3.1 Roots and Weights
3.1.2 Root Reflections
Lemma 3.1.4.Let G=SO(C2l,B). The subgroup Rl⊂WGis normal, and WGis the semidirect product of Rl andπ¯(Sl). The action of WGon the coordinate functions inO[H]is by xi7→ xσ(i)±1 (i=1, . . . ,l), for every permutationσand choice±1 of exponents with anevennumber of negative exponents.
Proof. By the same argument as in the proof of Lemma 3.1.2 we see that the normal- izer ofHinO(C2l,B)is given by theHcosets of the elementsβFπ(σ)asσ ranges overSl andF ranges over all subsets of{1, . . . ,l}. Sinceπ(σ)∈NormG(H),we haveβFπ(σ)∈NormG(H)if and only if Card(F)is even. ut
3.1 Roots and Weights 133
The action of the Weyl group onh∗can be expressed in terms of these reflection operators, as follows:
Lemma 3.1.6.Let W =NormG(H)/H be the Weyl group of G. Identify W with a subgroup ofGL(h∗)by the natural action of W onX(H).
1. For everyα∈Φthere exists w∈W such that w acts onh∗by the reflection sα . 2. W·∆=Φ .
3. W is generated by the reflections{sα : α∈∆}. 4. If w∈W and wΦ+=Φ+then w=1.
5. There exists a unique element w0∈W such that w0Φ+=−Φ+.
Proof. We proceed case by case using the enumeration of∆ from Section 2.4.3 and the description ofW from Section 3.1.1. In all cases we use the characterization (3.5) ofsα.
Type A (G=SL(l+1,C)): Here W ∼=Sl+1 acts on h∗ by permutations of ε1, . . . ,εl+1. Letα=εi−εj. Then
hεk,hαi=
1 ifk=i,
−1 ifk=j, 0 otherwise.
This shows that sαεk=εσ(k), where σ is the transposition (i,j)∈Sl+1. Hence sα∈W, which proves (1). Property (2) is clear, since the transpositionσ= (i+1,j) carries the simple rootεi−εi+1toεi−εjfor anyj6=i. Property (3) follows from the fact that any permutation is the product of transpositions(i,i+1). If the permutation σ preservesΦ+thenσ(i)<σ(j)for alli<jand henceσ is the identity, which proves (4). To prove (5), letσ∈Sl+1act byσ(1) =l+1,σ(2) =l, . . . ,σ(l+1) =1 and letw0∈W correspond toσ. Thenw0αi=εσ(i)−εσ(i+1)=−αl+1−i for i= 1, . . . ,l. Hencew0is the desired element. It is unique, by (4).
The root systems of typesBl,Cl, andDl each contain subsystems of typeAl−1 and the corresponding Weyl groups containSl. Furthermore, the set of simple roots for these systems is obtained by adjoining one root to the simple roots forAl−1. So we need consider only the roots and Weyl group elements not in theAl−1subsystem in these remaining cases. We use the same notation for elements ofW in these cases as in Section 3.1.1.
Type B (G=SO(C2l+1,B)): Forα=εiwe havesαεi=−εi, whereassαεj=εjif i6=j. Sosαgives the action ofγionh∗. Whenα=εi+εjwithi6=j, thensαεi=−εj
andsαfixesεkfork6=i,j. Hencesαhas the same action onh∗asγiγjϕ(σ), where σ= (i j), proving (1). Sinceγjtransformsεi−εjintoεi+εjand the transposition σ = (il)interchangesεi andεl, we obtain (2). We know from Lemma 3.1.3 that W contains elements that act on h∗ by sα1, . . . ,sαl−1 and generate a subgroup of W isomorphic toSl. Combining this with the relationγi=ϕ(σ)γlϕ(σ), where σ = (il), we obtain (3). HenceW acts by orthogonal transformations ofh∗R. Ifw preservesΦ+thenwmust permute the set of short positive roots, and sowεi=εσ(i)
for someσ ∈Sl. Arguing as in the case of typeAl we then conclude that σ=1.
Letw0be the product of the cosetsγiHfori=1, . . . ,l. Thenw0acts by−Ionh∗, so it is the desired element.
Type C (G=Sp(C2l,Ω)): Sinces2εi=sεi, we can use Lemma 3.1.2 and the same argument as for typeB, replacingγibyτi. In this case an elementw∈W that pre- servesΦ+will permute the set{2εi}oflongpositive roots. Againw0acts by−I.
Type D (G=SO(C2l,B)): Forα =εi+εj the reflectionsα has the same action onh∗asβiβjπ(σ), whereσ= (i j). This proves (1). We know from Lemma 3.1.4 thatWcontains elements that act onh∗bysα1, . . . ,sαl−1 and generate a subgroup of W isomorphic toSl. Since we can move the simple rootαl =εl−1+εl toα by a permutationσ, we obtain (2).
This same permutation action conjugates the reflectionsαl tosα, so we get (3).
