: Let α ∈ R+ . We know that eα acts nilpotently on L(Λ). We show first, that also e−α
does, to see that all of the root sl2 -copies act integrably. In fact, en−α vΛ = 0 for n = 2(Λ,α)
(α,α) =
∨
(Λ, α ) + 1, as if not, we would have L(Λ)Λ−nα 6= 0 and hence by Corollary 7.45 that

sα (Λ − nα) = sα (Λ) + nα
= Λ − (Λ, α∨ )α + ((Λ, α∨ ) + 1)α
=Λ+α>Λ

is also a weight in L(Λ), contradicting that Λ is the highest weight. Thus, by Exercise 7.44, we
see that e−α acts locally nilpotently on all of L(Λ). Therefore,

U (n− )vΛ = hek−β

1
1
. . . ek−β
r
v i = L(Λ),
r Λ
for R+ = {β1 , . . . βr },

is finite-dimensional.

Now we can prove the complete reducibility stated in Theorem 7.14. In order to do that, we
need the following lemmas:

Lemma 7.47: For the reflection si = sαi of the i-th simple root αi the condition si (R+ \{αi }) =
R+ \ {αi } holds.

kj αj ∈ R+ , then all kj ≥ 0. Now

P
Proof: Let α = j
 
X X
si α = kj αj −  (αj , αi∨ )kj + ki  αi ,
j6=i j6=i

but α 6= αi , so some kj > 0, j 6= i. Thus, the coefficient of αj in si α is still positive (as it is the
same coefficient kj ). But R = R+ (−R+ ), i.e. the disjoint union of roots with all coefficients
`

≥ 0 and roots where all coefficients are ≤ 0. This implies si α ∈ R+ , and si α 6= αi .

1 P
Recall ρ = 2 α∈R+ α.

Lemma 7.48: We have ρ(Hi ) = 1, for all i, i.e. ρ = ω1 + . . . + ωl ∈ P + .

Proof: Observe
 
1 1 X 
si ρ = si 
2 α i + α
2 
α6=αi
α∈R+
1 1 X
= − αi + α = ρ − αi .
2 2
α6=αi
α∈R+

But, si ρ = ρ − (ρ, αi∨ )αi , so (ρ, αi∨ ) = 1 for all i.

Robert Laugwitz & Henning Seidler 57

Lie Algebras and Their Representations

Lemma 7.49 (Key lemma): Let Λ ∈ P + , and µ ≤ Λ such that µ + ρ ∈ P + . Then

|Λ + ρ| = |µ + ρ| =⇒ Λ = µ.
P
Proof: Denote Λ − µ = ki αi , then for all i, ki ≥ 0, and we compute
0 = (Λ + ρ, Λ + ρ) − (µ + ρ, µ + ρ) = (Λ + ρ − (µ + ρ), Λ + ρ + µ + ρ)
X
= (Λ − µ, Λ + ρ + µ + ρ) = ki (αi , Λ + ρ + µ + ρ),
| {z }
∈P +

as Λ, ρ, µ + ρ ∈ P + . But (αi , (Λ + µ) + 2ρ) ≥ 1. This implies ki = 0 for all i.

Theorem 7.50 (Weyl complete reducibility, cf. 7.14): Let char k = 0, k = k and g be
a semisimple Lie algebra over k, then every finite-dimensional g-module V is a direct sum of
irreducibles.

Proof: Recall that V is completely reducible as an (sl2 )α -module. Write V = ⊕λ∈P Vλ . Consider
+ +
V n = {v ∈ V | n+ v = 0}. By Engel’s theorem, V n 6= 0, and [t, n+ ] ⊆ n+ . Hence t acts on
+ + + +
V n , and so V n = µ∈P Vµn , where Vµn = {x ∈ V | n+ x = 0 and tx = µ(t)x}. Therefore,
L
+
V n consists of singular vectors.
+
We claim that for every 0 6= vµ ∈ Vµn , the module L = U g · vµ is irreducible. To prove
this, note that L is a highest weight module with highest weight µ, so we must only show that
it has no other singular vectors. If λ is the weight of a singular vector in L, then λ ≤ µ, but
also |λ + ρ| = |µ + ρ| by considering the action of the Casimir. Since V , and therefore L, is
finite-dimensional, we must have λ, µ ∈ P + (by 7.14, and λ(hi ) = (λ, αi∨ )). So by the key
lemma, λ = µ.
+
It follows that V 0 = U g · V n is completely reducible (as if {v1 , . . . , vr } is a basis of weight
+
vectors for V n with weights λ1 , . . . , λr , then V 0 = L(λ1 ) ⊕ . . . ⊕ L(λr )). So to finish, we must
show that N = V /V 0 = 0.
+ +
If N 6= 0, then N n 6= 0. Let vλ ∈ Nλn be a singular vector, as N is finite-dimensional,
λ ∈ P + . Lift vλ to vλ ∈ Vλ , then Ei vλ ∈ Vλ+αi and there exist some i s.t. Ei vλ 6= 0 as otherwise
vλ is a singular vector and U gvλ is contained in V 0 , contradicting our choice of vλ 6= 0. But
then, as Ei v λ ∈ V 0 ,
7.25 2
ΩEi vλ = Ei Ωvλ = ( λ0 + ρ − |ρ|2 )Ei vλ , λ0 ∈ {λ1 , . . . , λr },
but on the other hand, as vλ is a singular vector in N , Ωvλ = |λ + ρ|2 −|ρ|2 vλ , so |λ0 + ρ| = |λ + ρ|
(by 7.36). Moreover, λ + αi is a weight in L(λ0 ), so λ + αi = λ0 −
P
kj αj , for some kj , i.e.
λ = λ0 + ki0 αi with not all ki0 zero contradicting the key lemma, so V = V 0 .
P

