Root systems

4.4.16 Remark With the notation above, u:=t+ X

α∈∆⁺

R(Eα−Fα) +Ri(Eα+Fα) is a compact real form ofg. Indeed, if

X=iH + X

α∈∆⁺

a_α(E_α−F_α) +ib_α(E_α+F_α)

withH∈h_Randa_α,b_α ∈R, then the Cartan-Killing formβofghas β(X, X) =−β(H, H)− 4

||α||² X

α∈∆⁺

a²_α+b²_α<0.

In the next section, we shall use our knowledge about representations ofg[α]to investigate the structure of root systems.

4.5. ROOT SYSTEMS 67 and

(4.5.3) The only multiples ofα ∈∆which are roots are±α.

Next, let α, β ∈ ∆, β 6= ±α, and consider ρ = ad|g[α] acting on V = P

k∈Zg_β+kα. ForX ∈g_β+kα, we have that

ρ( ¯H_α)X = [ ¯H_α, X] = (β+kα)( ¯H_α)X

=

2hβ, αi

||α||² + 2k

X.

Sincedimg_β+kα = 1forβ+kα ∈ ∆, we see thatV is irreducible and the sequence of eigenvalues ofρ( ¯H_α)has the form2^h_||^β,α_α||2ⁱ+ 2k, where−p≤k≤ q, for integersp,q ≥0. Owing to (4.5.1), this set is symmetric with respect to0. In particular,−

2^hβ,αi

||α||² −2p

= 2^hβ,αi

||α||² + 2q. It follows that

(4.5.4) Forα,β ∈∆,β 6=±α, there exist integersp,q ≥0such that 2^h_||^β,α_α||2ⁱ =p−qwhereβ+kα∈∆if and only if−p≤k≤q.

{β+kα∈∆| −p≤k≤q}is called theα-string of roots throughβ.

Further,

(4.5.5) [g_α,g_β] =g_α+βifα,β,α+β ∈∆.

In factα+β ∈∆impliesq ≥1. If it were[g_α,g_β] = 0, thenP

−p≤k≤0g_β+kα would be a proper invariant subspace ofV =P

−p≤k≤qg_β+kα, butV ad|g[α] -irreducible in view of the above discussion.

We further refine the information about the length of strings of roots.

(4.5.6) Ifα,β∈∆,β6=±α, then2^h_||^β,α_α||2ⁱ ∈ {0,±1,±2,±3}.

Indeed we already know2^h_||^β,α_α||2ⁱis an integer; alsoα,βare linearly indepen-dent by (4.5.3), so the Cauchy-Schwarz inequality says that

0≤

2hβ, αi

||α||² ·

2hα, βi

||β||²

= 4 hβ, αi²

||α||²||β||² <4.

Further:

(4.5.7) Ifα,β∈∆,β6=±α, then theα-string throughβhas length≤4.

In fact, that string coincides with theα-string throughγ = β−pα, that is, {γ+kα |0≤k≤p+q}, and due to (4.5.4) and (4.5.6),

−2hγ, αi

||α||² =p+q≤3.

Weyl group

Consider nexth^∗_R = P

α∈∆Rα(resp.h_R) equipped with the inner product h,iinduced from the Cartan-Killing form. Thereflectionin the hyperplane α^⊥(resp.kerα) is

(4.5.8) s_α(λ) :=λ−2hλ, αi

||α||²α resp.s_α(H_λ) :=H_λ−2hλ, αi

||α||²H_α . TheWeyl groupW ofgrelative tohis the group generated by the orthogonal transformations{s_α|α∈∆}.

(4.5.9) W permutes the roots; in particular,W is finite.

In fact, using (4.5.4), for eachβ ∈ ∆we have sα(β) = β + (q −p)α ∈ ∆, since−p≤q−p≤q.

For eachα ∈ ∆,kerα ⊂ h_R is called asingular hyperplane; ∪α∈∆kerα is called the set ofsingular elementsof h_R, and its complement h^′_R in h_R is called the set ofregular elementsofh_R. The connected components of ofh^′_R are calledWeyl chambers.

(4.5.10) W acts simply transitively on the set of Weyl chambers.

Indeed it follows from (4.5.9) thatW permutes the chambers. Given two chambersC andC^′, we choose H ∈ C, H^′ ∈ C^′ and join them by an arc in h_Rthat at each time meets at most one singular hyperplane; the product in order of the reflections in the hyperplanes met by the arc is an element of W that sendsCtoC^′.¹

Positive roots and simple roots

Fix a Weyl chamberC. It is clear that each root takes a constant sign onC. We define

∆⁺:={α∈∆|α(C)>0}.

In view of (4.4.9), this determines a partition of the root systems into two halves:

(4.5.11) ∆ = ∆⁺∪˙(−∆⁺); andα,β ∈∆⁺,α+β∈∆impliesα+β ∈

∆⁺.

∆⁺is called apositive root systemand its elements are calledpositive roots. A positive root is calledsimpleif it cannot be written as a sum of two positive roots. Denote byΠthe set of simple roots.

