nLab Lazard elimination

Context

∞\infty-Lie theory

∞-Lie theory (higher geometry)

Background

Smooth structure

Higher groupoids

Lie theory

∞-Lie groupoids

∞-Lie algebroids

Formal Lie groupoids

Cohomology

Homotopy

Related topics

Examples

∞\infty-Lie groupoids

∞\infty-Lie groups

∞\infty-Lie algebroids

∞\infty-Lie algebras

Contents

Idea

Lazard elimination (Lazard 1960, Bourbaki 1972, §II.2.9) is the statement that the free Lie algebra 𝔏(V⊕W)\mathfrak{L}(V \oplus W) on a direct sum of modules is the semidirect product of the free Lie algebra 𝔏(V)\mathfrak{L}(V) with the free Lie algebra on the module T(V)⊗WT(V) \otimes W, where T(V)T(V) denotes the tensor algebra on VV (Prop. below):

𝔏(V⊕W)≃𝔏(V)⋉𝔏(T(V)⊗W). \mathfrak{L}(V \oplus W) \;\simeq\; \mathfrak{L}(V) \ltimes \mathfrak{L}\big(T(V) \otimes W\big) \,.

In particular, the Lie ideal generated by WW is itself a free Lie algebra, freely generated by the iterated brackets [v 1,[v 2,⋯,[v n,w]⋯]]\big[v_1, [v_2, \cdots, [v_n, w] \cdots]\big] for n∈ℕn \in \mathbb{N}, v i∈Vv_i \in V, w∈Ww \in W.

In words: the generators in VV may be “eliminated”, at the cost of trading the generators in WW for all their iterated brackets with elements of VV (typically infinitely many even if VV and WW are of finite rank).

Iterating this elimination is a basic tool for constructing bases of free Lie algebras and for exhibiting free Lie subalgebras (Viennot 1978, Reutenauer 1993).

Conceptually, Lazard elimination is the special case 𝔥=𝔏(V)\mathfrak{h} = \mathfrak{L}(V) of the fact that the free product of any Lie algebra 𝔥\mathfrak{h} with a free Lie algebra 𝔏(W)\mathfrak{L}(W) is the semidirect product 𝔥⋉𝔏(U(𝔥)⊗W)\mathfrak{h} \ltimes \mathfrak{L}\big(U(\mathfrak{h}) \otimes W\big) (Prop. below, cf. Block & Leroux 1972, Lepowsky & Wilson 1995). In this form it has found notable application to Borcherds algebras (Rem. below).

Statement

Throughout, RR denotes a commutative ring, “module” means RR-module, and Lie algebras are over RR (with [x,x]=0[x,x] = 0 for all xx). We write

Given a Lie algebra 𝔥\mathfrak{h} and an 𝔥\mathfrak{h}-module MM, the action of each h∈𝔥h \in \mathfrak{h} on MM extends uniquely to a derivation D hD_h of 𝔏(M)\mathfrak{L}(M) (derivations of a free Lie algebra being freely determined by their values on generators), and h↦D hh \mapsto D_h is a Lie homomorphism 𝔥→Der(𝔏(M))\mathfrak{h} \to Der\big(\mathfrak{L}(M)\big) (since [D h,D h′][D_h, D_{h'}] and D [h,h′]D_{[h,h']} are derivations which agree on generators). This defines the semidirect product Lie algebra 𝔥⋉𝔏(M)\mathfrak{h} \ltimes \mathfrak{L}(M).

Proposition

(free product with a free Lie algebra) For 𝔥\mathfrak{h} a Lie algebra and WW a module, the coproduct (“free product”) of 𝔥\mathfrak{h} with 𝔏(W)\mathfrak{L}(W) is the semidirect product of 𝔥\mathfrak{h} with the free Lie algebra on the free U(𝔥)U(\mathfrak{h})-module on WW:

𝔥⊔𝔏(W)≃𝔥⋉𝔏(U(𝔥)⊗W). \mathfrak{h} \sqcup \mathfrak{L}(W) \;\simeq\; \mathfrak{h} \ltimes \mathfrak{L}\big(U(\mathfrak{h}) \otimes W\big) \,.