7.3 The Weyl character formula

Lemma 7.51: Let Λ ∈ t∗ and M (Λ) the Verma module of Λ. Then
eΛ
ch V (Λ) = Q −α )
. (10)
α∈R+ (1 − e

Proof: Let R+ = {β1 , . . . , βr }. The PBW Theorem 7.30 gives the basis {ek−β
1
1
. . . ek−β
r
v | ki ∈
r Λ
P r
Z≥0 } for the Verma module M (Λ) and the weight of such an element is Λ − i=1 ki βi . So the
P
dimension of a weight space M (Λ)Λ−β is the number of ways of writing β as ki βi . But this
−β α −1
Q
is the coefficient of e in α∈R+ (1 − e ) .

58 Robert Laugwitz & Henning Seidler

 7 Representations of Semisimple Lie Algebras

− e−α ). We have just shown that ch M (Λ) = eΛ /∆.

Q
Let us write ∆ = α∈R+ (1

Lemma 7.52: For all w ∈ W , w(eρ ∆) = det w · eρ ∆. Here, det : W → Z/2 is the determinant
of w acting on t∗ (= ±1).

Proof: Since W is generated by simple reflections, it is enough to show that si (eρ ∆) = −eρ ∆.
But
 
Y
−αi
si (eρ ∆) = si  (1 − e−α )
 ρ 
 e (1 − e ) 
α6=αi
α∈R+
Y
= eρ−αi (1 − e+αi ) (1 − e−α ) = −eρ ∆,
α6=αi
α∈R+

as si ρ = ρ − αi , and si (R+ \ {αi }) = R+ \ {αi }.

Lemma 7.53: For any highest weight module V (Λ) with highest weight Λ

(i) there exist coefficients aλ ≥ 0, λ ≤ Λ, such that

X
ch V (Λ) = aλ ch L(λ), with aΛ = 1; (11)
λ≤Λ
|λ+ρ|=|Λ+ρ|

(ii) there exist coefficients bλ ∈ Z with bΛ = 1 such that

X
ch L(Λ) = bλ ch M (λ). (12)
λ≤Λ
|λ+ρ|=|Λ+ρ|

Proof: (i)⇒(ii): We write B(Λ) = {λ ≤ Λ | |λ + ρ| = |Λ + ρ|}. Recall that B(Λ) is a finite set
(for Λ ∈ RR). We have a total order on B(Λ) = {λ1 , . . . , λn } so that if λi ≤ λj , then i ≤ j.
Then (i) is a system of equations relating ch M (λ) and ch L(λ), which is upper-triangular with
ones on the diagonal and therefore invertible. Inverting this system gives (ii).
(i): Recall that the weight spaces of a highest weight module are finite-dimensional. We induct
P
on µ∈B(Λ) dim V (Λ)µ . Note that if V (Λ) is irreducible, (i) is true with aΛ = 1, aλ = 0, if λ 6= Λ.
Otherwise, there exists a root µ ∈ B(Λ) with a singular vector vµ ∈ V (Λ)µ . Choose µ so that the
P P
height ( ki ) of Λ−µ = ki αi is maximal for all singular vectors. Then L(µ) := U g·vµ ⊆ V (Λ)
has no singular vectors, and is therefore irreducible. Set V (Λ) = V (Λ)/L(µ) (i.e there exists
an exact sequence 0 → L(µ) → V (Λ) → V (Λ) → 0), then we see that V (Λ) is a highest-weight
P
module with a smaller value of µ∈B(Λ) dim V (Λ)µ , and ch V (Λ) = ch V (Λ) + ch L(µ). So we
are done by induction.

We will now compute ch L(Λ) for Λ ∈ P + . We know that

X X eλ
ch L(Λ) = bλ ch M (λ) = bλ .
∆
λ∈B(Λ) λ∈B(Λ)

Robert Laugwitz & Henning Seidler 59

Lie Algebras and Their Representations

Further, we have seen before that w(ch L(Λ)) = ch L(Λ) for all w ∈ W , and w(∆eρ ) = det w·∆eρ .
Therefore, X
eρ ∆ ch L(Λ) = bλ eλ+ρ
λ∈B(Λ)
P  P 
is W -anti-invariant. So w b e λ+ρ = det w. b e λ+ρ . Let us rewrite this as
λ∈B(Λ) λ λ∈B(Λ) λ
X X X
bλ eλ+ρ = bλ det w · ew(λ+ρ) ,
λ∈B(Λ) λ1 ,...,λs w∈W

where λ1 , . . . , λs is a representative system for the orbits of W acting on B(Λ + ρ). Now, if
λ ∈ RR (which is true, since Λ ∈ P + ), then W (λ + ρ) intersects {x ∈ RR | (x, αi∨ , ∀i) ≥ 0} in
exactly one point (this set is a fundamental domain for the W -action on RR). (Note that for a
given λ ∈ RR, W (λ + ρ) defines a positive root system and W acts simply transitively on those
roots). Therefore, we can take a representative system for the orbits only containing dominant
weights. Note that one of these dominant weights is Λ and the other orbits are given by W
acting on {λ ∈ B(Λ) | λ 6= Λ, λ ∈ P + }. But the key lemma 7.49 implies that this set is empty,
so the only coefficient is bΛ = 1. This proves the following theorem:

