Bourbaki

§ 1. Primary decomposition of linear representations

1. Decomposition of a family of endomorphisms

Let V be a vector space, S a set, and rr a map from S to End(V). Denote by P the set of maps from S to kk. If λ∈P\lambda \in P, denote by Vλ(S)V_{\lambda}(S) (resp. Vλ(S)V^{\lambda}(S)) the set of v∈Vv\in V such that, for all s∈S,r(s)v=λ(s)vs\in S,r(s)v=\lambda (s)v (resp. (r(s)−λ(s))nv=0(r(s)-\lambda (s))^nv= 0 for nn sufficiently large). The sets Vλ(S)V_{\lambda}(S) and Vλ(S)V^{\lambda}(S) are vector subspaces of V, and Vλ(S)⊂Vλ(S)V_{\lambda}(S)\subset V^{\lambda}(S). We say that Vλ(S)V_{\lambda}(S) is the eigenspace of V relative to λ\lambda (and to rr), that Vλ(S)V^{\lambda}(S) is the primary subspace of V relative to λ\lambda (and to rr), and that V0(S)V^0(S) is the nilspace of V (relative to rr). We say that λ\lambda is a weight of S in V if Vλ(S)≠0V^{\lambda}(S)\not= 0.

In particular, if S reduces to a single element s,Ps, P can be identified with kk; we use the notations Vλ(s)(s)V_{\lambda(s)}(s) and Vλ(s)(s)V^{\lambda(s)}(s), or Vλ(s)(r(s))V_{\lambda(s)}(r(s)) and Vλ(s)(r(s))V^{\lambda(s)}(r(s)), instead of Vλ({s}),Vλ({s})V_{\lambda}(\{s\}), V^{\lambda}(\{s\}); we speak of eigenspaces, primary subspaces and the nilspace of r(s)r(s); an element vv of Vλ(s)(s)V_{\lambda(s)}(s) is called an eigenvector of r(s)r(s), and, if v≠0,λ(s)v\not= 0,\lambda (s) is called the corresponding eigenvalue (cf. Algebra, Chap. VII, §5).

For all λ∈P\lambda \in P, the following relations are immediate:

Vλ(S)=⋂s∈SVλ(s)(s)(1) V^{\lambda}(S) =\bigcap_{s\in S}V^{\lambda(s)}(s) \tag{1}
Vλ(S)=⋂s∈SVλ(s)(s)(2) V_{\lambda}(S) =\bigcap_{s\in S}V_{\lambda(s)}(s) \tag{2}

Let k′k' be an extension of kk. The canonical map from End(V) to End(V⊗kk′V\otimes_kk') gives, by composition with rr, a map r′:S→r': S\rightarrow End(V⊗kk′V\otimes_kk'). Similarly, every map λ\lambda from S to kk defines canonically a map, also denoted by λ\lambda, from S to k′k'. With these notations, we have the following proposition:

Proposition 1 00SJ

For all λ∈P\lambda \in P,

(V⊗kk′)λ(S)=Vλ(S)⊗kk′(V\otimes_kk')^{\lambda}(S) = V^{\lambda}(S)\otimes_kk' and (V⊗kk′)λ(S)=Vλ(S)⊗kk′(V\otimes_kk')_{\lambda}(S) = V_{\lambda}(S)\otimes_kk'.

Let (ai)(a_i) be a basis of the kk-vector space k′k'. If v∈V⊗kk′,vv\in V\otimes_kk',v can be expressed uniquely in the form ∑vi⊗ai\sum v_i\otimes a_i where (vi)(v_i) is a finitely-supported family of elements of V. For all s∈Ss\in S,

(r′(s)−λ(s))n(v)=∑(r(s)−λ(s))nvi⊗ai (r'(s)-\lambda (s))^n(v) =\sum(r(s)-\lambda (s))^nv_i\otimes a_i

It follows that

v∈(V⊗kk′)λ(S)⇐⇒vi∈Vλ(S)v\in (V\otimes_kk')^{\lambda}(S)\Leftarrow \Rightarrow v_i\in V^{\lambda}(S) for all ii,

v∈(V⊗kk′)λ(S)⇐⇒vi∈Vλ(S)v\in (V\otimes_kk')_{\lambda}(S)\Leftarrow \Rightarrow v_i\in V_{\lambda}(S) for all ii,

which implies the proposition.

Proposition 2 00SK

Let V,V′,WV,V',W be vector spaces. Let rr : S →\rightarrow End(V), r′:S→r': S\rightarrow End(V′') and q:S→q: S\rightarrow End(W) be maps.

(i) Let f:V→Wf: V\rightarrow W be a linear map such that q(s)f(v)=f(r(s)v)q(s)f(v) =f(r(s)v) for s∈Ss\in S and v∈Vv\in V. Then, for all λ∈P,f\lambda \in P,f maps Vλ(S)V^{\lambda}(S) (resp. Vλ(S)V_{\lambda}(S)) into Wλ(S)W^{\lambda}(S) (resp. Wλ(S)W_{\lambda}(S)).

(ii) Let B:V×V′→WB : V\times V'\rightarrow W be a bilinear map such that

q(s)B(v,v′)=B(r(s)v,v′)+B(v,r′(s)v′) q(s)B(v, v') = B(r(s)v, v') + B(v, r'(s)v')

for s∈S,v∈V,v′∈V′s\in S,v\in V,v'\in V'. Then, for all λ,μ∈P\lambda , \mu\in P, B maps Vλ(S)×V′μ(S)V^{\lambda}(S)\times {V'}^\mu(S) (resp. Vλ(S)×Vμ′(S)V_{\lambda}(S)\times V'_\mu(S)) into Wλ+μ(S)W^{\lambda+\mu}(S) (resp. Wλ+μ(S)W_{\lambda+\mu}(S)).

(iii) Let B:V×V′→WB : V\times V'\rightarrow W be a bilinear map such that

q(s)B(v,v′)=B(r(s)v,r′(s)v′) q(s)B(v, v') = B(r(s)v, r'(s)v')

for s∈S,v∈V,v′∈V′s\in S,v\in V,v'\in V'. Then, for all λ,μ∈P\lambda , \mu\in P, B maps Vλ(S)×V′μ(S)V^{\lambda}(S)\times {V'}^\mu(S) (resp. Vλ(S)×Vμ′(S)V_{\lambda}(S)\times V'_\mu(S)) into Wλμ(S)W^{\lambda \mu}(S) (resp. Wλμ(S)W_{\lambda \mu}(S)).

