Chapter 3 Semisimple and unipotent elements 3.1 Jordan decomposition 3.1.1 Jordan decomposition in GL(V ) Let us first recall some fact on linear algebra. See for example [Bou58] for proofs. Let V be a vector space. Definition 3.1.1 (ı) We call semisimple any endomorphism of V which is diagonalisable. Equiva- lently if dim V is finite, the minimal polynomial is separable. (ıı) We call nilpotent (resp. unipotent) any endomorphism x such that xn = 0 for some n (resp. x − Id is nilpotent). (ııı) We call locally finite any endomorphism x such that for all v ∈ V , the span of {xn(v) /n ∈ N} is of finite dimension. (ııı) We call locally nilpotent (resp. locally unipotent) any endomorphism x such that for all v ∈ V , there exists an n such that xn(v) = 0 (resp. Id − x is locally nilpotent). Fact 3.1.2 Let x and y in gl(V ) such that x and y commute. (ı) If x is semisimple, then it is locally finite. (ıı) If x and y are semisimple, then so are x + y and xy. (ııı) If x and y are locally nilpotent, then so are x + y and xy. (ıv) If x and y are locally unipotent, then so is xy. Theorem 3.1.3 (Additive Jordan decomposition) Let x ∈ gl(V ) be locally finite. (ı) There exists a unique decomposition x = xs + xn in gl(V ) such that xs is semisimple, xn is nilpotent and xs and xn commute. (ıı) There exists polynomial P and Q in k[T ] such that xs = P (x) and xn = Q(x). In particular xs and xn commute with any endomorphism commuting with x. (ııı) If U ⊂ W ⊂ V are subspaces such that x(W ) ⊂ U, then xs and xn also map W in U. (ıv) If x(W ) ⊂ W , then (x|W )s = (xs)|W and (x|W )n = (xn)|W and (x|V/W )s = (xs)|V/W and (x|V/W )n = (xn)|V/W . Definition 3.1.4 The elements xs (resp. xn) is called the semisimple part of x ∈ End(V ) (resp. nilpotent part The decomposition x = xs + xn is called the Jordan-Chevalley decomposition. Corollary 3.1.5 (Multiplicative Jordan decomposition) Let x ∈ gl(V ) be locally finite and in- vertible. 27 28 CHAPTER 3. SEMISIMPLE AND UNIPOTENT ELEMENTS (ı) There exists a unique decomposition x = xsxu in GL(V ) such that xs is semisimple, xu is unipotent and xs and xu commute. (ıı) The elements xs and xu commute with any endomorphism commuting with x. (ııı) If U ⊂ W ⊂ V are subspaces such that x(W ) ⊂ U, then xs and xn also map W in U. (ıv) If x(W ) ⊂ W , then (x|W )s = (xs)|W and (x|W )u = (xu)|W and (x|V/W )s = (xs)|V/W and (x|V/W )u = (xu)|V/W . Proof. We simply have to write x = xs + xn. Because x is inversible, so is xs thus we may set −1 xu = Id+ xs xn which is easily seen to be unipotent and satisfies the above properties. 3.1.2 Jordan decomposition in G Theorem 3.1.6 Let G be an algebraic group and let g be its Lie algebra. 2 (ı) For any g ∈ G, there exists a unique couple (gs, gu) ∈ G such that g = gsgu and ρ(gs)= ρ(g)s and ρ(gu)= ρ(g)u. 2 (ıı) For any η ∈ g, there exists a unique couple (ηs, ηn) ∈ g such that η = ηs + ηn and deG ρ(ηs)= deG ρ(η)s and deG ρ(ηn)= deG ρ(η)n. ′ (ııı) If φ : G → G is a morphism of algebraic groups, then φ(gs) = φ(g)s, φ(gu) = φ(g)u, deG φ(ηs)= deG φ(η)s and deG φ(ηn)= deG φ(η)n. Proof. Let us first note that because ρ and deρ are faithful, the unicity for g ∈ G and η ∈ g follows from the unicity of the Jordan decomposition for ρ(g) and deρ(η). We first prove (ı) and (ıı) for GL(V ). Proposition 3.1.7 Let g ∈ GL(V ) and X ∈ gl(V ). (ı) If g is semisimple, then so is ρ(g). (ıı) If X is semisimple, then so is deρ(X). Therefore, if g = gsgu and X = Xs + Xn are the Jordan decompositions in GL(V ) and gl(V ), then ρ(g) = ρ(gs)ρ(gu) and deρ(X) = deρ(Xs)+ deρ(Xn) are the Jordan decompositions of ρ(g) and deρ(X). Proof. Assume that g or X is semisimple (resp. unipotent or nilpotent), then let (fi) be a base of V such that these endomorphisms are diagonal (resp. upper triangular with 1 or 0 on the diagonal). Recall also that for f ∈ k[G] with ∆(f)= Pi ai ⊗ ui we have ρ(g)f = X aiui(g) and deρ(X)f = X aiX(g). i i Applying this to the elements Ti,j we get ρ(g)Ti,j = X Tk,j(g)Ti,k and deρ(X)Ti,j = X Tk,jX(Ti,k). k k But if g and X are diagonal, then Ti,j(g)= δi,jλi and X(Ti,j)= δi,jλi. We thus get ρ(g)Ti,j = λjTi,j and deρ(X)Ti,j = λjTi,j. Furthermore for det we have ∆(det) = det ⊗ det thus ρ(g) det = det(g)det and deρ(X)det = X(det) det = Tr(X) det . 