Theorem 7.54 (Weyl Character Formula): For all Λ ∈ P +

det w · ew(Λ+ρ)
P
ch L(Λ) = ρw∈W
Q −α )
(13)
e α∈R+ (1 − e
X
= det w · ch M (w(Λ + ρ) − ρ). (14)
w∈W

Example 7.55: Let g = sl2 , and write z = eα/2 , then C[P ] = C[z, z −1 ] and eρ = z, and we
have
α z m+1 − z −(m+1)
ch L(m ) =
2 z − z −1
as we saw earlier in this course.

Corollary 7.56 (Weyl denominator identity): As L(0) = C, we have ch L(0) = 1, so

Y X
eρ (1 − e−α ) = det w · ewρ .
α∈R+ w∈W

Exercise 7.57: Let g = sln . Show that the Weyl denominator identity is equivalent to the
Vandermonde determinant
 
1 1 ... 1
 z1 z2 ... zn  Y
det  . = (zi − zj ),
 
.. .. ..
 .. . . . 
i<j
z1n−1 z2n−1 . . . znn−1
where we write zi = eei .

Corollary 7.58 (Weyl dimension formula): For all Λ ∈ P + we have

Y (α, Λ + ρ)
dim L(Λ) = . (15)
+
(α, ρ)
α∈R

60 Robert Laugwitz & Henning Seidler

 7 Representations of Semisimple Lie Algebras

Example 7.59: g = sl3 (root system is of type A2 ), R+ = {α, β, α+β} with ρ = α+β = ω1 +ω2 .
Let Λ = m1 ω1 + m2 ω2 , then

α β α+β
(·, Λ + ρ) m1 + 1 m2 + 1 m1 + m2 + 2 .
(·, ρ) 1 1 2

Therefore, dim L(Λ) = 21 (m1 + 1)(m2 + 1)(m1 + m2 + 2).

Exercise 7.60: Compute the dimensions of all the finite-dimensional irreducible representa-
tions of B2 and G2 .

Remark 7.61: Let w ∈ W be written as w = si1 si2 . . . sir where sik are simple reflections.
Then det w = (−1)r . The minimal r such that w can be written in this form is called the length
of w, denoted l(w). The Monoid Lemma asserts that you can get from one minimal-length
expression for w to another by repeatedly applying the braid relations.

Exercise 7.62: l(w) = #{−R+ ∩ w−1 R+ } = l(w−1 ).

Proof: (Weyl dimension formula). We still have to prove the Weyl dimension formula 7.58.
We know ch L(Λ) = dim L(Λ)λ eλ ∈ C[P ]. We would like to set eλ 7→ 1, but then the denom-
P

inator in the Weyl character formula would become 0. Instead, consider the homomorphism

Fµ : C[P ] → C(q), eλ 7→ q −(λ,µ) .

For example, F0 (eλ ) = 1, so F0 (ch L(λ)) = dim L(λ). Now apply Fµ to the Weyl dominator
identity. Then
Y X X
q −(ρ,µ) (1 − q (α,µ) ) = det wq −(wρ,µ) = det wq −(ρ,wµ)
α∈R+ w∈W w∈W

as det w = det w−1 and (x, wy) = (w−1 x, y) (i.e. the Weyl group is a subgroup of the orthogonal
group of the inner product). We now apply Fµ to the Weyl character formula:
−(w(Λ+ρ),µ)
P
w∈W det wq
Fµ (ch L(Λ)) = −(ρ,µ) Q (α,µ) )
q α∈R+ (1 − q

if (α, µ) 6= 0 for all α ∈ R+ .

Now, take µ = ρ (recall that (ρ, αi ) = 1 > 0 for all simple roots αi , so (ρ, α) > 0 for all
α ∈ R+ ), so

q −(ρ,Λ+ρ) α∈R+ (1 − q (α,Λ+ρ) )

Q
X
−(λ,ρ)
Fρ (ch L(Λ)) = dim L(λ)λ q =
q −(ρ,ρ) α∈R+ (1 − q (α,ρ) )
Q

where we used our expression for the Weyl denominator identity and applied it to the numerator.
From this we can conclude the Weyl dimension formula
Y (Λ + ρ, α)
dim L(Λ) =
+
(ρ, α)
α∈R

by setting q = 1 and applying L’Hôpital’s rule.

Robert Laugwitz & Henning Seidler 61

Lie Algebras and Their Representations

Remark 7.63: We can now algorithmically answer all questions about finite dimensional rep-
resentations of semisimple Lie algebras by knowing the highest weight. For example, let us
P ν
decompose L(λ) ⊗ L(µ) = mλµ L(ν) (by complete reducibility). To compute the Littlewood-
Richardson coefficients mνλµ (recall that we had the Clebsch-Gordan rule for them in sl2 ) define

: Z[P ] → Z[P ], eλ 7→ e−λ ,

(
0 if λ 6= 0
CT : Z[P ] → Z, eλ 7→ , and
1 if λ = 0
1
(·, ·) : Z[P ] × Z[P ] → Z, (f, g) = CT (f g∆∆),
|W |

where ∆ = α∈R+ (1 − e−α ).

