Bourbaki

§ 3. Enveloping algebra of the free Lie algebra

In this paragraph, A(X)=AK(X)A(X) = A_K(X) denotes the free associative algebra Libas(X) of the set XX over the ring KK (Algebra, Chapter III, § 2, no. 7, Definition 2). XX is identified with its canonical image in A(X)A(X); recall that the KK-module A(X)A(X) admits as basis the free monoid Mo(X)Mo(X) derived from XX; A+(X)A^+(X) denotes the submodule of A(X)A(X) generated by the non-empty words.

1. Enveloping algebra of L(X)L(X)

Theorem 1

Let* α:L(X)→A(X)\alpha : L(X) \to A(X) be the unique Lie algebra homomorphism extending the canonical injection of XX into A(X)A(X) (§ 2, no. 2, Proposition 1). Let σ:L(X)→U(L(X))\sigma : L(X) \to U(L(X)) be the canonical mapping of L(X)L(X) into its enveloping algebra and let β:U(L(X))→A(X)\beta : U(L(X)) \to A(X) be the unique unital algebra homomorphism such that β∘σ=α\beta \circ \sigma = \alpha (Chapter I, § 2, no. 1, Proposition 1). Then: (a) α\alpha is injective and α(L(X))\alpha(L(X)) is a direct factor submodule of A(X)A(X). (b) β\beta is bijective.

Cites 4 · Cited by 4

Cites Proposition 1, I § 2, Proposition 1, II § 2, Corollary 3, I § 2, Corollary, II § 2

Cited by Remark, Theorem 2, II § 6. The Hausdorff series, Proposition 3, VIII § 4

Let B be a unital K-algebra and ϕ\phi a mapping of X into B; by Proposition 1 of § 2, no. 2, there exists a Lie algebra homomorphism ψ:L(X)→B\psi : L(X) \to B such that ψ∣X=ϕ\psi | X = \phi; by Proposition 1 of Chapter I, § 2, no. 1, there exists a unital algebra homomorphism θ:U(L(X))→B\theta : U(L(X)) \to B such that θ∘σ=ψ\theta \circ \sigma = \psi and hence such that (θ∘σ)∣X=ϕ(\theta \circ \sigma) | X = \phi. As σ(X)\sigma(X) generates the unital algebra U(L(X))U(L(X)), the homomorphism θ\theta is the unique unital algebra homomorphism satisfying the latter condition. This shows that the ordered pair (U(L(X)),σ∣X)(U(L(X)), \sigma | X) is a solution of the same universal mapping problem as A(X)A(X); taking ϕ\phi to be the canonical injection of X into A(X)A(X), we deduce that β\beta is an isomorphism, which proves (b).

Finally, as L(X)L(X) is a free K-module (§ 2, no. 11, Corollary to Theorem 1), σ\sigma is injective and σ(L(X))\sigma(L(X)) is a direct factor submodule of U(L(X))U(L(X)) (Chapter I, § 2, no. 7, Corollary 3 to Theorem 1). By (b), this proves (a).

Corollary 1

There exists on the algebra A(X)A(X) a unique coproduct making A(X)A(X) into a bigebra such that the elements of X are primitive. Further, β\beta is an isomorphism of the bigebra U(L(X))U(L(X)) onto A(X)A(X) with this bigebra structure.

Cited by 2

Cited by Corollary 2, Theorem 1, II § 6

This follows from assertion (b) of the theorem and the fact that X generates the unital algebra A(X)A(X).

Henceforth A(X)A(X) is given this bigebra structure and L(X)L(X) is identified with its image under α\alpha, that is with the Lie subalgebra of A(X)A(X) generated by X.

Corollary 2

If KK is a field of characteristic 0, L(X)L(X) is the Lie algebra of primitive elements of A(X)A(X).

Cites 2 · Cited by 2

Cites Corollary, II § 1, Corollary 1

Cited by Remark, Theorem 1, II § 6

Remark

(1) Let K′K' be a commutative ring containing K. If A(X)A(X), L(X)L(X) and LK′(X)L_{K'}(X) are identified with subsets of AK′(X)A_{K'}(X), we deduce from part (a) of Theorem 1 the relation

L(X)=LK′(X)∩A(X). L(X) = L_{K'}(X) \cap A(X).

(2) Corollary 2 to Theorem 1 remains valid if it only assumes that the additive group of the ring K is torsion-free. For suppose first that K=ZK = \mathbf{Z}; every primitive element of A(X)A(X) is a primitive element of Aq(X)A_q(X) and hence is in Lq(X)∩A(X)=L(X)L_q(X) \cap A(X) = L(X) (Corollary 2 and formula (1)). In the general case, K is flat over Z\mathbf{Z} and we apply Remark 2 of § 1, no. 2 and Proposition 3 of § 2, no. 5.

