Bourbaki

§ 7. Ado's theorem

Recall that KK denotes a field of characteristic 0 and that all Lie algebras are assumed to be finite-dimensional over KK.

1. Coefficients of a representation

Let U be an associative algebra with unit element over K, U* the dual of the vector space U and ρ\rho a representation of U on a vector space E. For e∈Ee \in E and e′∈E∗e' \in E^*, let θ(e,e′)∈U∗\theta(e, e') \in U^* be the corresponding coefficient of ρ\rho (Algebra, Chapter VIII, § 13, no. 3). Recall that θ(e,e′)(x)=⟨ρ(x)e,e′⟩\theta(e, e')(x) = \langle \rho(x)e, e' \rangle and that the mapping e↦θ(e,e′)e \mapsto \theta(e, e') is for fixed e′e' a homomorphism of the U-module E into the U-module U* of the coregular representation of U (loc. cit., Proposition 1); therefore the vector subspace of U* generated by the coefficients of ρ\rho (a subspace which we shall denote by C(ρ)C(\rho) in this paragraph) is a sub-U-module of U*. If (ei′)i∈I(e'_i)_{i \in I} is a family of elements generating E* over K, the mapping e↦(θ(e,ei′))e \mapsto (\theta(e, e'_i)) is an injective U-homomorphism of E into C(ρ)IC(\rho)^I; for θ(e,ei′)=0\theta(e, e'_i) = 0 for all ii implies ⟨e,ei′⟩=⟨ρ(1)e,ei′⟩=θ(e,ei′)(1)=0\langle e, e'_i \rangle = \langle \rho(1)e, e'_i \rangle = \theta(e, e'_i)(1) = 0 for all ii and hence e=0e = 0.

In particular, if U is the enveloping algebra of a Lie algebra g and ρ\rho is a representation of g (identified with a representation of U) on an n-dimensional vector space E, the g-module E is isomorphic to a sub-g-module of (C(ρ))n(C(\rho))^n.

2. The extension theorem

Let g=h+g′g = h + g' be a Lie algebra which is the direct sum of an ideal g′g' and a subalgebra hh, U the enveloping algebra of g and U′⊂UU' \subset U the enveloping algebra of g′g'. There exists one and only one g-module structure on U′U' such that: (α) for x∈g′x \in g' and u∈U′u \in U', xU′u=−uxx_{U'}u = -ux; (β) for x∈hx \in h and u∈U′u \in U', xU′u=xu−uxx_{U'}u = xu - ux (the latter element is certainly in U′U' since the inner derivation of U defined by x leaves g′g' and hence U′U' stable). For conditions (α) and (β) define uniquely a linear mapping x↦xU′x \mapsto x_{U'} of g into LK(U′)\mathscr{L}_K(U'). It therefore suffices to verify that [x,y]U′=[xU′,yU′][x, y]_{U'} = [x_{U'}, y_{U'}]; it is only necessary to consider the following cases:

(1) x∈g′,y∈g′x \in g', y \in g': then

[x,y]U′u=−u(xy−yx)=(xU′yU′−yU′xU′)u; [x, y]_{U'}u = -u(xy - yx) = (x_{U'}y_{U'} - y_{U'}x_{U'})u;

(2) x∈h,y∈g′x \in h, y \in g': then

[x,y]U′u=−u(xy−yx)=x(−uy)−(−uy)x+(xu−ux)y=(xU′yU′−yU′xU′)u; [x, y]_{U'}u = -u(xy - yx) = x(-uy) - (-uy)x + (xu - ux)y \\ = (x_{U'}y_{U'} - y_{U'}x_{U'})u;

(3) x∈h,y∈hx \in h, y \in h: then [x,y]U′[x, y]_{U'} and [xU′,yU′][x_{U'}, y_{U'}] are two derivations of U′U' whose restrictions to g′g' coincide with those of adg[x,y]\mathrm{ad}_g[x, y] and [adgx,adgy][\mathrm{ad}_g x, \mathrm{ad}_g y]; hence these derivations are equal.

We shall also consider the dual representation x↦txU′x \mapsto {}^t x_{U'} of g on U′∗{U'}^*. For x∈g′x \in g', txU′{}^t x_{U'} is the transpose of right multiplication by x in U′U'; the corresponding representation of U′U' is therefore the coregular representation of U′U'.

Definition 1

Let gg be a Lie algebra, g′g' a subalgebra of gg and ρ′\rho' a representation of g′g' on V′V'. A representation ρ\rho of gg on VV is called an extension of ρ′\rho' to gg if there exists an injective homomorphism of the g′g'-module into the g′g'-module VV. We also say that the gg-module VV is an extension of the g′g'-module V′V'.

If ρ′\rho' is finite-dimensional and g′g' is a solvable ideal of gg, it is necessary for the existence of a finite-dimensional extension that [g,g′][g, g'] be contained in the largest nilpotency ideal of ρ′\rho' (§ 5, no. 3, Theorem 1).