Q

Claim: if we let χλ = ch L(λ), then (χλ , χµ ) = δµν , and thus mνλµ = (χλ χµ , χν ).

Proof:
1 X
(χλ , χµ ) = CT ( ew(λ+ρ)−ρ ex(µ+ρ)−ρ det(wx))
|W |
x,w∈W

by Weyl. But CT (ew(λ+ρ)−x(µ+ρ) ) = δwx δµλ as for λ, µ ∈ R+ we have w(λ + ρ) = µ + ρ if and

only if w = 1, µ = λ (as the dominant weights are the lattice points in a fundamental domain of
the W -action), so x−1 w(λ + ρ) = µ + ρ precisely if x = w, λ = µ.

7.4 Principal sl2

Define ρ∨ ∈ t∗ by (ρ∨ , αi ) = 1 for all αi ∈ Π (recall that (ρ, αi∨ ) = 1, so ρ∨ can be seen as ρ for
R∨ ).

Exercise 7.64: Show ρ∨ = 21 α∈R+ α∨ . In particular, if R is simply laced, ρ = ρ∨ . This

P

implies (ρ∨ , α) = ht(α) = ki if we write α = ki αi .

P P

Exercise 7.65:
Y (1 − q (Λ+ρ,α∨ ) )
−(Λ,ρ∨ )
Fρ∨ (ch L(Λ)) = q .
+
(1 − q (ρ,α∨ ) )
α∈R

Hint: Apply Fρ∨ to the Weyl denominator identity of the irreducible representations of the Lie
∨)
algebra with root system R∨ . Note that (λ+ρ,α (λ+ρ,α) ∨ 2α
(ρ,α∨ ) = (ρ,α) as α = (α,α) , so we recover

Definition 7.66: We call Fρ∨ (ch L(Λ)) =: dimq L(Λ) the q-dimension of L(λ).

Proposition 7.67: The q-dimension dimq L(Λ) is a unimodal polynomial, i.e. it lives in
N[q 2 , q −2 ]Z/2 or qN[q 2 , q −2 ]Z/2 (depending on its degree), and the coefficients decrease as the
absolute value of the degree increases.

Proof: This follows if we show that dimq L(Λ) is the character of an sl2 -module in which the
length of all “strings” have the same parity. Let H = 2ν −1 (ρ∨ ) ∈ t ⊆ g, and set E =
P
Ei .
P
Check that [H, E] = 2E (Exercise). Write H = ci Hi for some ci ∈ C (H1 , . . . , Hl is a basis of
P
t), and set F = ci Fi . It is left as an exercise to show:

(i) Show that E, F, H generate a subalgebra isomorphic to sl2 , the so-called principal sl2 .

62 Robert Laugwitz & Henning Seidler

 8 Crystals

(ii) Show that if Λ − γ is a weight of L(Λ), then (Λ − γ, 2ρ∨ ) ≡ (Λ, 2ρ∨ ) (mod 2).

This implies the proposition.

q n −1
Exercise 7.68: Write [n] = q−1 . Show that the following polynomials are unimodal:
n [n]!
(i) k = [k]![n−k]! , where [n]! = [n][n − 1] . . . [1],

(ii) (1 + q)(1 + q 2 ) . . . (1 + q n ).

Hint: For (i), apply the above arguments to g = sln and V = S k Cn or Λk Cn+k . For (ii), apply
this to the spin representation of Bn = so2n+1 (which we will define in 8.16).

Remark 7.69: An isomorphism V ∼ = V ∗ implies a bilinear form (·, ·), but is this form in Λ2 V
2
or S V ? Consider for example sl2 , it can be shown that the bilinear form induced by L(n) is
alternating precisely if n is odd, and symmetry if n is even. Notice that L(λ) ∼ = L(λ)∗ if and
2
only if the lowest weight of L(λ) is −λ (for example, the C representations are always self-dual,
and C2 ⊗ C2 ∼ = C3 + C). Now, the question whether the bilinear form induced by L(λ) ⊗ L(λ)∗
is alternating or symmetric can be answered by checking this for the restriction to the principal
sl2 . This is equivalent to (λ, 2ρ∨ ) having odd or even parity.

Exercise 7.70:

(i) Compute dimq L(θ), where L(θ) is the adjoint representation, for A2 , B2 , and G2 . Then
do this for all the classical groups.

(ii) You will notice that L(θ)|principal sl2 = L(2e1 ) + . . . + L(2el ) where l = rank g = dim t, for
some e1 , . . . , el ∈ N with e1 = 1. The ei are called the exponents of the Weyl group. Note
that the order of the Weyl group is |W | = (e1 + 1) . . . (el + 1). If you are in the mood,
compute |W | for E8 .

