MATH 223A NOTES 2011 LIE ALGEBRAS 31

8.4. Root strings. This subsection is based on Erdmann and Wildon “Introduction
to Lie Algebras” an undergraduate textbook which is the place to look if you don’t
understand something. We first review what we have so far using an example.

8.4.1. example: sl(3, F ). Let L = sl(3, F ). This is 8 dimensional with H being the 2-
dimensional subalgebra of diagonal matrices with trace zero.
  
 h1 0 0 
H =  0 h2 0  : h1 + h2 + h3 = 0
 
0 0 h3

The off-diagonal entries have an obvious basis given by xij , the matrix with 1 in the ij
position and 0 elsewhere:
     
0 1 0 0 0 0 0 0 1
x12 = 0 0 0 , x23 = 0 0 1 , x13 = 0 0 0
0 0 0 0 0 0 0 0 0

and x21 , x32 , x31 . Note that each xij is an eigenvector: For example,

[hx12 ] = hx12 − x12 h = (h1 − h2 )x12

So,
(1) x12 ∈ Lα where α(h) = h1 − h2 ,
(2) x23 ∈ Lβ where β(h) = h2 − h3 and
(3) x13 ∈ Lα+β with (α + β)(h) = h1 − h3 .
We also have
(4) x21 ∈ L−α
(5) x32 ∈ L−β
(6) x31 ∈ L−α−β .
This gives the root space decomposition:

L = H ⊕ Lα ⊕ Lβ ⊕ Lα+β ⊕ L−α ⊕ L−β ⊕ L−α−β

What is hα ?
 
1 0 0
[x12 x21 ] = 0 −1 0 = hα
0 0 0

This matrix is hα since α(hα ) = h1 − h2 = 2. xa = x12 , yα = x21 and

Sα = span(x12 , x21 , hα )

Find the decomposition of L into irreducible Sα -modules.

32 MATH 223A NOTES 2011 LIE ALGEBRAS

One component is V (2) = Lα ⊕ F hα ⊕ L−α . To find the others we draw the root spaces
in the following pattern:
Lα+β ⊕ Lβ ∼
= V (1)

Lα ⊕ H ⊕ L−α ∼
= V (2) ⊕ V (0)

L−β ⊕ L−α−β ∼
= V (1)
Claim: Lα+β ⊕ Vβ ∼= V (1).
Pf: Since xα ∈ Lα , ad xα sends Lα+β to L2α+β = 0. Therefore, x13 = xα+β is a maximal
vector. But hα (x13 ) = (h1 − h3 )x13 = x13 . So, the weight is 1. So, x13 generates a
submodule isomorphic to V (1). (Recall that V (λ) is generated by a maximal vector with
maximal weight λ and that λ is a nonnegative integer.)
Similarly, L−β ⊕ L−α−β ∼ = V (1). This leaves only a one dimensional submodule V (0)
contained in H.
8.4.2. α-root string. We now return to the general case. Recall that we have a root space
decomposition: �
L∼=H⊕ Lα
α∈Φ

Definition 8.4.1. If α ∈ Φ and β ∈ Φ∪{0}, the α-root string through β is the Sα -module:
�
M= Lβ+cα
c∈F

(We will prove shortly that only integer values of c can occur in any root string.)
Proposition 8.4.2. For any α ∈ Φ, L is a direct sum of α-root strings.
From the example we know that the α-root string though 0 is not irreducible in general.
Proposition 8.4.3. If α ∈ Φ then Lα is 1-dimensional. Furthermore, the only multiples
of α which are roots are ±α.
Proof. Consider the α-root string though 0:
�
M =H⊕ Lα
cα∈Φ

Let K = ker(α : H → F ). Then H = K ⊕ F hα .

Claim: K is an Sα -module.
Pf: For any k ∈ K we have [xα k] = −α(k)xα = 0. Similarly, [yα k] = 0 and [hk] = 0
since h, k ∈ H. So, every nonzero element of K generated an Sα -module isomorphic to
V (0). This implies that
�
M/K ∼ = W = F hα ⊕ Lcα
cα∈Φ
 MATH 223A NOTES 2011 LIE ALGEBRAS 33