In case (i), (q(s)−λ(s))nf(v)=f((r(s)−λ(s))nv)(q(s)-\lambda (s))^nf(v) =f((r(s)-\lambda (s))^nv) for s∈Ss\in S and v∈Vv\in V, hence the conclusion. In case (ii),

(q(s)−λ(s)−μ(s))B(v,v′)=B((r(s)−λ(s))v,v′)+B(v,(r′(s)−μ(s))v′) (q(s)-\lambda (s)-\mu(s))B(v, v') = B((r(s)-\lambda (s))v, v') + B(v,(r'(s)-\mu(s))v')

for s∈S,v∈V,v′∈V′s\in S,v\in V,v'\in V', hence by induction on nn

′(n)′′ '(n)''

(q(s)−λ(s)−μ(s))nB(v,v)=∑B((r(s)−λ(s))iv,(r(s)−μ(s))jv)(q(s)-\lambda (s)-\mu(s))^nB(v, v) =\sum B((r(s)-\lambda (s))^iv,(r(s)-\mu(s))^jv).

i i
i+j=ni+j=n

The assertions in (ii) follow immediately. In case (iii), (q(s)−λ(s)μ(s))B(v,v′)=B((r(s)−λ(s))v,r′(s)v′)+B(λ(s)v,(r′(s)−μ(s))v′)(q(s)-\lambda (s)\mu(s))B(v, v') = B((r(s)-\lambda (s))v, r'(s)v')+B(\lambda (s)v,(r'(s)-\mu(s))v') for s∈S,v∈V,v′∈V′s\in S,v\in V,v'\in V', hence by induction on nn

(q(s)−λ(s)μ(s))nB(v,v′) (q(s)-\lambda (s)\mu(s))^nB(v, v')

= ∑(n)B(λ(s)j(r(s)−λ(s))iv,r′(s)i(r′(s)−μ(s))jv′)\sum(n)B(\lambda (s)^j(r(s)-\lambda (s))^iv, r'(s)^i(r'(s)-\mu(s))^jv').

i i
i+j=ni+j=n

The assertions in (iii) follow immediately.

Proposition 3 00SL

The sum ∑λ∈PVλ(S)\sum_{\lambda\in P}V^{\lambda}(S) is direct. The sum ∑λ∈PVλ(S)\sum_{\lambda\in P}V_{\lambda}(S) is direct.

The second assertion is a consequence of the first; hence it suffices to prove that. We distinguish several cases.

a)Sa) S is empty. The assertion is trivial.

b)Sb) S is reduced to a single element ss. Let λ0,λ1,...,λn\lambda_0, \lambda_1, . . . , \lambda_n be distinct elements of kk. For i=0,1,...,ni= 0,1, . . . , n, let vi∈Vλi(s)v_i\in V^{\lambda_i}(s) and assume that v0=v1+⋯+vnv_0=v_1+\cdots +v_n. It suffices to prove that v0v_0 = 0. For i=0,...,ni= 0, . . . , n, there exists an integer qi>q_i> 0 such that (r(s)−λi)qivi(r(s)-\lambda_i)^{q_i}v_i = 0. Consider the polynomials P(X)=∏i≥1(X−λi)qiP(X) =\prod_{i\geq 1}(X-\lambda_i)^{q_i} and Q(X)=(X−λ0)q0Q(X) = (X-\lambda_0)^{q_0}. We have Q(r(s))v0=0Q(r(s))v_0= 0, and

P(r(s))v0=∑i=1nP(r(s))vi=0P(r(s))v_0=\sum_{i=1}^nP(r(s))v_i= 0. Since P and Q are relatively prime, the Bezout identity proves that v0=0v_0= 0.

c)Sc) S is finite and non-empty. We argue by induction on the cardinal of S. Let s∈Ss\in S and S′=SS'= S-- {s}\{s\}. Let (vλ)λ∈P(v_{\lambda})_{\lambda\in P} be a finitely-supported family of elements of V such that ∑λ∈Pvλ=0\sum_{\lambda\in P}v_{\lambda}= 0 and vλ∈Vλ(S)v_{\lambda}\in V^{\lambda}(S). Let λ0∈P\lambda_0\in P. Let P′P' be the set of λ∈P\lambda \in P such that λ∣S′=λ0∣S′\lambda |S'=\lambda_0|S'. By the induction hypothesis applied to S′S', we have ∑λ∈P′vλ=0\sum_{\lambda\in P'}v_{\lambda}= 0. If λ,μ\lambda , \mu are distinct elements of P′,λ(s)≠μ(s)P',\lambda (s)\not=\mu(s).

Since the sum ∑α∈kVα(s)\sum_{\alpha\in k}V^{\alpha}(s) is direct by b), and since vλ∈Vλ(s)(s),vλ=0v_{\lambda}\in V^{\lambda(s)}(s),v_{\lambda}= 0 for all λ∈P′\lambda \in P', and in particular vλ0=0v_{\lambda_0}= 0, which we had to prove.

d) General case. Let (vλ)λ∈P(v_{\lambda})_{\lambda\in P} be a finitely-supported family of elements of V such that ∑λ∈Pvλ=0\sum_{\lambda\in P}v_{\lambda}= 0 and vλ∈Vλ(S)v_{\lambda}\in V^{\lambda}(S). Let P′P' be the finite set of λ∈P\lambda \in P such that vλ≠0v_{\lambda}\not= 0, and let S′S' be a finite subset of S such that the conditions λ∈P′\lambda \in P', μ∈P′,λ∣S′=μ∣S′\mu\in P',\lambda |S'=\mu|S' imply that λ=μ\lambda =\mu. We have vλ∈Vλ∣S′(S′)v_{\lambda}\in V^{\lambda|S'}(S'); applying c), we see that vλ=0v_{\lambda}= 0 for λ∈P′\lambda \in P', which completes the proof.

Recall that, if x∈x\in End(V), we denote by ad xx the map y→xy−yx=[x,y]y \rightarrow xy-yx= [x, y] from End(V) to itself.

Lemma 1 00SM

Let x,y∈x, y\in End(V).

(i) Assume that V is finite dimensional. Then xx is triangularizable if and only if V=∑a∈kVa(x)V =\sum_{a\in k}V^a(x).

(ii) If there exists an integer nn such that (ad xx)ny=0^ny= 0, each Va(x)V^a(x) is stable under yy.

(iii) Assume that V is finite dimensional. If V=∑a∈kVa(x)V =\sum_{a\in k}V^a(x) and if each

Va(x)V^a(x) is stable under yy, there exists an integer nn such that (ad xx)ny=0^ny= 0.

Part (i) follows from Algebra, Chap. VII, §5, no. 2, Prop. 3.