8 Crystals
Let g be a semisimple Lie algebra, Π = {α1 , . . . , αl } the simple roots, and P the weight lattice.

Definition 8.1: A crystal is a set B, 0 6∈ B, together with functions wt : B → P , ẽi : B →

B t {0}, f˜i : B → B t {0} such that

(i) If ẽi b 6= 0, then wt ẽi (b) = wt b + αi , and if f˜i b 6= 0, then wt f˜i (b) = wt b − αi .

(ii) For b and b0 ∈ B, ẽi b = b0 if and only if b = f˜i b0 .

(iii) ϕi (b) − εi (b) = hwt b, αi∨ i, for all αi ∈ Π, where

εi (b) = max{n ≥ 0 | ẽni b 6= 0},

ϕi (b) = max{n ≥ 0 | f˜n b 6= 0}.
i

We can draw B as a graph: The vertices are b ∈ B, and the edges are b −→ b0 if ẽi b0 = b.
i
We say that this edge is coloured by i. We call such a graph a crystal graph.

Robert Laugwitz & Henning Seidler 63

Lie Algebras and Their Representations

Example 8.2: Consider sl2 , then the string

n −→ n − 2 −→ n − 4 −→ . . . −→ −n

is a crystal, where the weight of vertex i is i α2 . Notice, for the crystal of the highest-weight
representation L(n) = L(nw1 ), we have that if b is of weight n − 2k, then ε(b) = k, and
ϕ(b) = n − k and the sum εi (b) + ϕi (b) is the length of the string
ε ϕ
z }| { z }| {
n −→ n − 2 −→ n − 4 −→ . . . −→ n − 2k −→ . . . −→ −n .

Define Bµ = {b ∈ B | wt b = µ}.
If B1 and B2 are crystals, can define the tensor product B1 ⊗ B2 = B1 × B2 as a set, with
wt(b1 ⊗ b2 ) = wt b1 + wt b2 , and
(
(ẽi b1 ) ⊗ b2 , if ϕi (b1 ) ≥ εi (b2 )
ẽi (b1 ⊗ b2 ) = whence
b1 ⊗ (ẽi )b2 , if ϕi (b1 ) < εi (b2 ),
(
(f˜i b1 ) ⊗ b2 , if ϕi (b1 ) > εi (b2 )
f˜i (b1 ⊗ b2 ) =
b1 ⊗ (f˜i )b2 , if ϕi (b1 ) ≤ εi (b2 ).

That is, in each colour i we have a graph of the form

i i i i i

the same form as we have seen for sl2 before.

Exercise 8.3:

(i) Check that B1 ⊗ B2 defines a crystal.

(ii) B1 ⊗ (B2 ⊗ B3 ) ∼= (B1 ⊗ B2 ) ⊗ B3 , b1 ⊗ (b2 ⊗ b3 ) 7→ (b1 ⊗ b2 ) ⊗ b3 . It suffices to prove this

for sl2 . Note that it is not true in general that B1 ⊗ B2  B2 ⊗ B1 .

Definition 8.4: B ∨ is the crystal obtained from B by reversing the arrows. That is, B ∨ =
{b∨ | b ∈ B}, wt b∨ = − wt b, εi (b∨ ) = ϕi (b) (and vice versa), and ẽi (b∨ ) = (f˜i b)∨ (and vice
versa). In pictures:
 ∨  
• −→ • −→ • = • −→ • −→ • .
1 2 2 1

64 Robert Laugwitz & Henning Seidler

 8 Crystals

Remark 8.5: If B corresponds to a basis of a representation V , then B ∨ corresponds to a basis

of the dual V ∗ as B → B ∨ comes from the Lie algebra anti-automorphism ei 7→ fi , fi 7→ ei , and
hi 7→ −hi . Notice that if L(λ) is a representation with highest weight λ, then L(λ)∗ has lowest
weight −λ.

Exercise 8.6: Show that (B1 ⊗ B2 )∨ = B2∨ ⊗ B1∨ .

Theorem 8.7 (Kashiwara): Let L(λ) be the irreducible highest-weight representation with
highest weight λ ∈ P + , then:

(i) There exists a crystal B(λ) whose elements are in 1-1 correspondence with a basis of L(λ)
(i.e. B(λ)µ parametrizes a basis of L(λ)µ ), so
X
ch L(λ) = ewt(b) . (16)
b∈B(λ)

(ii) For each simple root αi (i.e. a simple (sl2 )i ⊆ g), the decomposition of L(λ) as an (sl2 )i -
module is precisely given by the i-coloured strings in B(λ). (In particular, as an uncoloured
graph, B(λ) is connected, since it is spanned by elements of the form f˜1 . . . f˜l · vλ .)

(iii) The crystal B(λ)⊗B(µ) is precisely the crystal for L(λ)⊗L(µ), i.e. B(λ)⊗B(µ) decomposes
into connected components exactly in the way L(λ) ⊗ L(µ) decomposes into irreducible
representations.

Example 8.8: Let g = sl3 , V = C3 = L(ω1 ), then the weight spaces are 1-dimensional, so we
have no choice but to define the crystal as

w1 → w1 − α1 → w1 − α1 − α2 .
1 2

Let us compute V ⊗ V and V ⊗ V ∗ :

α 1 + α 2 − w1
h.w. 2w1
α1 + α2
w1
2w1 − α1

2w1 − α1

(a) V ⊗ V (b) V ⊗ V ∗

Figure 4: Crystals for V ⊗ V and V ⊗ V , V = C3 as sl3 -module

Here, we chose black as colour 1, and red as colour 2 in the graphic. This implies,

V ⊗ V = C3 ⊗ C3 = S 2 C3 + Λ2 C = S 2 C3 + (C3 )∗ as 2w1 − α1 = α1 + α2 − w1 , and

V ⊗ V ∗ = End V = C + sl3 (as α1 + α2 = θ).

