3.2 Irreducible Representations
3.2.1 Theorem of the Highest Weight
Throughout this sectiongwill be a semisimple Lie algebra. Fix a Cartan subalgebra hof gand letΦ be the root system ofgwith respect to h. Choose a set Φ+ of positive roots inΦ. Ifα∈Φ+then letgα denote the corresponding root space. As in Section 2.5.3, we definen+=Lα∈Φ+gα and n−=Lα∈Φ+g−α. We also write
s(α) =g−α+ [gα,g−α] +gα,
as in Section 2.4.2. Thens(α)is a Lie subalgebra ofgthat has basisx∈gα,h∈ [gα,g−α], andy∈g−αsuch that[x,y] =h,[h,x] =2x, and[h,y] =−2y. Recall that the elementh, called thecoroot associated withα, is unique, and is denoted byhα. Let(π,V)be a representation ofg(we do not assume that dimV <∞). Ifλ∈h∗ then we set
V(λ) ={v∈V :π(h)v=hλ,hivfor allh∈h}, as in the finite-dimensional case. We define
X(V) ={λ ∈h∗:V(λ)6=0}.
We call an element ofX(V)aweightof the representationπand the spaceV(λ)the λ weight spaceofV. We note that ifα∈Φ+andx∈gα thenxV(λ)⊂V(λ+α).
Furthermore, the weight spaces for different weights are linearly independent.
Put a partial order onh∗by definingµ≺λifµ=λ−β1−···−βrfor some (not necessarily distinct)βi∈Φ+and some integerr≥1. We call this theroot order.
From (3.13) we see that on pairsλ,µ withλ−µ∈Q(g)this partial order is the same as the partial order defined by the dual positive Weyl chamber (see the proof of Proposition 3.1.12).
IfV is infinite-dimensional, thenX(V)can be empty. We now introduce a class ofg-modules that have weight-space decompositions with finite-dimensional weight spaces. Ifgis a Lie algebra thenU(g)denotes its universal enveloping algebra; every Lie algebra representation (π,V)of gextends uniquely to an associative algebra representation(π,V)ofU(g)(see Section C.2.1). We use module notation and write π(T)v=T vforT ∈U(g)andv∈V.
Definition 3.2.1.A g-moduleV is a highest-weight representation (relative to a fixed set Φ+ of positive roots) if there areλ ∈h∗and a nonzero vector v0∈V such that (1) n+v0=0, (2) hv0=hλ,hiv0 for allh∈h, and (3) V=U(g)v0.
Setb=h+n+. A vectorv0satisfying properties (1) and (2) in Definition 3.2.1 is called ab-extremevector. A vectorv0satisfying property (3) in Definition 3.2.1 is called ag-cyclic vectorfor the representation.
Lemma 3.2.2.Let V be a highest-weight representation ofgas in Definition3.2.1.
Then
V=Cv0⊕M
µ≺λ
V(µ), (3.21)
3.2 Irreducible Representations 149
withdimV(µ)<∞for allµ. In particular,dimV(λ) =1andλis the unique max- imal element inX(V), relative to the root order; it is called thehighest weightof V .
Proof. Sinceg=n−⊕h⊕n+, the Poincar´e–Birkhoff–Witt theorem implies that there is a linear bijectionU(n−)NU(h)NU(n+) //U(g)such that
Y⊗H⊗X7→Y HX forY ∈U(n−),H∈U(h), andX∈U(n+)
(see Corollary C.2.3). We haveU(h)U(n+)v0=Cv0, so it follows thatV=U(n−)v0. Thus V is the linear span of v0 together with the elements y1y2···yrv0, for all yj∈g−β
j withβj∈Φ+andr=1,2, . . .. Ifh∈hthen hy1y2···yrv0=y1y2···yrhv0+
∑
ri=1
y1···yi−1[h,yi]yi+1···yrv0
= hλ,hi −
r
i=1
∑
hβi,hiy1···yrv0.
Thusy1y2···yrv0∈V(µ), whereµ=λ−β1−···−βrsatisfiesµ≺λ. This implies (3.21). We have dimg−βj =1, and for a fixedµ there is only a finite number of choices ofβj∈Φ+such thatµ=λ−β1− ··· −βr. Hence dimV(µ)<∞. ut Corollary 3.2.3.Let(π,V)be a nonzero irreducible finite-dimensional representa- tion ofg. There exists a unique dominant integralλ ∈h∗such thatdimV(λ) =1.