Let E = End(V). Let B be the bilinear map (u,v)→u(v)(u, v) \rightarrow u(v) from E×VE\times V to V. By the definition of ad xx,

x(B(u,v))=B(u,x(v))+x(B(u, v)) =B(u, x(v)) + B((ad xx)(u),v(u), v)

for x∈E,u∈E,v∈Vx\in E,u\in E,v\in V. Let xx operate on E via ad xx. By Prop. 2 (ii), B(E0(x),Va(x))⊂Va(x)B(E^0(x),V^a(x))\subset V^a(x) for all a∈ka\in k. If (ad xx)ny=0^ny= 0, then y∈E0(x)y\in E^0(x), so y(Va(x))⊂Va(x)y(V^a(x))\subset V^a(x), which proves (ii).

To prove (iii), we can replace V by Va(x)V^a(x) and xx (resp. yy) by its restriction to Va(x)V^a(x). Replacing xx by x−ax-a, we can assume that xx is nilpotent. Then, (ad xx)2dimV−1=0^{2 dim V-1}= 0 (Chap. I, §4, no. 2), which proves (iii).

Remark 00SN

The argument proves that, if V is finite dimensional and if there exists an integer nn such that (ad xx)ny=0^ny= 0, then (ad xx)2dimV−1y=0^{2 dim V-1}y= 0.

In the sequel, we shall say that the map r:S→r: S\rightarrow End(V) satisfies condition (AC) (“almost commutative”) if:

(AC) For every pair (s,s′)(s, s') of elements of S, there exists an integer nn such that

(ad r(s)r(s))nr(s′)=0^nr(s') = 0.

Theorem 1 00SO

Assume that V is finite dimensional. The following conditions are equivalent:

(i) Condition (AC) is satisfied and, for all s∈S,r(s)s\in S,r(s) is triangularizable.

(ii) For all λ∈P\lambda \in P, Vλ(S)V^{\lambda}(S) is stable under r(S)r(S), and V=∑λ∈PVλ(S)V =\sum_{\lambda\in P}V^{\lambda}(S).

If V=∑λ∈PVλV =\sum_{\lambda\in P}V^{\lambda}(S), then V=∑a∈kVa(s)V =\sum_{a\in k}V^a(s) for all s∈Ss\in S, and it follows from

Lemma 1 that (ii) implies (i). Assume that condition (i) is satisfied. Lemma 1 and formula (1) imply that each Vλ(S)V^{\lambda}(S) is stable under r(S)r(S). It remains to prove that V=∑λ∈PVλ(S)V =\sum_{\lambda\in P}V^{\lambda}(S). We argue by induction on dim V. We distinguish two cases.

a) For all s∈S,r(s)s\in S,r(s) has a single eigenvalue λ(s)\lambda (s). Then V=Vλ(S)V = V^{\lambda}(S).

b) There exists s∈Ss\in S such that r(s)r(s) has at least two distinct eigenvalues. Then V is the direct sum of the Va(s)V^a(s) for a∈ka\in k, and dim Va(s)<V^a(s)< dim V for all aa. Each Va(s)V^a(s) is stable under rr(S), and it suffices to apply the induction hypothesis.

Corollary 1 00SP

Assume that V is finite dimensional and that condition (AC) is satisfied. Let k′k' be an extension of kk. Assume that, for all s∈Ss\in S, the endomorphism r(s)⊗1r(s)\otimes 1 of V⊗kk′V\otimes_kk' is triangularizable. Let P′P' be the set of maps from S to k′k'. Then V⊗kk′=∑λ′∈P′(V⊗kk′)λ′(S)V\otimes_kk'=\sum_{\lambda'\in P'}(V\otimes_kk')^{\lambda'}(S).

Let r′:S→r': S\rightarrow End(V⊗kk′\otimes_kk') be the map defined by rr. If s1,s2∈Ss_1, s_2\in S, there exists an integer nn such that (ad r(s1)r(s_1))nr(s2)=0^nr(s_2) = 0, hence (ad r′(s1)r'(s_1))nr′(s2)=0^nr'(s_2) = 0. It now suffices to apply Th. 1.

Corollary 2 00SQ

Assume that V is finite dimensional and that condition (AC) is satisfied. Denote by V+(S)V^+(S) the vector subspace ∑s∈S((⋂i≥1r(s)iV)\sum_{s\in S}((\bigcap_{i\geq 1}r(s)^iV). Then:

(i) V0(S)V^0(S) and V+(S)V^+(S) are stable under r(S)r(S);

(ii) V=V0(S)⊕V+(S)V = V^0(S)\oplus V^+(S);

(iii) every vector subspace W of V, stable under r(S)r(S) and such that W0(S)=0W^0(S) = 0, is contained in V+(S)V^+(S);

(iv) ∑s∈Sr(s)V+(S)=V+(S)\sum_{s\in S}r(s)V^+(S) = V^+(S).

Moreover, V+(S)V^+(S) is the only vector subspace of V with properties (i) and (ii). For any extension k′k' of k,(V⊗kk′)+(S)=V+(S)⊗kk′k,(V\otimes_kk')^+(S) = V^+(S)\otimes_kk'.

The last assertion is immediate. Thus, taking Prop. 1 into account, in proving the others we can assume that kk is algebraically closed. By Th. 1, V = ∑λ∈PVλ\sum_{\lambda\in P}V^{\lambda}(S), and the Vλ(S)V^{\lambda}(S) are stable under r(S)r(S). If s∈Ss\in S, the characteristic

polynomial of r(s)∣Vλ(S)r(s)|V^{\lambda}(S) is (X−λ(s))dimVλ(S)(X-\lambda (s))^{dim V^{\lambda}(S)}; it follows that ⋂i≥1r(s)iVλ(s)\bigcap_{i\geq 1}r(s)^iV^{\lambda}(s)

is zero if λ(s)=0\lambda (s) = 0 and is equal to Vλ(S)V^{\lambda}(S) if λ(s)≠0\lambda (s)\not= 0; hence,

V+(S)=∑λ∈P,λ≠0Vλ(S)(3) V^+(S) =\sum_{\lambda\in P,\lambda\not=0}V^{\lambda}(S) \tag{3}

which proves (i), (ii) and (iv). If W is a vector subspace of V stable under rr(S), then W=∑λ∈PWλ(S)W =\sum_{\lambda\in P}W^{\lambda}(S) and Wλ(S)=W∩Vλ(S)W^{\lambda}(S) = W\cap V^{\lambda}(S). If W0(S)=0W^0(S) = 0, we see that W⊂V+W\subset V^+(S), which proves (iii).

Let V′V' be a vector subspace of V stable under r(S)r(S) and such that V′∩V0(S)=0V'\cap V^0(S) = 0. Then V′0(S)=0{V'}^0(S) = 0, so V′⊂V+(S)V'\subset V^+(S) by (iii). If, in addition, V=V0(S)+V′V = V^0(S) + V', we see that V′=V+(S)V'= V^+(S). Q.E.D.