Proof

For any Lie algebra 𝔤\mathfrak{g}, expanding out the bracket of the semidirect product shows that Lie homomorphisms 𝔥⋉𝔏(M)→𝔤\mathfrak{h} \ltimes \mathfrak{L}(M) \to \mathfrak{g} are equivalently pairs of Lie homomorphisms ϕ:𝔥→𝔤\phi \colon \mathfrak{h} \to \mathfrak{g} and ψ:𝔏(M)→𝔤\psi \colon \mathfrak{L}(M) \to \mathfrak{g} satisfying

ψ(D h(n))=[ϕ(h),ψ(n)]for allh∈𝔥,n∈𝔏(M). \psi\big(D_h(n)\big) \;=\; \big[\phi(h),\, \psi(n)\big] \;\;\;\;\;\; \text{for all}\; h \in \mathfrak{h},\, n \in \mathfrak{L}(M) \,.

For fixed hh, both sides of this equation are ψ\psi-derivations in nn, so that the locus of nn on which it holds is a Lie subalgebra?. Therefore the condition holds as soon as it holds for all n∈Mn \in M, which means that the restriction ψ| M\psi\vert_M is a homomorphism of 𝔥\mathfrak{h}-modules from MM to 𝔤\mathfrak{g}, the latter regarded as an 𝔥\mathfrak{h}-module via ad∘ϕ\mathrm{ad} \circ \phi. For M=U(𝔥)⊗WM = U(\mathfrak{h}) \otimes W the free U(𝔥)U(\mathfrak{h})-module on WW, such module homomorphisms are in turn equivalent to plain linear maps W→𝔤W \to \mathfrak{g}, hence to Lie homomorphisms 𝔏(W)→𝔤\mathfrak{L}(W) \to \mathfrak{g}. In summary, there are bijections

Hom LieAlg(𝔥⋉𝔏(U(𝔥)⊗W),𝔤) ≃ Hom LieAlg(𝔥,𝔤)×Hom LieAlg(𝔏(W),𝔤) ≃ Hom LieAlg(𝔥⊔𝔏(W),𝔤) \array{ Hom_{LieAlg}\Big( \mathfrak{h} \ltimes \mathfrak{L}\big(U(\mathfrak{h}) \otimes W\big) ,\, \mathfrak{g} \Big) &\simeq& Hom_{LieAlg}\big(\mathfrak{h},\, \mathfrak{g}\big) \times Hom_{LieAlg}\big(\mathfrak{L}(W),\, \mathfrak{g}\big) \\ &\simeq& Hom_{LieAlg}\big( \mathfrak{h} \sqcup \mathfrak{L}(W) ,\, \mathfrak{g} \big) }

natural in 𝔤\mathfrak{g}, and the claim follows by the Yoneda lemma.

Proposition

(Lazard elimination) For V,WV, W modules, there is an isomorphism of Lie algebras

𝔏(V⊕W)≃𝔏(V)⋉𝔏(T(V)⊗W), \mathfrak{L}(V \oplus W) \;\simeq\; \mathfrak{L}(V) \ltimes \mathfrak{L}\big(T(V) \otimes W\big) \,,

where 𝔏(V)\mathfrak{L}(V) acts on T(V)⊗WT(V) \otimes W by left multiplication on the first tensor factor (via 𝔏(V)↪U(𝔏(V))≃T(V)\mathfrak{L}(V) \hookrightarrow U\big(\mathfrak{L}(V)\big) \simeq T(V)), given on generators by

v ↦ (v,0) w ↦ (0,1⊗w) \array{ v &\mapsto& (v,\, 0) \\ w &\mapsto& (0,\, 1 \otimes w) }

with inverse given on generators by

(v,0) ↦ v (0,v 1⊗⋯⊗v n⊗w) ↦ [v 1,[v 2,⋯,[v n,w]⋯]]. \array{ (v,\, 0) &\mapsto& v \\ \big(0,\, v_1 \otimes \cdots \otimes v_n \otimes w\big) &\mapsto& \big[v_1, [v_2, \cdots, [v_n, w] \cdots]\big] \,. }