3.2. SEMISIMPLE, UNIPOTENT AND NILPOTENT ELEMENTS 29 We thus have in this case a base of eigenvectors. If g and X are unipotent of nilpotent, then the same will be true because in the lexicographical order base of the monomials, we also have a triangular matrix whose diagonal coefficients are those of g or Tr(X) = 0. We are therefore left to prove (ııı) to conclude. We deal with to cases which are enough: φ : G → G′ is injective or surjective. Any morphism can be decomposed in such two morphisms by taking the factorisation through the image. Assume that φ is a closed immersion. Then we have k[G′] → k[G] = k[G′]/I. Let g ∈ G resp. ′ ′ η ∈ g and let g = gsgu resp. η = ηs +ηn the Jordan decomposition of g resp. η in G resp. g . We need to prove that these decompositions are in G resp. in g. For this we check that ρ(gs)I = I, ρ(gu)I = I, ′ deρ(ηs)I ⊂ I and deρ(ηn)I ⊂ I. But I is a vector subspace of k[G ] which is stable under g resp. X thus it is also stable under all these maps. This applied to the inclusion of any algebraic group G in some GL(V ) implies the existence of the decomposition. Assume now that φ is surjective. This in particular implies that φ♯ : k[G′] → k[G] is injective. Let g ∈ G resp. η ∈ g and let g = gsgu resp. η = ηs + ηn the Jordan decomposition of g resp. η in G resp. g. We may realise k[G′] as a ρ(G)-submodule of k[G]. For f ∈ k[G′], g ∈ G and g′ ∈ G′, we have: ′ ′ ρ(g)f(g )= f(g φ(g)). We thus have the formula ρ(g)|k[G′] = ρ(φ(g)). Applying this to g, gs and gu, we have ρ(φ(g)) = ρ(φ(gs))ρ(φ(gu)) = ρ(gs)|k[G′]ρ(gu)|k[G′] but as ρ(gs) and ρ(gu) are semisimple and nipotent, so are their restriction thus this is the Jordan decomposition of ρ(φ(g)) and thus of φ(g). ′ The above submodule structure means that we have an action aG′ of G on G whose action is given by ♯ ♯ ′ aG′ = (Id ⊗ φ ) ◦ ∆G . ′ Note that aG′ =∆G|k[G′]. Thus for f ∈ k[G ] we have deρ(deφ(η))f = (Id ⊗ deφ(η)) ◦ ∆G′ (f) ♯ = (id ⊗ η) ◦ (Id ⊗ φ ) ◦ ∆G′ (f) ♯ = (Id ⊗ η)aG′ (f) = (Id ⊗ η)∆G(f) = η · f. Therefore we have deρ(deφ(η)) = deρ(η)|k[G′] and the same argument as above applies. 3.2 Semisimple, unipotent and nilpotent elements Definition 3.2.1 (ı) Let g ∈ G, then g is called semisimple, resp. unipotent if g = gs resp. g = gu. (ıı) Let η ∈ g, then η is called semisimple, resp. nilpotent if η = ηs resp. η = ηn. (ııı) We denote by Gs resp. Gu the set of semisimple, resp. unipotent elements in G. (ıv) We denote by gs resp. gu the set of semisimple, resp. nilpotent elements in g. Fact 3.2.2 If g ∈ G resp. η ∈ g is semisimple and unipotent (resp. semisimple and nilpotent), then g = e (resp. η = 0). 30 CHAPTER 3. SEMISIMPLE AND UNIPOTENT ELEMENTS Remark 3.2.3 Note that in the case of general Lie algebras, the Jordan decomposition does not always exists. This proves that any Lie algebra is not the Lie algebra of an algebraic group. In general, if char(k) = p > 0, for an algebraic group G defined over the field k with Lie algebra λ(G) g = Derk(k[G], k[G]) , we have an additional structure called the p-operation and given by taking the p-th power of the derivation (p-th composition). This maps invariant derivations to invariant derivations. Definition 3.2.4 A p-Lie algebra is a Lie algebra g with a linear map x → x[p] called the p-operation such that • (λx)[p] = λpx[p], • ad (x[p]) = ad (x)p, ′ [p] [p] ′[p] p−1 −1 ′ • (x + x ) = x + x + Pi=1 i si(x,x ) ′ ′ i ′ p−1 ′ where x,x ∈ g, λ ∈ k and si(x,x ) is the coefficient of a in ad (ax + x ) (x ).