Furthermore, everyµ∈X(V)withµ6=λ satisfiesµ≺λ.
Proof. By Theorem 3.1.16 we know thatX(V)⊂P(g). Letλ be any maximal ele- ment (relative to the root order) in the finite setX(V). Ifα∈Φ+andx∈gα then xV(λ)⊂V(λ+α). ButV(λ+α) =0 by the maximality ofλ. Henceπ(n+)V(λ) = 0. Let 06=v0∈V(λ). ThenL=U(g)v06= (0)is ag-invariant subspace, and hence L=V. Now apply Lemma 3.2.2. The fact thatλ is dominant follows from the rep- resentation theory ofsl(2,C)(Theorem 2.3.6 applied to the subalgebras(α)). ut Definition 3.2.4.If(π,V)is a nonzero finite-dimensional irreducible representation ofgthen the elementλ ∈X(V)in Corollary 3.2.3 is called thehighest weight of V. We will denote it byλV.
We now use the universal enveloping algebraU(g)to prove that the irreducible highest-weight representations ofg(infinite-dimensional, in general) are in one-to- one correspondence withh∗via their highest weights. By the universal property of U(g), representations ofgare the same as modules forU(g). We define aU(g)- module structure onU(g)∗by
g f(u) =f(ug) forg,u∈U(g)and f ∈U(g)∗. (3.22) Clearlyg f∈U(g)∗and the mapg,f7→g fis bilinear. Also forg,g0∈U(g)we have
g(g0f)(u) =g0f(ug) = f(ugg0) = (gg0)f(u),
so definition (3.22) does give a representation ofU(g)onU(g)∗. Here the algebra U(g)is playing the role previously assigned to the algebraic group G, while the vector spaceU(g)∗is serving as the replacement for the space of regular functions onG.
Theorem 3.2.5.Letλ∈h∗.
1. There exists an irreducible highest-weight representation (σ,Lλ) of g with highest weightλ.
2. Let (π,V) be an irreducible highest-weight representation of g with highest weightλ. Then(π,V)is equivalent to(σ,Lλ).
Proof. To prove (1) we useλ to define a particular element ofU(g)∗as follows.
Sincehis commutative, the algebraU(h)is isomorphic to the symmetric algebra S(h), which in turn is isomorphic to the polynomial functions on h∗. Using these isomorphisms, we define λ(H), forH ∈h, to be evaluation at λ. This gives an algebra homomorphismH7→λ(H)fromU(h)toC. IfH=h1···hnwithhj∈h, then
λ(H) =hλ,h1i···hλ,hni.
For any Lie algebramthere is a unique algebra homomorphismε:U(m) //C such that
ε(1) =1 and ε(U(m)m) =0.
WhenZ∈U(m)is written in terms of a P–B–W basis,ε(Z)is the coefficient of 1. Combining these homomorphisms (whenm=n±) with the linear isomorphism U(g)∼=U(n−)⊗U(h)⊗U(n+)already employed in the proof of Lemma 3.2.2, we can construct a unique element fλ∈U(g)∗that satisfies
fλ(Y HX) =ε(Y)ε(X)λ(H) for Y ∈U(n−),H∈U(h), and X∈U(n+). Ifh∈h, thenY HX h=Y HhX+Y HZwithZ= [X,h]∈n+U(n+). Sinceε(Z) =0 andλ(Hh) =hλ,hiλ(H), we obtain
h fλ(Y HX) =fλ(Y HhX) +fλ(Y HZ) =hλ,hifλ(Y HX). (3.23) Furthermore, ifx∈n+thenε(x) =0, so we also have
x fλ(Y HX) =fλ(Y HX x) =ε(Y)λ(H)ε(X)ε(x) =0. (3.24) DefineLλ ={g fλ : g∈U(g)}to be theU(g)-cyclic subspace generated by fλ relative to the action (3.22) ofU(g)onU(g)∗, and letσbe the representation ofgon Lλ obtained by restriction of this action. Then (3.23) and (3.24) show that(σ,Lλ) is a highest-weight representation with highest weightλ. SinceHX∈U(h)U(n+) acts on fλ by the scalarλ(H)ε(X), we have
3.2 Irreducible Representations 151
Lλ ={Y fλ :Y∈U(n−)}. (3.25) We now prove thatLλ is irreducible. Suppose 06=M⊂Lλ is ag-invariant sub- space. By Lemma 3.2.2 we see thatMhas a weight-space decomposition underh:
M=M
µλ
M∩Lλ(µ).