We sometimes call (V0(S),V+(S))(V^0(S),V^+(S)) the Fitting decomposition of V, or of the map r:S→r: S\rightarrow End(V). If S reduces to a single element ss, we write V+(s)V^+(s) or V+(r(s))V^+(r(s)) instead of V+({s})V^+(\{s\}). We have that V=V0(s)⊕V+(s),V0(s)V = V^0(s)\oplus V^+(s), V^0(s) and V+(s)V^+(s) are stable under r(s),r(s)∣V0(s)r(s),r(s)|V^0(s) is nilpotent and r(s)∣V+(s)r(s)|V^+(s) is bijective.

Corollary 3 00SR

Let V and V′V' be finite dimensional vector spaces, and let r:S→r: S\rightarrow End(V) and r′:S→r': S\rightarrow End(V′') be maps satisfying condition (AC). Let f:V→V′f: V\rightarrow V' be a surjective linear map such that f(r(s)v)=r′(s)f(v)f(r(s)v) =r'(s)f(v) for s∈Ss\in S and v∈Vv\in V. Then f(Vλ(S))=V′λ(S)f(V^{\lambda}{(S)) = V'}^{\lambda}(S) for all λ∈P\lambda \in P.

In view of Prop. 1, we are reduced to the case in which kk is algebraically closed. We have V=⨁λ∈PVλV =\bigoplus_{\lambda\in P}V^{\lambda}(S), V′=⨁λ∈PV′λ(S)V'=\bigoplus_{\lambda\in P}{V'}^{\lambda}(S) by Th. 1, and V′=f(V)=V'=f(V) = ∑λ∈Pf(Vλ(S))\sum_{\lambda\in P}f(V^{\lambda}(S)). Finally, f(Vλ(S))⊂V′λ(S)f(V^{\lambda}(S))\subset {V'}^{\lambda}(S) by Prop. 2 (i), hence the corollary.

Proposition 4 00SS

Assume that kk is perfect. Let V be a finite dimensional vector space, uu an element of End(V),us,un,u_s, u_n the semi-simple and nilpotent components of u(u(Algebra, Chap. VII, §5, no. 8).

(i) For all λ∈k,Vλ(u)=Vλ(us)=Vλ(us)\lambda \in k,V^{\lambda}(u) = V^{\lambda}(u_s) = V_{\lambda}(u_s).

(ii) If V has an algebra structure and if uu is a derivation of V,us,u_s and unu_n are derivations of V.

(iii) If V has an algebra structure and if uu is an automorphism of V, then usu_s and 1+us−1un1 +u^{-1}_su_n are automorphisms of V.

In view of Prop. 1, we can assume that kk is algebraically closed, so

V=∑λ∈kVλ(u) V =\sum_{\lambda\in k}V^{\lambda}(u)

The semi-simple component of u∣Vλ(u)u|V^{\lambda}(u) is the homothety with ratio λ\lambda in Vλ(u)V^{\lambda}(u). This proves (i).

Assume from now on that V has an algebra structure. Let x∈Vλ(u)x\in V^{\lambda}(u), y∈Vμ(u)y\in V^\mu(u).

If uu is a derivation of V, then xy∈Vλ+μ(u)xy\in V^{\lambda+\mu}(u) (Prop. 2 (ii)), so

us(xy)=(λ+μ)(xy)=(λx)y+x(μy)=(usx)y+x(usy) u_s(xy) = (\lambda +\mu)(xy) = (\lambda x)y+x(\mu y) = (u_sx)y+x(u_sy)

This proves that usu_s is a derivation of V. Then un=u−usu_n=u-u_s is a derivation of V.

If uu is an automorphism of V, Ker(usu_s) =V0(u)=0= V^0(u) = 0, so usu_s is bijective. On the other hand, xy∈Vλμ(u)xy\in V^{\lambda \mu}(u) (Prop. 2 (iii)), so

us(xy)=(λμ)(xy)=(λx)(μy)=us(x).us(y) u_s(xy) = (\lambda \mu)(xy) = (\lambda x)(\mu y) =u_s(x).u_s(y)

This proves that usu_s is an automorphism of V; but then so is

1+us−1un=us−1u 1 +u^{-1}_su_n=u^{-1}_su

2. The case of a linear family of endomorphisms

Assume now that S has a vector space structure, that the map r:S→r: S\rightarrow End(V) is linear, and that V and S are finite dimensional.

Proposition 5 00ST

Assume that condition (AC) is satisfied, and let λ:S→k\lambda : S\rightarrow k be such that Vλ(S)≠0V^{\lambda}(S)\not= 0. If kk is of characteristic 0, the map λ\lambda is linear. If kk is of characteristic p≠0p\not= 0, there exists a power qq of pp dividing dim Vλ(S)V^{\lambda}(S), and a homogeneous polynomial function P:S→kP : S\rightarrow k of degree qq, such that λ(s)q=P(s)\lambda (s)^q= P(s) for all s∈Ss\in S.

Since Vλ(S)V^{\lambda}(S) is stable under r(S)r(S) (Lemma 1 and formula (1) of no. 1), we can assume that V=Vλ(S)V = V^{\lambda}(S). Let n=n= dim V. Thus, for s∈Ss\in S,

det(X −r(s)-r(s)) =(X−λ(s))n= (X-\lambda (s))^n.

On the other hand, the expansion of the determinant shows that

det(X −r(s)-r(s)) =Xn+a1(s)Xn−1+⋯+ai(s)Xn−i+⋯= X^n+a_1(s)X^{n-1}+\cdots +a_i(s)X^{n-i}+\cdots

where ai:S→ka_i: S\rightarrow k is a homogeneous polynomial function of degree ii. Write n=qmn=qm where qq is a power of the characteristic exponent of kk and (q,m)=1(q, m) = 1. Then (X−λ(s))n=(Xq−λ(s)q)m(X-\lambda (s))^n= (X^q-\lambda (s)^q)^m; hence −mλ(s)q=aq(s)-m\lambda (s)^q=a_q(s), which implies the result.