Theorem 1 (Zassenhaus)

Let g=g′+hg = g' + h be a Lie algebra which is the direct sum of an ideal g′g' and a subalgebra hh and ρ′\rho' a finite-dimensional representation of g′g' whose largest nilpotency ideal contains [h,g′][h, g'].

(a) There exists a finite-dimensional extension ρ\rho of ρ′\rho' to gg whose largest nilpotency ideal contains that of ρ′\rho'.

(b) If for all x∈hx \in h the restriction to g′g' of ad⁡gx\operatorname{ad}_gx is nilpotent, ρ\rho can be chosen such that moreover the largest nilpotency ideal of ρ\rho contains hh.

Let U′U' be the enveloping algebra of g′g'. Suppose that U′U' and U′∗{U'}^* have the gg-module structures defined at the beginning of this no.

U′∗Let I⊂U′ be the kernel of ρ′ (identified with a representation of U′). It is a two-sided ideal of U′ of finite codimension. The subspace C(ρ′) of U′∗ (cf. no. 1) is orthogonal to I. Let S be the sub-g-module of U′∗ generated by C(ρ′).C(ρ′) \begin{array}{ll} {U'}^* & \text{Let } I \subset U' \text{ be the kernel of } \rho' \text{ (identified with a representation of } U'). \text{ It is a two-sided ideal of } U' \text{ of finite codimension. The subspace } C(\rho') \text{ of } {U'}^* \text{ (cf. no. 1) is orthogonal to } I. \text{ Let } S \text{ be the sub-} g \text{-module of } {U'}^* \text{ generated by } C(\rho'). \\ C(\rho') \end{array}

We now show that SS is finite-dimensional over KK. Let V′V' be the space on which ρ′\rho' operates and V′=V0′⊃V1′⊃⋯⊃Vd′={0}V' = V'_0 \supset V'_1 \supset \cdots \supset V'_d = \{0\} a Jordan-Hölder series of the g′g'-module V′V'. Let ρi′\rho'_i be the representation of g′g' on Vi−1′/Vi′V'_{i-1}/V'_i derived from ρ′\rho' (1≤i≤d1 \leq i \leq d). Let I′⊂U′I' \subset U' be the intersection of the kernels of the ρi′\rho'_i (identified with representations of U′U'). Then

I′d⊂I⊂I′ {I'}^d \subset I \subset I'

and I′∩g′I' \cap g' is the largest nilpotency ideal of ρ′\rho'. By § 2, no. 6, Corollary to Proposition 6, I′d{I'}^d is of finite codimension in U′U'. If x∈hx \in h, the derivation u↦xu−uxu \mapsto xu - ux of U′U' maps g′g' into [h,g′]⊂I′[h, g'] \subset I', hence U′U' into I′I' and hence I′d{I'}^d into I′d{I'}^d. On the other hand, clearly I′d{I'}^d is a sub-g′g'-module of U′U'. Hence I′d{I'}^d is a sub-gg-module of U′U'. The orthogonal of I′d{I'}^d in U′∗{U'}^* is a finite-dimensional sub-gg-module which contains C(ρ′)C(\rho') and therefore SS. This shows that SS is finite-dimensional over KK. For x∈I′∩g′x \in I' \cap g', xdx^d is obviously contained in the annihilator of the gg-module U′/I′dU'/{I'}^d and hence also in the annihilator of the gg-module SS.

We saw in no. 1 that the g′g'-module V′V' is isomorphic to a sub-g′g'-module of a product (C(ρ′))n(C(\rho'))^n. Hence the gg-module SnS^n provides a finite-dimensional extension ρ\rho of ρ′\rho' to gg. Moreover, ρ(x)\rho(x) is nilpotent for x∈I′∩g′x \in I' \cap g'; as I′∩g′I' \cap g' is an ideal of gg (for it contains [h,g′][h, g'] by hypothesis), we see that I′∩g′I' \cap g' is contained in the largest nilpotency ideal of ρ\rho. Thus (a) is proved.

Suppose finally that for all x∈hx \in h the restriction to g′g' of adψx\mathrm{ad}_{\psi}x is nilpotent. As the elements of hh operate on U′U' by derivations, there exists for all u∈U′u \in U' and all x∈hx \in h an integer ee such that (xU′)e.u=0(x_{U'})^e.u = 0; the endomorphisms derived from xU′x_{U'} on U′/I′dU'/{I'}^d and on SS (which are finite-dimensional spaces) are therefore nilpotent. Thus ρ(x)\rho(x) is nilpotent for all x∈hx \in h. We have seen earlier that this is also true for x∈I′∩g′x \in I' \cap g'. As I′∩g′I' \cap g' is an ideal of g′g' containing [h,g′][h, g'], the sum h+(I′∩g′)h + (I' \cap g') is also an ideal of gg. Assertion (b) of Theorem 1 then results from the following lemma:

Lemma 1

Let g=g′+hg = g' + h be a Lie algebra which is the sum of an ideal g′g' and a subalgebra hh. Let σ\sigma be a finite-dimensional representation of gg. Suppose that σ(x)\sigma(x) is nilpotent for all x∈g′x \in g' and all x∈hx \in h. Then σ(x)\sigma(x) is nilpotent for all x∈gx \in g.