If µλ then there are only finitely many ν∈h∗such that µνλ. (This is clear by writingλ−µ,ν−µ, and λ−ν as nonnegative integer combinations of the simple roots.) Hence there exists a weight ofM, sayµ, that is maximal. Take 06= f ∈M(µ). Then n+f =0 by maximality of µ, and henceX f =ε(X)f for allX∈U(n+). By (3.25) we know that f =Y0fλ for someY0∈U(n−). Thus for Y ∈U(n−),H∈U(h), andX∈U(n+)we can evaluate
f(Y HX) =X f(Y H) =ε(X)f(Y H) =ε(X)fλ(Y HY0).
ButHY0=Y0H+[H,Y0], and[H,Y0]∈n−U(n−). Thus by definition of fλwe obtain f(Y HX) =ε(X)ε(Y)ε(Y0)λ(H) =ε(Y0)fλ(Y HX).
Thusε(Y0)6=0 and f =ε(Y0)fλ. This implies thatM=Lλ, and henceLλ is irre- ducible, completing the proof of (1).
To prove (2), we note by Lemma 3.2.2 thatV has a weight-space decomposition and dim(V/n−V) =1. Letp:V //V/n−V be the natural projection. Then the re- striction ofptoV(λ)is bijective. Fix nonzero elementsv0∈V(λ)andu0∈V/n−V such that p(v0) =u0. The representationπ extends canonically to a representa- tion ofU(g)onV. ForX∈U(n+)we haveX v0=ε(X)v0, forH∈U(h)we have Hv0=λ(H)v0, while forY ∈U(n−)we haveY v0≡ε(Y)v0 modn−V. Thus
p(Y HX v0) =ε(X)λ(H)p(Y v0) =ε(Y)ε(X)λ(H)u0=fλ(Y HX)u0. (3.26) For anyg∈U(g)andv∈V, the vectorp(gv)is a scalar multiple ofu0. If we fix v∈V, then we obtain a linear functionalT(v)∈U(g)∗such that
p(gv) =T(v)(g)u0 for allg∈U(g)
(since the mapg7→p(gv)is linear). The mapv7→T(v)defines a linear transforma- tionT :V //U(g)∗, and we calculated in (3.26) thatT(v0) = fλ. Ifx∈gthen
T(xv)(g)u0=p(gxv) =T(v)(gx)u0= (xT(v))(g)u0.
HenceT(xv) =xT(v). SinceVis irreducible, we haveV=U(g)v0, and soT(V) = U(g)fλ=Lλ. ThusT is ag-homomorphism fromVontoLλ.
We claim that Ker(T) =0. Indeed, ifT(v) =0 then p(gv) =0 for allg∈U(g).
HenceU(g)v⊂n−V. Sincen−V 6=V, we see thatU(g)vis a proper g-invariant
subspace ofV. ButV is irreducible, soU(g)v=0, and hencev=0. ThusT gives
an isomorphism betweenVandLλ. ut
The vector spaceLλ in Theorem 3.2.5 is infinite-dimensional, in general. From Corollary 3.2.3 a necessary condition for Lλ to be finite-dimensional is thatλ be dominant integral. We now show that this condition is sufficient.
Theorem 3.2.6.Let λ ∈h∗ be dominant integral. Then the irreducible highest weight representation Lλ is finite-dimensional.
Proof. We writeX(λ)for the set of weights ofLλ. Sinceλ is integral, we know from Lemma 3.2.2 thatX(λ)⊂P(g)⊂h∗
R. Fix a highest-weight vector 06=v0∈ Lλ(λ).
Letα∈∆be a simple root. We will show thatX(λ)is invariant under the action of the reflectionsα onh∗R. The argument proceeds in several steps. Fix a TDS basis x∈gα,y∈g−α, andh= [x,y]∈h for the subalgebra s(α). Setn=hα,hiand vj=yjv0. Thenn∈N, sinceλ is dominant integral.