Proposition 6 00SU

Assume that kk is infinite and that condition (AC) is satisfied. Let k′k' be an extension of kk. Put V′=V⊗kk′,S′=S⊗kk′V'= V\otimes_kk',S'= S\otimes_kk'. Let r′:S′→r': S'\rightarrow End(V′') be the map obtained from rr by extension of scalars. Then

V0(S)⊗kk′=V′0(S)=V′0(S′) V^0(S)\otimes_kk'{= V'}^0{(S) = V'}^0(S')

The first equality follows from Prop. 1. To prove the second, we can assume that V=V0(S)V = V^0(S) and so V′=V′0(S)V'{= V'}^0(S). Let (s1,...,sm)(s_1, . . . , s_m) be a basis of S and (e1,...,en)(e_1, . . . , e_n) a basis of V. There exist polynomials Pij(X1,...,Xm)P_{ij}(X_1, . . . ,X_m) such that

r′(a1s1+⋯+amsm)nej=∑i=1nPij(a1,...,am)ei r'(a_1s_1+\cdots +a_ms_m)^ne_j=\sum_{i=1}^nP_{ij}(a_1, . . . , a_m)e_i

for 1≤j≤n1\leq j\leq n and a1,...,am∈k′a_1, . . . , a_m\in k'. By hypothesis, r′(s)n=0r'(s)^n= 0 for all s∈Ss\in S, in other words Pij(a1,...,am)=0P_{ij}(a_1, . . . , a_m) = 0 for 1≤i,j≤n1\leq i, j\leq n and a1,...,am∈ka_1, . . . , a_m\in k. Since kk is infinite, Pij=0P_{ij}= 0. Consequently, every element of r′(S′)r'(S') is nilpotent and V′=V′0(S′)V'{= V'}^0(S').

Proposition 7 00SV

Assume that kk is infinite and that condition (AC) is satisfied. Let S~\widetilde{S} be the set of s∈Ss\in S such that V0(s)=V0(S)V^0(s) = V^0(S). If s∈Ss\in S, let P(s)P(s) be the determinant of the endomorphism of V/V0(S)V/V^0(S) defined by r(s)r(s) (no. 1, Cor. 2 (i) of Th. 1).

(i) The function s→P(s)s \rightarrow P(s) is polynomial on S. We have S~={s∈S∣P(s)≠0}\widetilde{S} =\{s\in S|P(s)\not= 0\}; this is an open subset of S in the Zariski topology (App. 1).

(ii) S~\widetilde{S} is non-empty, and V+(s)=V+(S)V^+(s) = V^+(S) for all s∈S~s\in \widetilde{S}.

The fact that s→P(s)s \rightarrow P(s) is polynomial follows from the linearity of rr. If s∈S,V0(s)⊃V0s\in S, V^0(s)\supset V^0(S), with equality if and only if r(s)r(s) defines an automorphism of V/V0V/V^0(S), hence (i).

Now let k′k' be an algebraic closure of kk, and introduce V′,S′,r′V',S', r' as in Prop. 6. We remark that S′S' satisfies condition (AC) by continuation of the polynomial identity (ad r(s1)r(s_1))2dimV−1r(s2)=0^{2 dim V-1}r(s_2) = 0 valid for s1,s2∈Ss_1, s_2\in S (no. 1, Remark ). Applying Th.1, we deduce a decomposition

V′=V′0(S′)⊕∑i=1mV′λi(S′) V'{= V'}^0(S')\oplus \sum_{i=1}^m{V'}^{\lambda_i}(S')

with λi≠0\lambda_i\not= 0 for 1≤i≤m1\leq i\leq m. For 1≤i≤m1\leq i\leq m, there exists a polynomial function PiP_i non-zero on S′S' and an integer qiq_i such that λiqi=Pi\lambda^{q_i}_i= P_i (Prop. 5). Since kk is infinite, there exists s∈Ss\in S such that (P1...Pm)(s)≠0(P_1. . .P_m)(s)\not= 0, cf. Algebra, Chap. IV, §2, no. 3, Cor. 2 of Prop. 9. Then λi(s)≠0\lambda_i(s)\not= 0 for all ii, so V′0(S′)=V′0(s){V'}^0(S'{) = V'}^0(s) and consequently V0(S)=V0(s)V^0(S) = V^0(s) (Prop. 6), which shows that S~≠∅\widetilde{S}\not=\emptyset. If s∈S~s\in \widetilde{S}, the fact that V+(S)V^+(S) is stable under r(s)r(s) and is a complement of V0(s)V^0(s) in V implies that V+(S)=V+(s)V^+(S) = V^+(s) (Cor. 2 of Th. 1).

3. Decomposition of representations of a nilpotent Lie algebra

Let h\mathfrak{h} be a Lie algebra and M an h\mathfrak{h}-module. For any map λ\lambda from h\mathfrak{h} to kk, we have defined in no. 1 vector subspaces Mλ(h)M^{\lambda}(\mathfrak{h}) and Mλ(h)M_{\lambda}(\mathfrak{h}) of M. In particular, if g\mathfrak{g} is a Lie algebra containing h\mathfrak{h} as a subalgebra, and if x∈gx\in \mathfrak{g}, we shall often employ the notations gλ(h)\mathfrak{g}^{\lambda}(\mathfrak{h}) and gλ(h)\mathfrak{g}_{\lambda}(\mathfrak{h}); it will then be understood that h\mathfrak{h} operates on g\mathfrak{g} by the adjoint representation adg_{\mathfrak{g}}.

Proposition 8 00SW

Let h\mathfrak{h} be a Lie algebra, and L,M,NhL,M,N\mathfrak{h}-modules. Denote by P the set of maps from h\mathfrak{h} to kk.

(i) The sum ∑λ∈PLλ(P)\sum_{\lambda\in P}L^{\lambda}(P) is direct.

(ii) If f:L→Mf: L\rightarrow M is a homomorphism of h\mathfrak{h}-modules, f(Lλ(h))⊂Mλ(h)f(L^{\lambda}(\mathfrak{h}))\subset M^{\lambda}(\mathfrak{h}) for all λ∈P\lambda \in P.

(iii) If f:L×M→Nf: L\times M\rightarrow N is a bilinear h\mathfrak{h}-invariant map,

f(Lλ(h)×Mμ(h))⊂Nλ+μ(h) f(L^{\lambda}(\mathfrak{h})\times M^\mu(\mathfrak{h}))\subset N^{\lambda+\mu}(\mathfrak{h})

for all λ,μ∈P\lambda , \mu\in P.

This follows from Props. 2 and 3.

Proposition 9 00SX

Let h\mathfrak{h} be a nilpotent Lie algebra and M a finite dimensional h\mathfrak{h}-module. Denote by P the set of maps from h\mathfrak{h} to kk.

(i) Each Mλ(h)M^{\lambda}(\mathfrak{h}) is an h\mathfrak{h}-submodule of M. If xMx_M is triangularizable for all x∈hx\in \mathfrak{h}, then M=∑λ∈PMλ(h)M =\sum_{\lambda\in P}M^{\lambda}(\mathfrak{h}).

(ii) If kk is infinite, there exists x∈hx\in \mathfrak{h} such that M0(x)=M0(h)M^0(x) = M^0(\mathfrak{h}).

(iii) If kk is of characteristic 0, and if λ∈P\lambda \in P is such that Mλ(h)≠0M^{\lambda}(\mathfrak{h})\not= 0, then λ\lambda is a linear form on h\mathfrak{h} vanishing on [h,h][\mathfrak{h},\mathfrak{h}], and Mλ(h)≠0M_{\lambda}(\mathfrak{h})\not= 0.