IfwpreservesΦ+then for 1≤i<j≤lwe have w(εi+εj) =εσ(i)±εσ(j)
for someσ∈Sl withσ(i)<σ(j). Henceσ is the identity andwεi=εifor 1≤ i≤l−1. Sincewcan only change the sign of an even number of theεi, it follows thatwfixesεl, which proves (4). Iflis even, letw0be the product of all the cosets βiHfor 1≤i≤l. Thenw0acts by−I. Iflis odd takew0to be the product of these cosets for 1≤i≤l−1. In this case we havew0αl−1=−αl,w0αl=−αl−1, and w0αi=−αifori=1, . . . ,l−2, which shows thatw0is the desired element. ut Now we consider a semisimple Lie algebragoverC(see Section 2.5); the reader who has omitted Section 2.5 can take gto be a classical semisimple Lie algebra in all that follows. Fix a Cartan subalgebrah⊂gand letΦ be the set of roots of hong(the particular choice ofhis irrelevant, due to Theorem 2.5.28). Choose a setΦ+of positive roots and let∆ ⊂Φ+ be the simple roots (see Section 2.5.3).
Enumerate∆={α1, . . . ,αl}. We introduce a basis forh∗whose significance will be more evident later.
Definition 3.1.7.Thefundamental weights(relative to the choice of simple roots∆) are elements{ϖ1, . . . ,ϖl} ofh∗dual to the coroot basis{αˇ1, . . . ,αˇl}forh∗. Thus (ϖi,αˇj) =δi j for i,j=1, . . . ,l.
Forα ∈Φ define the root reflection sα by equation (3.4). Then sα is the or- thogonal reflection in the hyperplaneα⊥ and acts as a permutation of the setΦ by Theorem 2.5.20 (4). In particular, the reflections in the simple root hyperplanes transform the fundamental weights by
sαiϖj=ϖj−δi jαi. (3.7) Definition 3.1.8.TheWeyl groupof(g,h)is the groupW =W(g,h)of orthogonal transformations ofh∗
Rgenerated by the root reflections.
We note thatW is finite, sincew∈W is determined by the corresponding permuta- tion of the finite setΦ. In the case of the Lie algebra of a classical group, Lemma 3.1.6 (3) shows that Definition 3.1.8 is consistent with that of Section 3.1.1.
3.1 Roots and Weights 135
Theorem 3.1.9.LetΦ be the roots,∆ ⊂Φ a set of simple roots, and W the Weyl group for a semisimple Lie algebragand Cartan subalgebrah⊂g. Then all the properties(1)–(5)of Lemma 3.1.6are satisfied.
Proof. Property (1) is true by definition ofW. LetW0⊂W be the subgroup gener- ated by the reflections{sα1, . . . ,sαl}. The following geometric property is basic to all arguments involving root reflections:
Ifβ∈Φ+\ {αk}, then sαkβ ∈Φ+\ {αk}. (3.8) (Thus the reflection in a simple root hyperplane sends the simple root to its negative and permutes the other positive roots.) To prove (3.8), letβ =∑iciαi. There is an index j6=ksuch thatcj>0, whereas insαkβ the coefficients of the simple roots other thanαkare the same as those ofβ. Henceallthe coefficients ofsαkβ must be nonnegative.
Proof of (2): Letβ ∈Φ+\∆. We shall prove by induction on ht(β)that β ∈ W0·∆. We can writeβ =∑iciαi withci≥0 andci6=0 for at least two indicesi.
Furthermore, there must exist an indexksuch that(β,αk)>0, since otherwise we would have(β,β) =∑ici(β,αi)≤0,forcingβ=0. We havesαkβ∈Φ+by (3.8), and we claim that
ht(sαkβ)<ht(β). (3.9)
Indeed,sαkβ=β−dkαk, wheredk=2(β,αk)/(αk,αk)>0 by (3.6). Thus insαkβ the coefficient ofαkisck−dk<ck. This proves (3.9)
By inductionsαkβ∈W0·∆, henceβ∈W0·∆. Thus we can writeβ =wαjfor somew∈W0. Hence
−β=w(−αj) =w sαjαj
∈W0∆. This completes the proof thatΦ=W0·∆, which implies (2).
Proof of (3): Letβ∈Φ. Then by (2) there existw∈W0and an indexisuch that β =wαi. Hence forγ∈h∗
Rwe have sβγ =γ− 2(wαi,γ)
(wαi,wαi)wαi=w
w−1γ−2(αi,w−1γ) (αi,αi) αi
= wsαiw−1 γ.
This calculation shows thatsβ =wsαiw−1∈W0, proving (3).
Proof of (4): Letw∈W and supposewΦ+=Φ+. Assume for the sake of con- tradiction thatw6=1. Then by (3) and (3.8) we can writew=s1···sr, wheresj is the reflection relative to a simple rootαijandr≥2. Among such presentations ofw we choose one with the smallest value ofr. We have
s1···sr−1αir=−s1···sr−1srαir=−wαir ∈ −Φ+. Sinceαir∈Φ+, there must exist an index 1≤ j<rsuch that
sjsj+1···sr−1αir∈ −Φ+ and sj+1···sr−1αir ∈Φ+.