Proof

Since 𝔏\mathfrak{L} is a left adjoint, it preserves coproducts, 𝔏(V⊕W)≃𝔏(V)⊔𝔏(W)\mathfrak{L}(V \oplus W) \simeq \mathfrak{L}(V) \sqcup \mathfrak{L}(W), so that the isomorphism is the special case of Prop. for 𝔥≔𝔏(V)\mathfrak{h} \coloneqq \mathfrak{L}(V). The formulas follow by tracing identity morphisms through the bijections in the proof there. For instance, the inverse map corresponds to the canonical inclusion 𝔏(V)↪𝔏(V⊕W)\mathfrak{L}(V) \hookrightarrow \mathfrak{L}(V \oplus W) together with the T(V)T(V)-linear extension of W↪𝔏(V⊕W)W \hookrightarrow \mathfrak{L}(V \oplus W), which sends v 1⋯v n⊗wv_1 \cdots v_n \otimes w to ad v 1∘⋯∘ad v n(w)\mathrm{ad}_{v_1} \circ \cdots \circ \mathrm{ad}_{v_n}(w).

Remark

(classical formulation) Under the isomorphism of Prop. , the projection onto 𝔏(V)\mathfrak{L}(V) corresponds to 𝔏(pr V)\mathfrak{L}(pr_V), whose kernel is the Lie ideal ⟨W⟩\langle W \rangle generated by WW. Hence Lazard elimination says equivalently that:

  1. ⟨W⟩⊂𝔏(V⊕W)\langle W \rangle \subset \mathfrak{L}(V \oplus W) is the free Lie algebra on T(V)⊗W≃⨁n∈ℕV ⊗n⊗WT(V) \otimes W \simeq \underset{n \in \mathbb{N}}{\bigoplus} V^{\otimes n} \otimes W, embedded via v 1⊗⋯⊗v n⊗w↦[v 1,[v 2,⋯,[v n,w]⋯]]v_1 \otimes \cdots \otimes v_n \otimes w \mapsto \big[v_1, [v_2, \cdots, [v_n,w]\cdots]\big];

  2. as RR-modules, 𝔏(V⊕W)≃𝔏(V)⊕⟨W⟩\mathfrak{L}(V \oplus W) \,\simeq\, \mathfrak{L}(V) \oplus \langle W \rangle.

For free modules on sets YY and ZZ, this is the form in which the statement is traditionally given (Bourbaki 1972, §II.2.9, Prop. 10):

𝔏(Y⊔Z)=𝔏(Y)⊕𝔏({[y 1,[y 2,⋯,[y n,z]⋯]]|n∈ℕ,y i∈Y,z∈Z}), \mathfrak{L}(Y \sqcup Z) \;=\; \mathfrak{L}(Y) \,\oplus\, \mathfrak{L}\Big( \big\{ [y_1, [y_2, \cdots, [y_n, z] \cdots]] \,\big\vert\, n \in \mathbb{N},\, y_i \in Y,\, z \in Z \big\} \Big) \,,

where the first summand is a Lie subalgebra and the second is the Lie ideal generated by ZZ.

Examples

Example

(eliminating a single generator) For V=RxV = R x free of rank 1, we have 𝔏(Rx)=Rx\mathfrak{L}(R x) = R x (since [x,x]=0[x,x] = 0) and T(Rx)≃R[x]T(R x) \simeq R[x], so that Prop. gives

𝔏(Rx⊕W)≃Rx⋉𝔏(R[x]⊗W), \mathfrak{L}(R x \oplus W) \;\simeq\; R x \,\ltimes\, \mathfrak{L}\big(R[x] \otimes W\big) \,,

where x n⊗wx^n \otimes w corresponds to ad x n(w)=[x,[x,⋯,[x,w]⋯]]\mathrm{ad}^n_x(w) = [x, [x, \cdots, [x, w]\cdots]]. For instance, in the free Lie algebra 𝔏({x,y})\mathfrak{L}\big(\{x,y\}\big) on two generators, the Lie ideal generated by yy is the free Lie algebra on the countably many generators {ad x n(y)} n∈ℕ\big\{\mathrm{ad}^n_x(y)\big\}_{n \in \mathbb{N}}.