(iv) If f:M→Nf: M\rightarrow N is a surjective homomorphism of finite dimensional h\mathfrak{h}-modules, then f(Mλ(h))=Nλ(h)f(M^{\lambda}(\mathfrak{h})) = N^{\lambda}(\mathfrak{h}) for all λ∈P\lambda \in P.

(v) If N is a finite dimensional h\mathfrak{h}-module, and B a bilinear form on M×NM\times N invariant under h\mathfrak{h}, then Mλ(h)M^{\lambda}(\mathfrak{h}) and Nμ(h)N^\mu(\mathfrak{h}) are orthogonal relative to B if λ+μ≠0\lambda +\mu\not= 0. Moreover, if B is non-degenerate then so is its restriction to Mλ(h)×N−λ(h)M^{\lambda}(\mathfrak{h})\times N^{-\lambda}(\mathfrak{h}) for all λ∈P\lambda \in P.

Part (i) follows from no. 1, Lemma 1 and Th. 1. Part (ii) follows from no. 2, Prop. 7. Part (iv) follows from no. 1, Cor. 3 of Th. 1. We prove (iii). We can assume that M=Mλ(h)M = M^{\lambda}(\mathfrak{h}). Then, for all x∈h,λ(x)=x\in \mathfrak{h},\lambda (x) = (dim MM)−1^{-1}Tr(xMx_M); this proves that λ\lambda is linear (which also follows from Prop. 5) and that λ\lambda vanishes on [h,h][\mathfrak{h},\mathfrak{h}]. Consider the map ρ:h→\rho :\mathfrak{h}\rightarrow Endk(M)_k(M) defined by

ρ(x)=xM−λ(x)1M \rho (x) =x_M-\lambda (x)1_M

from the above, this is a representation of h\mathfrak{h} on M, and ρ(x)\rho (x) is nilpotent for all x∈hx\in \mathfrak{h}. By Engel’s theorem (Chap. I, §4, no. 2, Th. 1), there exists m≠0m\not= 0 in M such that ρ(x)m=0\rho (x)m= 0 for all x∈hx\in \mathfrak{h}, so m∈Mλ(h)m\in M_{\lambda}(\mathfrak{h}).

The first assertion of (v) follows from no. 1, Prop. 2 (ii). To prove the second, we can assume that kk is algebraically closed in view of Prop. 1 of no. 1; it then follows from the first and the fact that M=∑λMλ(h),N=∑λNλ(h)M =\sum_{\lambda}M^{\lambda}(\mathfrak{h}), N =\sum_{\lambda}N^{\lambda}(\mathfrak{h}), cf. (i).

Remark 00SY

Assume that kk is perfect and of characteristic 2. Let h=sl(2,k)\mathfrak{h}=\mathfrak{s}\mathfrak{l}(2, k), and

2(ab) _2(ab)

let M be the h\mathfrak{h}-module kk (for the identity map of h\mathfrak{h}). If x=x= is an

ca ca

arbitrary element of h\mathfrak{h}, denote by λ(x)\lambda (x) the unique λ∈k\lambda \in k such that λ2=a2+bc\lambda^2=a^2+bc. A calculation shows immediately that M=Mλ(h)M = M^{\lambda}(\mathfrak{h}); on the other hand, Mλ(h)=0M_{\lambda}(\mathfrak{h}) = 0 and λ\lambda is neither linear nor zero on [h,h][\mathfrak{h},\mathfrak{h}], even though h\mathfrak{h} is nilpotent.

Corollary 00SZ

Let h\mathfrak{h} be a nilpotent Lie algebra, and M a finite dimensional h\mathfrak{h}-module such that M0(h)=0M^0(\mathfrak{h}) = 0. Let f:h→Mf:\mathfrak{h}\rightarrow M be a linear map such that

f([x,y])=x.f(y)−y.f(x)f([x, y]) =x.f(y)-y.f(x) for x,y∈hx, y\in \mathfrak{h}.

There exists a∈Ma\in M such that f(x)=x.af(x) =x.a for all x∈hx\in \mathfrak{h}.

Let N=M×kN = M\times k. Make h\mathfrak{h} operate on N by the formula

x.(m,λ)=(x.m−λf(x),0) x.(m, \lambda ) = (x.m-\lambda f(x),0)

The identity satisfied by ff implies that N is an h\mathfrak{h}-module (Chap. I, §1, no. 8, Example 2). The map (m,λ)→λ(m, \lambda ) \rightarrow \lambda from N to kk is a homomorphism from N to the trivial h\mathfrak{h}-module kk. By Prop. 9 (iv), it follows that N0(h)N^0(\mathfrak{h}) contains an element of the form (a,1)(a,1) with a∈Ma\in M. In view of the hypothesis on M,

(M×0)∩N0(h)=0 (M\times 0)\cap N^0(\mathfrak{h}) = 0

so N0(h)N^0(\mathfrak{h}) is of dimension 1 and hence is annihilated by h\mathfrak{h}. Thus, x.a−f(x)=0x.a-f(x) = 0 for all x∈hx\in \mathfrak{h}, which proves the corollary.

Proposition 10 00T0

Let g\mathfrak{g} be a Lie algebra, h\mathfrak{h} a nilpotent subalgebra of g\mathfrak{g}. Denote by P the set of maps from h\mathfrak{h} to kk.

(i) For λ,μ∈P\lambda , \mu\in P, [gλ(h),gμ(h)]⊂gλ+μ(h)[\mathfrak{g}^{\lambda}(\mathfrak{h}),\mathfrak{g}^\mu(\mathfrak{h})]\subset \mathfrak{g}^{\lambda+\mu}(\mathfrak{h}); in particular, g0(h)\mathfrak{g}^0(\mathfrak{h}) is a Lie subalgebra of g\mathfrak{g} containing h\mathfrak{h}, and the gλ(h)\mathfrak{g}^{\lambda}(\mathfrak{h}) are stable under ad g0(h)\mathfrak{g}^0(\mathfrak{h}). Moreover, g0(h)\mathfrak{g}^0(\mathfrak{h}) is its own normalizer in g\mathfrak{g}.

(ii) If M is a g\mathfrak{g}-module, gλ(h)Mμ(h)⊂Mλ+μ(h)\mathfrak{g}^{\lambda}(\mathfrak{h})M^\mu(\mathfrak{h})\subset M^{\lambda+\mu}(\mathfrak{h}) for λ,μ∈P\lambda , \mu\in P; in particular, each Mλ(h)M^{\lambda}(\mathfrak{h}) is a g0(h)\mathfrak{g}^0(\mathfrak{h})-module.