(i) Ifβ is a simple root andz∈gβ, thenzvn+1=0 .
To prove (i), suppose first thatβ6=α. Thenβ−αis not a root, since the coefficients of the simple roots are of opposite signs. Hence [z,y] =0, so in this casezvj= yjzv0=0 . Now suppose thatβ=α. Then we may assume thatz=x. We know that xvj=j(n+1−j)vj−1by equation (2.16). Thusxvn+1=0.
(ii) The subspaceU(s(α))v0is finite-dimensional.
We have n+vn+1=0 by (i), since the subspacesgβ withβ ∈∆ generaten+ by Theorem 2.5.24. Furthermorevn+1∈Lλ(µ)withµ=λ−(n+1)α. Hence theg- submoduleZ=U(g)vn+1 is a highest-weight module with highest weightµ. By Lemma 3.2.2 every weightγofZsatisfiesγµ. Sinceµ≺α, it follows thatZis a properg-submodule ofLλ. ButLλis irreducible, soZ=0. The subspaceU(s(α))v0 is spanned by{vj: j∈N}, since it is a highest-weights(α)module. Butvj∈Zfor
j≥n+1, sovj=0 for j≥n+1. HenceU(s(α))v0=Span{v0, . . . ,vn}. (iii) Letv∈Lλ be arbitrary and letF=U(s(α))v. Then dimF<∞.
SinceLλ =U(g)v0, there is an integer jsuch thatv∈Uj(g)v0, whereUj(g)is the subspace ofU(g)spanned by products of j or fewer elements ofg. By Leibniz’s rule[g,Uj(g)]⊂Uj(g). HenceF⊂Uj(g)U(s(α))v0. SinceUj(g)andU(s(α))v0 are finite-dimensional, this proves (iii).
(iv) sαX(λ)⊂X(λ).
Letµ∈X(λ)and let 06=v∈Lλ(µ). Since[h,s(α)]⊂s(α), the spaceF=U(s(α))v is invariant underhand the weights ofhonFare of the formµ+kα, wherek∈Z. We know thatF is finite-dimensional by (iii), so Theorem 2.3.6 implies thatF is equivalent to
F(k1)L···LF(kr) (3.27)
3.2 Irreducible Representations 153
as an s(α)-module. Now hµ,himust occur as an eigenvalue of h in one of the submodules in (3.27), and hence−hµ,hialso occurs as an eigenvalue by Lemma 2.3.2. It follows that there exists a weight of the formµ+kαinX(λ)with
hµ+kα,hi=−hµ,hi.
Thus 2k=−2hµ,hi. Hencesαµ=µ+kα∈X(λ), which proves (iv).
We can now complete the proof of the theorem. The Weyl groupW is generated by the reflectionssα withα ∈∆ by Theorem 3.1.9. Since (iv) holds for all simple rootsα, we conclude that the setX(λ)is invariant underW. We already know from Lemma 3.2.2 that dimLλ(µ)<∞for allµ. Thus the finite-dimensionality ofLλ is a consequence of the following property:
(v) The cardinality ofX(λ)is finite.
Indeed, letµ∈X(λ). Thenµ∈h∗
R. By Proposition 3.1.12 there existss∈W such that ξ =sµ is in the positive Weyl chamberC. Since ξ ∈X(λ), we know from Lemma 3.2.2 thatξ =λ−Q, whereQ=β1+···+βr withβi∈Φ+. Let(·,·)be the inner product onh∗
Ras in Remark 3.1.5. Then
(ξ,ξ) = (λ−Q,ξ) = (λ,ξ)−(Q,ξ)
≤(λ,ξ) = (λ,λ−Q)
≤(λ,λ).
Here we have used the inequalities(λ,βi)≥0 and(ξ,βi)≥0, which hold for ele- ments ofC. SinceW acts by orthogonal transformations, we have thus shown that
(µ,µ)≤(λ,λ). (3.28)
This implies thatX(λ)is contained in the intersection of the ball of radiuskλkwith the weight latticeP(g). This subset ofP(g)is finite, which proves (v). ut