Passing to the quotient by the kernel of σ\sigma, σ\sigma may be assumed to be faithful. Then g′g' and hh are nilpotent and hence gg, which is an extension of a quotient of hh by g′g', is solvable. Then hh and g′g' are contained in the largest nilpotency ideal of σ\sigma (§ 5, no. 3, Corollary 6 to Theorem 1).

For an improvement of Theorem 1, cf. Exercise 4.

3. Ado's theorem

Proposition 1

Let gg be a Lie algebra, nn its largest nilpotent ideal, aa a nilpotent ideal of gg and ρ\rho a finite-dimensional representation of aa such that every element of ρ(a)\rho(a) is nilpotent. Then ρ\rho admits a finite-dimensional extension σ\sigma to gg such that every element of σ(n)\sigma(n) is nilpotent.

Let a=n0⊂n1⊂⋯⊂np=na = n_0 \subset n_1 \subset \cdots \subset n_p = n be a sequence of subalgebras of nn such that ni−1n_{i-1} is an ideal of nin_i of codimension 1 for 1≤i≤p1 \leq i \leq p (§ 4, no. 1, Proposition 1 (e)). The algebra nin_i is therefore the direct sum of ni−1n_{i-1} and a 1-dimensional subalgebra. As adnx\mathrm{ad}_{n}x is nilpotent for all x∈nx \in n, it is possible (Theorem 1) to find one by one finite-dimensional extensions ρ1,ρ2,…,ρp=ρ′\rho_1, \rho_2, \ldots, \rho_p = \rho' of ρ\rho to n1,n2,…,np=nn_1, n_2, \ldots, n_p = n such that every element of ρ′(n)\rho'(n) is nilpotent.

Let rr be the radical of gg and let n=r0⊂r1⊂⋯⊂rq=rn = r_0 \subset r_1 \subset \cdots \subset r_q = r be a sequence of subalgebras of rr such that ri−1r_{i-1} is an ideal of rir_i of codimension 1 for 1≤i≤q1 \leq i \leq q (§ 5, no. 1, Proposition 2 (d)). The algebra rir_i is thus the direct sum of ri−1r_{i-1} and a 1-dimensional subalgebra. As [r,r]⊂n[r, r] \subset n, it is possible (Theorem 1) to find one by one finite-dimensional extensions ρ1′,ρ2′,…,ρq′=ρ′′\rho'_1, \rho'_2, \ldots, \rho'_q = \rho'' of ρ′\rho' to r1,r2,…,rq=rr_1, r_2, \ldots, r_q = r such that every element of ρ′′(n)\rho''(n) is nilpotent.

Finally gg is the direct sum of rr and a subalgebra ss (§ 6, no. 8, Theorem 5). As [s,r]⊂n[s, r] \subset n, it is possible (Theorem 1) to find a finite-dimensional extension σ\sigma of ρ′′\rho'' to gg such that every element of σ(n)\sigma(n) is nilpotent.

Theorem 2

Every Lie algebra has a finite-dimensional faithful linear representation.

More precisely:

Theorem 3 (Ado)

Let gg be a Lie algebra and nn its largest nilpotent ideal. There exists a finite-dimensional faithful representation φ\varphi of gg such that every element of φ(n)\varphi(n) is nilpotent.

The 1-dimensional Lie algebra KK admits finite-dimensional faithful representations τ\tau such that every element of τ(K)\tau(K) is nilpotent, for example the representation

λ↦(00λ0). \lambda \mapsto \begin{pmatrix} 0 & 0 \\ \lambda & 0 \end{pmatrix}.

It is easily deduced that the centre cc of gg, which is a product of 1-dimensional algebras, admits a finite-dimensional faithful representation σ\sigma such that every element of σ(c)\sigma(c) is nilpotent. Let σ1\sigma_1 be a finite-dimensional extension of σ\sigma to gg such that every element of σ1(n)\sigma_1(n) is nilpotent (Proposition 1); if kk denotes the kernel of σ1\sigma_1, then k∩c={0}k \cap c = \{0\}. On the other hand, let σ2\sigma_2 be the adjoint representation of gg, whose kernel is cc; every element of σ2(n)\sigma_2(n) is nilpotent. The direct sum ρ\rho of σ1\sigma_1 and σ2\sigma_2 is finite-dimensional; every element of ρ(n)\rho(n) is nilpotent; and the kernel of ρ\rho, contained in kk and in cc, is zero, so that ρ\rho is faithful.

Exercises

The conventions of § 7 remain valid unless otherwise mentioned.

See the exercises for § 7.

Transcribed from the printed volume. Not translated.