(4.5.12) Ifα,β ∈Π,α6=βthenα−β 6∈∆andhα, βi ≤0.

1Missing argument for simple transitivity. Humphreys, use repr theory.

4.5. ROOT SYSTEMS 69 Indeedα =β+ (α−β)andβ =α+ (β−α); ifα−βis a root, then either it is positive or its negative is positive; in either case we reach a contradiction to the fact thatαandβare simple. Nowp= 0in (4.5.4), so2^hβ,αi

||α||² =−q≤0.

Since all simple roots lie in the same open half-space (as they are all positive), with all mutual angles non-acute by (4.5.12), an elementary argu-ment in linear algebra shows that they are linearly independent. Therefore:

(4.5.13) Every positive root can be written uniquely as a non-negative integral linear combination of simple roots; andΠ is a basis ofh^∗_RoverR.

The fact that every positive root is a positive sum of simple roots follows immediately from the definition of simple root; the uniqueness part fol-lows from the linear independence ofΠ; and the last statement now follows from (4.4.10).

Now we enumerateΠ ={α1, . . . , αr}, wherer= dim_Ch, and define a_ij = 2hα_j, α_ii

||α_i||² .

The aij are called Cartan integers and the r ×r matrix (aij) is called the Cartan matrix.

A positive root is built up through roots by starting with a simple root and adding one simple root at a time. Indeed, define the level of β = P_r

i=1m_iα_ito beP_r

i=1m_i.

(4.5.14) Every positive root of levelm+ 1 can be written in at least one way as a the sum of a positive root of level m and a simple root.

To check this, supposeβ=Pr

i=1m_iα_ihas levelm+ 1. Then 0<hβ, βi=

Xr i−1

m_i0hβ, α_ii

shows thathβ, α_i0iandm_i0 >0for somei₀. By (4.5.4),β−α_i0 ∈∆is a root, as wished.

Now (4.5.11) and (4.5.14) imply thatΠand(aij)determine∆.

Isomorphism theorem

Consider a standard tripleHi,Ei,Fispanningg[αi]for eachi= 1, . . . , r, as in (4.4.15). Then we have the relations

[H_i, H_j] = 0, [E_i, F_i] =δ_ijH_i, [H_i, E_j] =a_ijE_j, [H_i, F_j] =−a_ijF_j,

for alli,j. For eachα ∈∆⁺, use (4.5.13) to write (4.5.15) α=α_i1 +· · ·+α_is,

sum of simple roots, where each partial sum is a root; this decomposition is unique, up to reordering. Now define

Eα= [Eis,[Eis−1,· · ·,[Ei2, Ei1]· · ·]]

F_α= [F_is,[F_is−1,· · ·,[F_i2, F_i1]· · ·]].

(4.5.16)

In view of (4.5.5),

(4.5.17) {Hi}¹≤i≤r∪ {Eα, Fα}α∈∆⁺

is a basis ofg. Now a delicate argument of induction onsshows that the multiplication table of this basis has rational entries determined by the Car-tan matrix. Here the difficulty lies in the fact that a different order in (4.5.15) entails different vectorsE_α^′,F_α^′ in (4.5.16), and the main step is to show that E_α^′ (resp.F_α^′) is a rational multiple ofE_α (resp.F_α), where the multiplica-tive constant is determined by the permutation and the Cartan matrix. This proves:

4.5.18 Theorem (Isomorphism Theorem) Letgandg^′be complex semisimple Lie algebras with corresponding CSAhandh^′. Denote by∆and∆^′the associated root systems. IfΦ :h→h^′is a linear isometry with respect to the Cartan-Killing forms such thatΦ^∗∆^′ = ∆, thenΦextends to an isomorphismg→g^′.

4.5.19 Corollary Every element of the Weyl groupW of a complex semisimple Lie algebragis induced by an automorphism ofg.

Proof. Each generator sα of W as in (4.5.8) is a linear isometry of the

CSA that preserves the root system.

In particular, a complex semisimple Lie algebra is determined, up to isomorphism, byΠand(aij). However, three arbitrary choices have been made in the construction of these invariants:

(a) A CSAhofg.

(b) A Weyl chamberCofgrelative toh.

(c) A labeling of the simple roots inΠ.

In case of (a), any two CSA’s ofg are conjugate under an inner automor-phism (Proposition 4.2.1). In case of (b), any two Weyl chambers are related by the action of an element of the Weyl group, and the latter is induced by an automorphism ofg(Corollary 4.5.19). We avoid the choice in (c) by in-troducing theDynkin diagramof g: taker nodes, one for each simple root inΠ; ifi6=j, joinαi toαj byaijaji = 4 cos²∠(αi, αj) ∈ {0,1,2,3}lines; if

||α_i||>||α_j||, we further orient the lines in the direction of the shorter root.

Note thata_ij 6= 0if and onlya_ji6= 0, and in this case ^a_a^ij

ji = ^||_||^α_α^j||2

i||².

4.6. CLASSIFICATION OF ROOT SYSTEMS 71