(iii) If B is a bilinear form on g\mathfrak{g} invariant under h,gλ(h)\mathfrak{h},\mathfrak{g}^{\lambda}(\mathfrak{h}) and gμ(h)\mathfrak{g}^\mu(\mathfrak{h}) are orthogonal relative to B for λ+μ≠0\lambda +\mu\not= 0. Assume that B is non-degenerate. Then, for all λ∈P\lambda \in P, the restriction of B to gλ(h)×g−λ(h)\mathfrak{g}^{\lambda}(\mathfrak{h})\times \mathfrak{g}^{-\lambda}(\mathfrak{h}) is non-degenerate; in particular, the restriction of B to g0(h)×g0(h)\mathfrak{g}^0(\mathfrak{h})\times \mathfrak{g}^0(\mathfrak{h}) is non-degenerate.

(iv) Assume that kk is of characteristic 0. Then, if x∈gλ(h)x\in \mathfrak{g}^{\lambda}(\mathfrak{h}) with λ≠0\lambda \not= 0, ad xx is nilpotent.

The map (x,y)→[x,y](x, y) \rightarrow [x, y] from g×g\mathfrak{g}\times \mathfrak{g} to g\mathfrak{g} is g\mathfrak{g}-invariant by the Jacobi identity, hence h\mathfrak{h}-invariant. The first part of (i) thus follows from Prop. 2 (ii). Part (ii) is proved similarly.

If xx belongs to the normalizer of g0(h)\mathfrak{g}^0(\mathfrak{h}) in g\mathfrak{g}, (ad yy).x=−[x,y]∈g0(h).x=-[x, y]\in \mathfrak{g}^0(\mathfrak{h}) for all y∈hy\in \mathfrak{h}, so (ad yy)n.x=0^n.x= 0 for nn sufficiently large. This proves that x∈g0(h)x\in \mathfrak{g}^0(\mathfrak{h}). Assertion (i) is now completely proved.

Assertion (iii) follows from Prop. 9 (v).

To prove (iv), we can assume that kk is algebraically closed. Let x∈gλ(h)x\in \mathfrak{g}^{\lambda}(\mathfrak{h}), with λ≠0\lambda \not= 0. For all μ∈P\mu\in P and any integer n≥0n\geq 0, (ad xx)ngμ(h)⊂gμ+nλ(h)^n\mathfrak{g}^\mu(\mathfrak{h})\subset \mathfrak{g}^{\mu+n\lambda}(\mathfrak{h}); let P1P_1 be the finite set of μ∈P\mu\in P such that gμ(h)≠0\mathfrak{g}^\mu(\mathfrak{h})\not= 0; if kk is of characteristic 0 and λ≠0,(P1+nλ)∩P1=∅\lambda \not= 0, (P_1+n\lambda )\cap P_1=\emptyset for nn sufficiently large, so (ad xx)n=0^n= 0.

Lemma 2 00T1

Assume that kk is of characteristic 0. Let g\mathfrak{g} be a semi-simple Lie algebra over kk, B the Killing form of g,m\mathfrak{g},\mathfrak{m} a subalgebra of g\mathfrak{g}. Assume that the following conditions are satisfied:

1) the restriction of B to m\mathfrak{m} is non-degenerate;

2) if x∈mx\in \mathfrak{m}, the semi-simple and nilpotent components1^1 of xx in g\mathfrak{g} belong to m\mathfrak{m}.

Then m\mathfrak{m} is reductive in g\mathfrak{g} (Chap. I, § 6, no. 6).

By Chap. I, §6, no. 4, Prop. 5d),m5d),\mathfrak{m} is reductive. Let c\mathfrak{c} be the centre of m\mathfrak{m}. If x∈cx\in \mathfrak{c} is nilpotent, then x=0x= 0; indeed, for all y∈my\in \mathfrak{m}, ad xx and ad yy commute, their composition ad x∘x\circ ad yy is nilpotent, and B(x,y)=0B(x, y) = 0, so x=0x= 0. Now let xx be an arbitrary element of c\mathfrak{c}; let ss and nn be its semi-simple and nilpotent components. We have n∈mn\in \mathfrak{m}. Since ad nn is of the form P(ad xx), where P is a polynomial with no constant term, (ad nn).m=0.\mathfrak{m}= 0 and so n∈cn\in \mathfrak{c}, and then n=0n= 0 by the above. Thus ad xx is semi-simple. Consequently, the restriction to m\mathfrak{m} of the adjoint representation of g\mathfrak{g} is semi-simple (Chap. I, §6, no. 5, Th. 4b4b))).

Proposition 11 00T2

Assume that kk is of characteristic 0. Let g\mathfrak{g} be a semi-simple Lie algebra, h\mathfrak{h} a nilpotent subalgebra of g\mathfrak{g}. The algebra g0(h)\mathfrak{g}^0(\mathfrak{h}) satisfies conditions (1) and (2) of Lemma 2 ; it is reductive in g\mathfrak{g}.

1^1 By Chap. I, §6, no. 3, Th. 3, every x∈gx\in \mathfrak{g} can be written uniquely as the sum of a semi-simple element ss and a nilpotent element nn that commute with each other; the element ss (resp. nn) is called the semi-simple (resp. nilpotent) component of xx.

Let x,x′∈g,sx, x'\in \mathfrak{g},s and s′s' their semi-simple components, nn and n′n' their nilpotent components. We have

x′∈g0(x)⇐⇒x'\in \mathfrak{g}^0(x)\Leftarrow \Rightarrow (ad ss)(x′)=0(x') = 0 (Prop.4.4)

⇐⇒\Leftarrow \Rightarrow (ad x′x')(s)=0(s) = 0

=⇒=\Rightarrow (ad s′s')(s)=0(s) = 0

⇐⇒\Leftarrow \Rightarrow (ad ss)(s′)=0(s') = 0

⇐⇒s′∈g0(x)\Leftarrow \Rightarrow s'\in \mathfrak{g}^0(x) (Prop.4.4)

so x′∈g0(x)⇒n′∈g0(x)x'\in \mathfrak{g}^0(x)\Rightarrow n'\in \mathfrak{g}^0(x) and (2) is proved. The Killing form of g\mathfrak{g} is non-degenerate, so its restriction to g0(h)\mathfrak{g}^0(\mathfrak{h}) is non-degenerate (Prop. 10 (iii)). The fact that g0(h)\mathfrak{g}^0(\mathfrak{h}) is reductive in g\mathfrak{g} thus follows from Lemma 2.

4. Decomposition of a Lie algebra relative to an automorphism

Proposition 12 00T3

Let g\mathfrak{g} be a Lie algebra, aa an automorphism of g\mathfrak{g}.