(If j=r−1 then the productsj+1···sr−1equals 1.) Hence by (3.8) we know that sj+1···sr−1αir =αij . (3.10) Setw1=sj+1. . .sr−1∈W. Then (3.10) implies thatw1sr(w1)−1=sj(as in the proof of (3)). We now use this relation to writewas a product ofr−2 simple reflections:
w=s1···sj−1 w1sr(w1)−1
w1sr=s1···sj−1sj+1···sr−1
(since(sr)2=1). This contradicts the minimality ofr, and hence we conclude that w=1, proving (4).
Proof of (5): Leth∈h∗
Rbe a regular element. Define ρ=ϖ1+···+ϖl and choosew∈W to maximize(s(h),ρ). We claim that
(w(h),αj)>0 for j=1, . . . ,l. (3.11) To prove this, note thatsαjρ=ρ−αjfor j=1, . . . ,lby (3.7). Thus
(w(h),ρ)≥(sαjw(h),ρ) = (w(h),ρ)−(w(h),αˇj)(αj,ρ).
Hence (w(h),αj)≥0, since (αj,ρ) = (αj,αj)(αˇj,ρ)/2= (αj,αj)/2>0. But (w(h),αj) = (h,w−1αj)6=0, sincehis regular. Thus we have proved (3.11).
In particular, taking h=−h0, where h0 is a regular element definingΦ+ (see Section 2.5.3), we obtain an elementw∈Wsuch that(w(h0),α)<0 for allα∈Φ+. Setw0=w−1. Thenw0Φ+=−Φ+. Ifw1∈W is another element sendingΦ+to
−Φ+thenw0(w1)−1preserves the setΦ+, and hence must be the identity, by (4).
Thusw0is unique. ut
Remark 3.1.10.The proof of Theorem 3.1.9 has introduced new arguments that will be used in the next proposition and in later sections. Forga classical Lie algebra the proof also furnishes a more geometric and less computational explanation for the validity of Lemma 3.1.6.
Definition 3.1.11.LetC={µ∈h∗
R:(µ,αi)≥0 fori=1, . . . ,l}. ThenCis a closed convex cone in the Euclidean space h∗
R that is called the positive Weyl chamber (relative to the choiceΦ+of positive roots).
Ifµ=∑li=1ciϖi, thenci= (µ,αˇi) =2(µ,αi)/(αi,αi). Hence
µ∈C if and only ifci≥0 fori=1, . . . ,l. (3.12) We shall also need thedual cone
3.1 Roots and Weights 137
C∗={λ ∈h∗R :(λ,ϖi)≥0 fori=1, . . . ,l}. Ifλ=∑li=1diαi, thendi=2(λ,ϖi)/(αi,αi). Hence
λ∈C∗ if and only ifdi≥0 fori=1, . . . ,l. (3.13) We now prove that the positive Weyl chamber is afundamental domainfor the action ofW onh∗
R.
Proposition 3.1.12.Ifλ∈h∗
Rthen there existµ∈C and w∈W such that w·λ=µ.
The elementµis unique, and ifλ is regular then w is also unique.
Proof. (For a case-by-case proof of this result for the classical Lie algebras, see Proposition 3.1.20.) We use the dual coneC∗to define a partial order onh∗R: say thatλµifµ−λ∈C∗. This partial order is compatible with addition and multi- plication by positive scalars. For a fixedλthe set
{w·λ :w∈W andλ w·λ}
is nonempty and finite. Letµbe a maximal element in this set (relative to the partial order). Sincesαiµ=µ−(µ,αˇi)αiandµis maximal, the inequality(µ,αˇi)<0 is impossible by (3.13). Thusµ∈C.
To prove uniqueness ofµ, we may assume thatλ,µ∈C\ {0}andw·λ=µfor somew∈W. We will use the same type of argument as in the proof of Theorem 3.1.9 (4). Writew=s1···sr, wheresjis the reflection relative to a simple rootαij. Assumew6=1. Then there exists an indexksuch thatwαk∈ −Φ+(since otherwise w∆⊂Φ+, which would implywΦ+=Φ+, contradicting Theorem 3.1.9 (4)). Thus there exists an index 1≤j≤rsuch that
sjsj+1···srαk∈ −Φ+ and sj+1···srαk∈Φ+.
(Ifj=rthen the productsj+1···srequals 1.) Hence by (3.8) we havesj+1···srαk= αij. Setw1=sj+1···sr∈W. Thenw1sαk(w1)−1=sj. We can use this relation to writewas
w=s1···sj−1 w1sαk(w1)−1
w1=s1···sj−1sj+1···srsαk.
Hencew1=wsαk=s1···sj−1sj+1···sris a product ofr−1 simple reflections. Since λ,µ∈Candwαk∈ −Φ+, we have
0≥(µ,wαk) = (w−1µ,αk) = (λ,αk)≥0.
Hence(λ,αk) =0; thussαkλ =λ andw1λ =µ. Ifw16=1 we can continue this shortening process until we reach the identity. This proves thatλ =µ and thatw is the product of simple reflections that fixλ. In particular, ifλ is regular, then no
simple reflection fixes it, sow=1 in this case. ut