Example

(derived subalgebra of the free Lie algebra on two generators) Applying Ex. once more, now eliminating the generator y=ad x 0(y)y = \mathrm{ad}^0_x(y) from {ad x n(y)} n∈ℕ\big\{\mathrm{ad}^n_x(y)\big\}_{n \in \mathbb{N}}, gives a decomposition of RR-modules

𝔏({x,y})≃Rx⊕Ry⊕𝔏(S),S≔{ad y mad x n(y)} m∈ℕ,n≥1. \mathfrak{L}\big(\{x,y\}\big) \;\simeq\; R x \,\oplus\, R y \,\oplus\, \mathfrak{L}(S) \,, \;\;\;\;\; S \;\coloneqq\; \Big\{ \mathrm{ad}^m_y \, \mathrm{ad}^n_x(y) \Big\}_{ m \in \mathbb{N},\, n \geq 1 } \,.

Since the derived subalgebra? [𝔏,𝔏][\mathfrak{L}, \mathfrak{L}] contains SS and has trivial intersection with the degree-1 summand Rx⊕RyR x \oplus R y, it follows that [𝔏,𝔏]=𝔏(S)[\mathfrak{L}, \mathfrak{L}] = \mathfrak{L}(S) is free on SS.

Properties

Remark

(bases of free Lie algebras) For XX a finite set, repeatedly eliminating (Ex. ) a free generator of lowest degree, all remaining free generators eventually have degree exceeding any given bound. Hence the eliminated generators form a basis of 𝔏(X)\mathfrak{L}(X) as an RR-module; in particular 𝔏(X)\mathfrak{L}(X) is a free RR-module. This is the Lazard elimination process. The resulting “Lazard sets” coincide with the Hall sets which index Hall bases, including the Lyndon basis (Viennot 1978, cf. Reutenauer 1993).

Remark

(variants)

  • For universal enveloping algebras over a field, using U(𝔥⋉𝔫)≃U(𝔫)⊗U(𝔥)U(\mathfrak{h} \ltimes \mathfrak{n}) \simeq U(\mathfrak{n}) \otimes U(\mathfrak{h}), Lazard elimination gives T(V⊕W)≃T(T(V)⊗W)⊗T(V)T(V \oplus W) \simeq T\big(T(V) \otimes W\big) \otimes T(V). On bases of words this is the unique factorization (Y⊔Z) *≃(Y *Z) *×Y *(Y \sqcup Z)^\ast \simeq (Y^\ast Z)^\ast \times Y^\ast of words in letters from Y⊔ZY \sqcup Z. The Hopf-algebraic formulation of elimination goes back to Block & Leroux 1972, see Lepowsky & Wilson 1995.

  • The analogous statement for free groups (with the same proof, using GG-sets in place of modules) says that the free product of a group GG with the free group F(Z)F(Z) on a set ZZ is the semidirect product G⋉F(G×Z)G \ltimes F(G \times Z). For instance, the normal subgroup of F(Y⊔Z)F(Y \sqcup Z) generated by ZZ is free on the conjugates gzg −1g z g^{-1}, for g∈F(Y)g \in F(Y) and z∈Zz \in Z.

  • There are versions for restricted Lie algebras (Stöhr 2001), for free partially commutative Lie algebras (Duchamp & Krob 1993), and for quotients of 𝔏(V⊕W)\mathfrak{L}(V \oplus W) by suitable ideals (Jurisich & Wilson 2004).

  • The proofs above apply verbatim to graded Lie algebras over ℚ\mathbb{Q} (with Koszul signs).

Remark

(relation to the Hilton-Milnor theorem) For connected pointed CW-complexes XX, YY, the first step of the Hilton-Milnor theorem? is the weak homotopy equivalence

ΩΣ(X∨Y)≃ΩΣX×ΩΣ(⋁n∈ℕX ∧n∧Y). \Omega \Sigma (X \vee Y) \;\simeq\; \Omega \Sigma X \,\times\, \Omega \Sigma \Big( \underset{n \in \mathbb{N}}{\bigvee} X^{\wedge n} \wedge Y \Big) \,.