Robert Laugwitz & Henning Seidler 65

Lie Algebras and Their Representations

Remark 8.9: There are three proof approaches to the Kashiwara’s theorem. The first one is
due to Kashiwara, and is in the lecturer’s opinion the most instructive.
Note that while the crystals give the decomposition of the representation into irreducibles,
they do not correspond directly to a basis. That is, there is no sl2 -invariant basis that we could
use here. Kashiwara’s proof of the theorem uses the quantum group Uq sl2 , which is an algebra
over C[q, q −1 ] and a deformation of the universal enveloping algebra U sl2 . The two algebras
U sl2 and Uq sl2 have the same representations, but over C[q, q −1 ] there is a very nice basis which
satisfies ei b = ẽi b + q · (“some mess”). Therefore, setting q = 0 (“freezing”) will give the crystal.
A second proof approach is due to Lusztig. We will later look at the third proof using
Littlemann paths, which give a purely combinatorial way of proving this theorem (which, on the
face of it, is a purely combinatorial statement).

Definition 8.10: A crystal is called integrable if it is a crystal of an integrable highest-weight

module with highest weight in P + .

For two integrable crystals B1 , B2 , we do in fact have B1 ⊗ B2 = B2 ⊗ B1 (in general, this

is false).
There is a combinatorial condition on crystals which implies that a crystal is integrable (due
to Stembridge); it is a degeneration of the Serre relations.

8.1 Semi-standard Young tableaux

Consider sln :
• −→ • −→ • −→ . . . −→ • −→
e1 =w1 1 w1 −α1 2 w1 −α1 −α2 3 n−1 n−1

is the crystal of the standard representation L(w1 ) = Cn .

From this, we can construct the
crystals for all sln -representations:
Let λ ∈ P + , λ = k1 w1 + ... + kn−1 wn−1 , then L(λ) is a summand of L(w1 )⊗k1 ⊗ . . . ⊗
L(wn−1 )⊗kn−1 as vw ⊗k1 ⊗ . . . ⊗ v ⊗kn−1 is the highest weight vector of weight λ (if v
1 wn−1 wi is the
highest weight vector of L(wi )). But L(wi ) = Λi Cn is a summand of (Cn )⊗i , so L(wi ) occurs
in some (Cn )⊗N , N > 0. Therefore, the crystal of Cn and the rule for the tensor product of a
crystal determine the crystal for every representation L(λ) of sln .
Now, we can introduce the semi-standard Young tableau of a representation (due to Hodge
(∼1930), Schur (∼1900), and Young (∼1900)). Write

B(w1 ) = 1 −→ 2 −→ 3 −→ . . . −→ n
1 2 3 n−1

for the crystal of the standard representation Cn . Now, if i < n, denote

bi = 1 ⊗ 2 ⊗ . . . ⊗ i ∈ B(w1 )⊗i .

The element bi corresponds to the basis vector v1 ∧ v2 ∧ . . . ∧ vi ∈ Λi Cn , where Cn has the basis
v1 , . . . , v n .

66 Robert Laugwitz & Henning Seidler

 8 Crystals

Exercise 8.11:
(i) The vector bi is a highest weight vector in B(w1 )⊗i of weight wi = e1 + . . . + en . (Recall
that b ∈ B is a highest weight vector if ẽi b = 0 for all i). Hence, the connected component
of B(w1 )⊗i containing bi is B(wi ).
(ii) The connected component B(wi ) consists precisely of
{ a1 ⊗ a2 ⊗ . . . ⊗ ai | 1 ≤ a1 < . . . < ai ≤ n} ⊂ B(w1 )⊗i .

We can write elements of the form a1 ⊗ a2 ⊗ . . . ⊗ ai as column vectors

a1 1
a2 2
.. , so the highest weight vectors are denoted .. .
. .
ai i
ki wi and embed B(λ) ,→ B(w1 )⊗k1 ⊗ . . . ⊗ B(wn−1 )⊗kn−1 by mapping the
P
Now, let λ =
⊗k ⊗k
highest weight vector bλ 7→ b1⊗k1 ⊗ . . . ⊗ bn−1n−1 . Note that b⊗k
1
1
⊗ . . . ⊗ bn−1n−1 actually is a highest
weight vector in B(w1 )⊗k1 ⊗ . . . ⊗ B(wn−1 )⊗kn−1 . Now as B(wi ) ,→ B(w1 )⊗k1 and hence
n−1
X
B(w1 )⊗k1 ⊗ . . . ⊗ B(wn−1 )⊗kn−1 ,→ B(w1 )N , N := ki ,
i=1

we can represent any element in B(w1 )⊗k1 ⊗ . . . ⊗ B(wn−1 )⊗kn−1 by a sequence of column vectors

···
··· k1
n−1

kn−2
kn−1
Pn
where the entries are strictly increasing down columns, the length of the i-th row is j=i kj .
We say this young tableau has shape λ.