is an Sα -module. But the weight zero part of W is W0 = F hα . So, W contains only one
irreducible V (even). Since we know that W contains Sα ∼ = V (2), there cannot be any
other V (even). So, L2α = 0. In other words, twice a root cannot be a root. But then 12 α
also cannot be a root since twice of it is a root. And this implies that W1 = Lα/2 = 0.
So, W = V (2) is irreducible. This also implies that Lα is one dimensional. �
Corollary 8.4.4. If α, β ∈ Φ and β is not ±α then the α-root string through β is irre-
ducible and has the form:
V (m) ∼
= Lβ+qα ⊕ Lβ+(q−1)α ⊕ · · · ⊕ Lβ−rα
and m = q + r. And β(hα ) = q − r is an integer. (β(hα ) are the Cartan integers.)
Proof. Let M be the α-root string through β. Then M0 = 0 since the root string does
not go through 0. (The proof of the last proposition showed that Sα = Lα ⊕ F hα ⊕ L−α
is the only α-root string through 0.) Therefore, M is a direct sum of V (odd)s. So, the
α-root string has only Modd but a difference of 2 in hα -weights corresponds to a difference
of α in roots since α(hα ) = 2. So, the root string contains only Lβ+kα for integer k. One
of these is M1 and thus is 1-dimensional. So, M is irreducible.
If M ∼
= V (m) then Mm = Lβ+qα and M−m = Lβ−rα . The dimension of M is q + r + 1 =
m + 1. This implies that
q + r = m = (β + qα)(hα ) = β(hα ) + 2q
So, β(hα ) = r − q ∈ Z. �

8.5. Inner product on H ∗ . We just proved that β(hα ) ∈ Z. We will rephrase this in
terms of the inner product on H ∗ .
Recall that hα = κ(t2tαα,tα ) . And, by definition of tβ we have:

2κ(tβ , tα ) 2(β, α)
β(hα ) = κ(tβ , hα ) = = ∈Z
κ(tα , tα ) (α, α)
where we used the definition of the inner product on H ∗ :
(α, β) := κ(tα , tβ ).
Proposition 8.5.1. The set of roots Φ spans H ∗ .
Proof. Suppose not. Then there is some h ∈ H so that α(h) = 0 for all α ∈ Φ. But then
[hx] = α(h)x = 0 for all x ∈ Lα and [hx] = 0 for all x ∈ H since H is abelian. So, h
commutes with all the generators of L and is therefore in Z(L). But Z(L) = 0 since L is
semisimple. �
∗
�means we can choose a basis for H consisting of roots: α1 , · · · , αn . If β ∈ Φ then
This
β = ci αi where ci ∈ F . Recall that we are assuming F is algebraically closed with
characteristic 0. Thus F contains Q, the rational numbers.
Claim 1: ci ∈ Q.
34 MATH 223A NOTES 2011 LIE ALGEBRAS
�
Pf: (β, αj ) = ci (αi , αj ). Therefore,
2(β, αj ) � 2(αi , αj )
= ci
(αj , αj ) (αj , αj )
These fractions are Cartan integers. Also the matrix (αi , αj ) is nonsingular since the form
is nondegenerate. Therefore, ci is the ratio of two determinants of integer matrices and
therefore a rational number.
This implies that the roots lie in the Q-span of α1 , · · · , αn . Let EQ denote this rational
vector space.
Claim 2: The inner product (·, ·) on EQ has rational values and is positive definite.
Pf: Take any λ ∈ H ∗ . Then
� �
(λ, λ) = κ(tλ , tλ ) = Tr(ad tλ ad tλ ) =(1) β(tλ )2 =(2) (β, λ)2
β∈Φ β∈Φ

where (1) follows from the fact that ad tλ acts on Lβ by multiplication by β(tλ ) and
(2) follows from the observation that β(tλ ) = κ(tβ , tλ ) = (β, λ). Thus (·, ·) is positive
semi-definite. Since it is nondegenerate, it must be positive definite.
Let E = EQ ⊗ R.
Theorem 8.5.2. (1) E is a real vector space with a positive definite inner product,
i.e., E is Euclidean space.
(2) Φ spans E and 0 ∈ / Φ.
(3) If α ∈ Φ then −α ∈ Φ and not other real multiple of α is in Φ.
(4) If α, β ∈ Φ then 2(β,α)
(α,α)
∈ Z.
(5) If α, β ∈ Φ then
2(β, α)
β− α∈Φ
(α, α)
Proof. We proved (1)-(4). The class figured out the proof of (5):
2(β, α)
= β(hα ) = r − q
(α, α)
in the notation of the α-root string through β. So,
2(β, α)
β− α = β + (q − r)α
(α, α)
and q ≥ q − r ≥ −r. So, this is one of the roots that occur in the α string through β. �