The rational homotopy groups π •(ΩΣZ)⊗ℚ\pi_\bullet(\Omega \Sigma Z) \otimes \mathbb{Q} form, under the Samelson product, the free graded Lie algebra on H˜ •(Z;ℚ)\widetilde{H}_\bullet(Z;\mathbb{Q}). Moreover, H˜ •(⋁ nX ∧n∧Y;ℚ)≃T(V)⊗W\widetilde{H}_\bullet\big(\bigvee_n X^{\wedge n} \wedge Y; \mathbb{Q}\big) \simeq T(V) \otimes W for V≔H˜ •(X;ℚ)V \coloneqq \widetilde H_\bullet(X;\mathbb{Q}) and W≔H˜ •(Y;ℚ)W \coloneqq \widetilde H_\bullet(Y;\mathbb{Q}). Hence this equivalence is a space-level counterpart of the underlying module decomposition 𝔏(V⊕W)≃𝔏(V)⊕𝔏(T(V)⊗W)\mathfrak{L}(V \oplus W) \simeq \mathfrak{L}(V) \oplus \mathfrak{L}\big(T(V) \otimes W\big) of graded Lazard elimination. Iterating it yields the full Hilton-Milnor theorem, with factors indexed by a Hall basis of the free Lie algebra on two generators (cf. Rem. ).

Applications

Remark

(to Borcherds algebras) Using Lazard elimination, Jurisich 1998 showed that every Borcherds algebra 𝔤\mathfrak{g} without mutually orthogonal imaginary simple roots decomposes as

𝔤=𝔲 +⊕(𝔤 J+𝔥)⊕𝔲 −, \mathfrak{g} \;=\; \mathfrak{u}^+ \oplus (\mathfrak{g}_J + \mathfrak{h}) \oplus \mathfrak{u}^- \,,

where 𝔤 J\mathfrak{g}_J is a Kac-Moody algebra and 𝔲 ±\mathfrak{u}^\pm are free Lie algebras on explicit generating modules. For the Monster Lie algebra 𝔪\mathfrak{m} of monstrous moonshine this gives 𝔪=𝔲 +⊕𝔤𝔩 2⊕𝔲 −\mathfrak{m} = \mathfrak{u}^+ \oplus \mathfrak{gl}_2 \oplus \mathfrak{u}^-. Seen through semidirect products 𝔥⋉𝔏(M)\mathfrak{h} \ltimes \mathfrak{L}(M) as in Prop. , with 𝔤𝔩 2\mathfrak{gl}_2 in the role of 𝔥\mathfrak{h}, this becomes transparent (Lepowsky & Wilson 1995). The free Lie algebra structure of 𝔲 −\mathfrak{u}^- serves to simplify part of Borcherds‘ proof of the Conway-Norton conjectures (Jurisich, Lepowsky & Wilson 1994).

References

The elimination method goes back to Lazard’s work on nilpotent groups and Lie rings:

Textbook accounts:

  • Nicolas Bourbaki, §2.9 in: Algèbres de Lie libres, Chap. II of: Groupes et algèbres de Lie, Hermann (1972); English translation: Free Lie Algebras, Chap. II of: Lie groups and Lie algebras – Chapters 1-3, Springer (1975, 1989) [ISBN:9783540642428]

  • Christophe Reutenauer: Free Lie algebras, London Mathematical Society Monographs 7, Oxford University Press (1993) [ISBN:9780198536796]

On Lazard sets and their relation to Hall sets:

  • Gérard Viennot?: Algèbres de Lie libres et monoïdes libres, Lecture Notes in Mathematics 691, Springer (1978)

Hopf-algebraic formulation:

  • Richard E. Block?, Pierre Leroux?: Free Hopf module algebras and related functors, unpublished preprint (1972)

  • James Lepowsky, Robert L. Wilson: On Hopf algebras and the elimination theorem for free Lie algebras, in: Moonshine, the Monster and Related Topics, Contemporary Mathematics, AMS [arXiv:q-alg/9505022]

Variants and generalizations:

Application to Borcherds algebras and the Monster Lie algebra:

Created on November 8, 2026 at 19:22:27. See the history of this page for a list of all contributions to it.