Definition 8.12: A semi-standard Young tableaux is an array of numbers as above, such that
(i) the numbers are strictly increasing down columns, and
(ii) decreasing along rows.

Theorem 8.13 (Exercise):

(i) The semi-standard Young tableau of shape λ are precisely elements of the connected
component of B(λ) in B(w1 )⊗k1 ⊗ . . . ⊗ B(wn−1 )⊗kn−1 .
(ii) Describe ẽi , f˜i explicitly in terms of tableaux.

In the following, we will construct the Young tableau for the classical Lie algebras.

Robert Laugwitz & Henning Seidler 67

Lie Algebras and Their Representations

Example 8.14: so2n+1 : (Type Bn root systems) For the standard representation C2n+1 we
have the crystal

1 −→ 2 −→ 3 −→ . . . −→ n −→ 0 −→ n −→ . . . −→ 2 −→ 1 .
1 2 4 n−1 n n n−1 2 1

so2n : (Type Dn root systems) For the standard representation C2n we have the crystal

>
n−1 n

>
1 2 n−2 n−2 2 1
1 > 2 > ... > n−1 n−1 > ... > 2 > 1.

>
n n−1

>
n

sp2n : (Type Cn root systems) For the standard representation C2n we have the crystal

1 −→ 2 −→ 3 −→ . . . −→ n −→ n −→ . . . −→ 2 −→ 1 .
1 2 4 n−1 n n−1 2 1

Exercise 8.15:

(i) Show that these are indeed the crystals of the standard representations of the classical Lie
algebras.

(ii) What subcategory of the category of representations of g do these representations generate?

Consider the highest weight λ of the standard representation. This gives an element
λ ∈ P/Q = Z(G), a finite group (G is the simply connected group attached to g). Consider
the subgroup hλi ≤ P/Q. We do not obtain all the representations unless P/Q is cyclic,
generated by λ. For the classical examples we have P/Q = Z/2×Z/2 for D2n , P/Q = Z/4
for D2n+1 , P/Q = Z/2 for Bn and Cn .

(iii) (Optional) Write down a combinatorial set like Young tableaux that is the crystal of B(λ)
with λ obtained from the standard representation.

For Bn , we need one more representation, the spin representation. Recall that for Bn we
had the dynkin diagram

α1 α2 αn−2 αn−1 αn

Definition 8.16: The irreducible highest weight so2n+1 -representation L(wn ), where wn is the
n-th fundamental weight, is called the spin representation.

Exercise 8.17: Use the Weyl dimension formula 7.58 to show that dim L(wn ) = 2n .

68 Robert Laugwitz & Henning Seidler

 8 Crystals

1 Pn
Define B = {(i1 , . . . , in ) | ij ∈ {±1}}, wt(i1 , . . . , in ) = 2 j=1 ij ej ∈ P , and for 1 ≤ j ≤ n−1

(i1 , . . . , +1, − 1, . . . , in ) if (ij , ij+1 ) = (−1, +1)
ẽj (i1 , . . . , in ) = j j+1
0 otherwise,
(
(i1 , . . . , in−1 , +1) if in = −1
ẽn (i1 , . . . , in ) =
0 otherwise,

so always ẽ2i = 0.

Fact 8.18: This is the crystal of the spin representation L(wn ).

 
A
Remark 8.19: We have dim L(wn ) = dim Λ• Cn . In fact, gln ⊂ so2n+1 , A 7→  0 ,
−JAT J −1
and L(wn )|gln = Λ• Cn .

Exercise 8.20: Check that B|gln is a crystal of Λ• Cn .

For type Dn , the situation is more complex. We can define representations

V + = L(wn ), V − = L(wn−1 ).

These are called half-spin representations. B ± = {(i1 , . . . , in ) | ij ∈ {±1}, πij = ±1} + if B +

and − if B − . wt, ei , and fi (i < n) are defined as above, and
(
(i1 , . . . , in−2 , +1, +1) if (in−1 , in ) = (−1, −1)
en (i1 , . . . , in ) =
0 otherwise.

8.2 Littelmann paths

Set PR = P ⊗Z R. By a path we mean a piecewise linear continuous map [0, 1] → PR . We
consider paths up to reparametrisation, i.e. π ∼
= π ◦ φ, where φ : [0, 1] → [0, 1] is a piecewise-
linear isomorphism.
Let P = {paths π s.t. π(0) = 0, π(1) ∈ P }. We can define a crystal structure on P. For
π ∈ P define
wt(π) = π(1).

To define ẽi (π), let

hi = min Z ∩ {hπ(t), αi∨ i | 0 ≤ t ≤ 1} ≤ 0.

That is, hi is the smallest integer in hαi∨ , π[0, 1]i (note that since π(0) = 0, we have hi ≤ 0). If
hi = 0, set ẽi (π) = 0 (this is not the path that stays at 0, but rather the extra element in the
crystal). Otherwise hi < 0, then take the smallest t1 > 0 such that hπ(t1 ), αi∨ i = hi (i.e. the first
time the path crosses hi ). Moreover, let t0 be the largest t0 < t1 such that hπ(t0 ), αi∨ i = hi + 1.
We will define ẽi π as the path reflecting π[t0 , t1 ] in the hyperplane {λ ∈ PR | hλ, α∨ i = hi + 1},
and then translating π[t1 , 1] while leaving π[0, t0 ] unchanged.