(i) For λ,μ∈k,[gλ(a),gμ(a)]⊂gλμ(a)\lambda , \mu\in k,[\mathfrak{g}^{\lambda}(a),\mathfrak{g}^\mu(a)]\subset \mathfrak{g}^{\lambda \mu}(a); in particular, g1(a)\mathfrak{g}^1(a) is a subalgebra of g\mathfrak{g}.

(ii) If B is a symmetric bilinear form on g\mathfrak{g} invariant under a,gλ(a)a,\mathfrak{g}^{\lambda}(a) and gμ(a)\mathfrak{g}^\mu(a) are orthogonal relative to B for λμ≠1\lambda \mu\not= 1. Assume that B is non-degenerate. Then, if λ≠0\lambda \not= 0, the restriction of B to gλ(a)×g1/λ(a)\mathfrak{g}^{\lambda}(a)\times \mathfrak{g}^{1/\lambda}(a) is non-degenerate.

Assertion (i) and the first half of (ii) follow from Prop. 2 (iii) applied to the composition law g×g→g\mathfrak{g}\times \mathfrak{g}\rightarrow \mathfrak{g} and the bilinear form B. To prove the second half of (ii), we can assume that kk is algebraically closed. Then g=⨁ν∈kgν(a)\mathfrak{g}=\bigoplus_{\nu\in k}\mathfrak{g}^{\nu}(a). In view of the above, gλ(a)\mathfrak{g}^{\lambda}(a) is orthogonal to gν(a)\mathfrak{g}^{\nu}(a) if λν≠1\lambda \nu \not= 1; since B is non-degenerate, it follows that its restriction to gλ(a)×g1/λ(a)\mathfrak{g}^{\lambda}(a)\times \mathfrak{g}^{1/\lambda}(a) is also.

Corollary 00T4

Assume that kk is of characteristic zero and that g\mathfrak{g} is semi-simple. Then the subalgebra g1(a)\mathfrak{g}^1(a) satisfies conditions (1) and (2) of Lemma 2 ; it is reductive in g\mathfrak{g}.

Condition (1) follows from part (ii) of Prop. 12; condition (2) follows from Prop. 4 of no. 1.

5. Invariants of a semi-simple Lie algebra relative to a semi-simple action

In this no., kk is assumed to be of characteristic zero.

Proposition 13 00T5

Let g\mathfrak{g} be a semi-simple Lie algebra, a\mathfrak{a} a subalgebra of g\mathfrak{g} reductive in g\mathfrak{g}, and m\mathfrak{m} the commutant of a\mathfrak{a} in g\mathfrak{g}. The subalgebra m\mathfrak{m} satisfies conditions (1) and (2) of Lemma 2 of no. 3; it is reductive in g\mathfrak{g}.

By Prop. 6 of Chap. I, §3, no. 5, applied to the a\mathfrak{a}-module g\mathfrak{g}, we have g=m⊕[a,g]\mathfrak{g}=\mathfrak{m}\oplus [\mathfrak{a},\mathfrak{g}]. Let B be the Killing form of g\mathfrak{g}, and let x∈a,y∈m,z∈gx\in \mathfrak{a}, y\in \mathfrak{m}, z\in \mathfrak{g}. Then,

B([z,x],y)=B(z,[x,y])=0B([z, x], y) = B(z,[x, y]) = 0 since [x,y]=0[x, y] = 0,

which shows that m\mathfrak{m} is orthogonal to [a,g][\mathfrak{a},\mathfrak{g}] relative to B. Since B is non-degenerate, and since g=m⊕[a,g]\mathfrak{g}=\mathfrak{m}\oplus [\mathfrak{a},\mathfrak{g}], this implies that the restriction of B to m\mathfrak{m} is non-degenerate; condition (1) of Lemma 2 is thus satisfied.

Now let x∈mx\in \mathfrak{m} and let ss and nn be its semi-simple and nilpotent components. The semi-simple component of ad xx is ad ss, cf. Chap. I, §6, no. 3. Since ad xx is zero on a\mathfrak{a}, so is ad ss, by Prop. 4 (i). Thus s∈ms\in \mathfrak{m}, so n=x−s∈mn=x-s\in \mathfrak{m}, and condition (2) of Lemma 2 is satisfied.

Remark 00T6

The commutant of m\mathfrak{m} in g\mathfrak{g} does not necessarily reduce to a\mathfrak{a}, cf. Exerc. 4.

Proposition 14 00T7

Let g\mathfrak{g} be a semi-simple Lie algebra, A a group and rr a homomorphism from A to Aut(g\mathfrak{g}). Let m\mathfrak{m} be the subalgebra of g\mathfrak{g} consisting of the elements invariant under r(A)r(A). Assume that the linear representation rr is semi-simple. Then m\mathfrak{m} satisfies conditions (1) and (2) of Lemma 2 of no. 3; it is reductive in g\mathfrak{g}.

The proof is analogous to that of the preceding proposition:

Let g+\mathfrak{g}^+ be the vector subspace of g\mathfrak{g} generated by the r(a)x−x,a∈A,x∈gr(a)x-x,a\in A, x\in \mathfrak{g}. The vector space g′=m+g+\mathfrak{g}'=\mathfrak{m}+\mathfrak{g}^+ is stable under r(A)r(A). Let n\mathfrak{n} be a complement of g′\mathfrak{g}' in g\mathfrak{g} stable under r(A)r(A). If x∈n,a∈A,r(a)x−x∈n∩g+=0x\in \mathfrak{n}, a\in A,r(a)x-x\in \mathfrak{n}\cap \mathfrak{g}^+= 0, so x∈mx\in \mathfrak{m} and then x=0x= 0 since m∩n=0\mathfrak{m}\cap \mathfrak{n}= 0. Thus, g=g′=m+g+\mathfrak{g}=\mathfrak{g}'=\mathfrak{m}+\mathfrak{g}^+. Let B be the Killing form of g\mathfrak{g} and let y∈m,a∈A,x∈gy\in \mathfrak{m}, a\in A, x\in \mathfrak{g}. Then

B(y,r(a)x−x)=B(y,r(a)x)−B(y,x) B(y, r(a)x-x) = B(y, r(a)x)-B(y, x)
=B(r(a−1)y,x)−B(y,x) = B(r(a^{-1})y, x)-B(y, x)
=B(y,x)−B(y,x)=0 = B(y, x)-B(y, x) = 0

Thus m\mathfrak{m} and g+\mathfrak{g}^+ are orthogonal relative to B. It follows that the restriction of B to m\mathfrak{m} is non-degenerate; hence condition (1) of Lemma 2. Condition (2) is immediate by transport of structure.

Exercises

All Lie algebras and modules over them are assumed to be finite dimensional over kk; from §3 onwards, kk is assumed to be of characteristic zero.

See the exercises for § 1.

Transcribed from the printed volume. Not translated.