Robert Laugwitz & Henning Seidler 69

Lie Algebras and Their Representations

π(1)

ẽi π

t1
t0 0

hi hi + 1 hi + 2

αi

Expressed in formulas we have


π(t),

 if 0 ≤ t ≤ t0
ẽi (π)(t) = π(t ) + sαi (π(t) − π(t0 )) = π(t) − hπ(t) − π(t0 ), αi∨ iαi , if t0 ≤ t ≤ t1
 0
if t ≥ t1 .

π(t) + α ,
i

Exercise 8.21: Show that εi (π) = −hi .

Example 8.22: Let us compute some examples for sl2 :

!
ẽi • ←− •
α
= • −→ α• ,
− 2i 0 0 i
2
!
ẽi • −→ α• = 0, and
0 i
2
 
ẽi • ←− • ←− • = •α  •,
−αi 0 − 2i 0
!
ẽi α
• • = • −→ • −→ • ,
− 2i 0 0 αi
 
ẽi • −→ • −→ • = 0.
0 αi

If π is a path, let π ∨ be the reversed path, i.e. t 7→ π(1 − t) − π(1). Define

f˜i (π) = (ẽi (π ∨ ))∨ .

Exercise 8.23: P is a crystal with wt, ẽi , f˜i defined as above.

Now, define

P + = paths π s.t. π[0, 1] ⊂ PR+ = {x ∈ PR | hx, αi∨ i ≥ 0 ∀i} .



Observe that if π ∈ P + , then ẽi (π) = 0 for all i.

For π ∈ P + let Bπ be the subcrystal of P generated by π, i.e. Bπ = {f˜i1 f˜i2 . . . f˜ir π}.

70 Robert Laugwitz & Henning Seidler

 8 Crystals

Theorem 8.24 (Littelmann):

(i) If π, π 0 ∈ P + , then
Bπ ∼
= Bπ0 ⇐⇒ π(1) = π 0 (1)
(i.e. crystals with the same endpoint of highest weight paths are isomorphic).

(ii) There is a unique isomorphism of crystals B(π(1)) → Bπ (where B(π(1)) is the crystal of
the irreducible representation L(π(1))) sending the highest weight π(1) to a path π with
endpoint π(1).

Moreover, for paths of the form π(t) = λt, λ ∈ P + , Littlemann give an explicit combinatorically
description of the paths in Bπ .

Example 8.25 (Exercise):

(i) Consider sl3 with simple roots α, β. We want to compute the crystal of the adjoint repre-
sentation. First, show that
 
•
0
  

ẽβ   = • ←− • ,
 
  −α 0
<


•

−(α+β)

and then compute the rest of the crystal and show that you obtain the adjoint represen-
tation of sl3 .

(ii) Consider the root system type G2 . You might have seen before that the smallest non-trivial
representation is 7-dimensional. Compute the crystal for the 7-dimensional representation
of G2 . Further, note that the second smallest non-trivial representation is 14-dimensional
(the adjoint representation). Calculate the crystal for the 14-dimensional representation,
and the tensor product of these two representations, if you feel like.

Remark 8.26: Littlemann’s Theorem 8.24 allows us to define B(λ) explicitly, without using
L(λ), and we can also prove Weyls character formula

det wew (λ + ρ) − ρ
P
dim B(λ) = w∈W Q
(1 − e−α )

without the use of L(λ). This gives a proof of the existence of crystals (Theorem 8.7) without
quantum groups. To prove this, we can build ch L(λ), and indeed L(λ) (and the crystal variants),
one root at a time. This is called the Demazure character formula.

Robert Laugwitz & Henning Seidler 71

Lie Algebras and Their Representations

For every w ∈ W , there is an approximation to L(λ) given by Lw (λ) if Lw (λ) is the n+ -

submodule of L(λ) generated by vectors vwλ , where vλ is the highest weight vector of L(λ), and
vwλ is the vector in wL(λ)λ (1-dimensional submodule).

Theorem 8.27 (Demazure character formula):

ch Lw (λ) = Dw (eλ ),

where w = si1 · . . . · sir is a reduced simple reflection decomposition of w (i.e. r minimal), and
Dw = Dsi1 · . . . · Dsir with Dsi : Z[P ] → Z[P ] defined by

f − si (f )
Dsi (f ) =
1 − e−αi 
f
= (Id +si )
1 − e−αi
1 αi αi
= αi −αi (f e 2 − si (f e 2 )), ∀f ∈ Z[P ].
e 2 −e 2

Note that 
eλ + eλ−αi + . . . + esi λ ,

 if hλ, αi∨ i ≥ 0,
Dsi (eλ ) = 0 if hλ, αi∨ i = −1,

−(eλ+αi + . . . + esi λ−αi ), if hλ, αi∨ i < −1.


Additional sources
Grojnowski, I. (2010), ‘Introduction to lie algebras and their representations, lecture notes’.

Kac, V. (2010), ‘Introduction to lie algebras, lecture notes’, [Link]

18.745/[Link].

Schweigert, C. (2004), ‘Einfhrung in die theorie der lieschen algebren, vorlesungsscript’, http:
//[Link]/home/schweigert/.

72 Robert Laugwitz & Henning Seidler