(3) Let Δ\Delta be a commutative monoid, ϕ0\phi_0 a mapping of XX into Δ\Delta and ϕ:Mo(X)→Δ\phi : Mo(X) \to \Delta the homomorphism of the associated monoid; if A(X)A(X) is given the graduation (Aδ(X))δ∈Δ(A^\delta(X))_{\delta \in \Delta} defined in Algebra, Chapter III, § 3, no. 1, Example 3 and L(X)L(X) the graduation (Lδ(X))δ∈Δ(L^\delta(X))_{\delta \in \Delta} defined in § 2, no. 6, we have immediately, for δ∈Δ\delta \in \Delta, Lδ(X)⊂L(X)∩Aδ(X)L^\delta(X) \subset L(X) \cap A^\delta(X). As LL is the sum of the Lδ(X)L^\delta(X) for δ∈Δ\delta \in \Delta, and the sum of the L(X)∩Aδ(X)L(X) \cap A^\delta(X) for δ∈Δ\delta \in \Delta is direct, this implies

Lδ(X)=L(X)∩Aδ(X). L^\delta(X) = L(X) \cap A^\delta(X).

(4) Let AA be a unital associative algebra and t=(ti)i∈It = (t_i)_{i \in I} a family of elements of AA. We have a diagram

L(I)→ftAi↓↓gtA(I) \begin{array}{ccc} L(I) & \xrightarrow{f_t} & A \\ i \downarrow & & \downarrow g_t \\ A(I) \end{array}

where ii is the canonical injection, ftf_t is the Lie algebra homomorphism defined by tt and gtg_t is the unital algebra homomorphism such that gt(i)=tig_t(i) = t_i for i∈Ii \in I. The diagram is commutative for gt∘ig_t \circ i and ftf_t coincide on II. It follows that if P∈L(I)P \in L(I), the element P((ti)i∈I)P((t_i)_{i \in I}) defined in § 2, no. 4, coincides with the element P((ti)i∈I)P((t_i)_{i \in I}) defined in Algebra, Chapter III, § 2, no. 8, Example 2.

Cites 6

Cites Theorem 1, Proposition 3, II § 2, Corollary 2, A III § 3. Graded algebras, II § 2. Free Lie algebras, A III § 2. Examples of algebras

2. Projector of A+(X)A^+(X) onto L(X)L(X)

Let π\pi be the linear mapping of A+(X)A^+(X) into L(X)L(X) defined by

π(x1…xn)=(ad⁡(x1)∘⋯∘ad⁡(xn−1))(xn) \pi(x_1 \ldots x_n) = (\operatorname{ad}(x_1) \circ \cdots \circ \operatorname{ad}(x_{n-1}))(x_n)

for n>0,x1,…,xnn > 0, x_1, \ldots, x_n in XX.

Proposition 1

(a) The restriction π0\pi_0 of π\pi to L(X)L(X) is a derivation of L(X)L(X). (b) *For every integer n≥1n \geq 1 and all uu in Ln(X)L^n(X), π(u)=n.u\pi(u) = n.u.

Cites 1 · Cited by 1

Cites A III § 10. Derivations

Cited by Corollary

(a) Let EE be the endomorphism algebra of the module L(X)L(X) and θ\theta the homomorphism of A(X)A(X) into EE such that θ(x)=ad⁡x\theta(x) = \operatorname{ad} x for all x∈Xx \in X. The restriction of θ\theta to L(X)L(X) is a Lie algebra homomorphism of L(X)L(X) into EE, which coincides on XX with the adjoint representation of L(X)L(X), whence

θ(u).v=[u,v]for u,v in L(X). \theta(u).v = [u, v] \quad \text{for } u, v \text{ in } L(X).

Let aa be in A(X)A(X) and bb in A+(X)A^+(X); then

π(a.b)=θ(a).π(b). \pi(a.b) = \theta(a).\pi(b).

It suffices to consider the case a=x1…xp, b=xp+1…xp+qa = x_1 \ldots x_p, \ b = x_{p+1} \ldots x_{p+q} with p≥0,q≥1p \geq 0, q \geq 1 and x1,…,xp+qx_1, \ldots, x_{p+q} in XX; but then (5) follows immediately from (3) since θ(x)=ad⁡x\theta(x) = \operatorname{ad} x for x∈Xx \in X.

Let uu and vv be in L(X)L(X); by (4) and (5),

π0([u,v])=π(uv−vu)=θ(u)⋅π(v)−θ(v)⋅π(u)=[u,π(v)]−[v,π(u)]=[u,π0(v)]+[π0(u),v], \pi_0([u, v]) = \pi(uv - vu) = \theta(u) \cdot \pi(v) - \theta(v) \cdot \pi(u) = [u, \pi(v)] - [v, \pi(u)] = [u, \pi_0(v)] + [\pi_0(u), v],

hence π0\pi_0 is a derivation of L(X)L(X).

(b) Let π1\pi_1 be the endomorphism of the module L(X)L(X) which coincides on Ln(X)L^n(X) with multiplication by the integer n≥1n \geq 1. The formula [Ln(X),Lm(X)]⊂Ln+m(X)[L^n(X), L^m(X)] \subset L^{n+m}(X) shows that π1\pi_1 is a derivation (Algebra, Chapter III, § 10, no. 3, Example 6). The derivation π1−π0\pi_1 - \pi_0 of L(X)L(X) is zero on XX and, as XX generates L(X)L(X), π0=π1\pi_0 = \pi_1, whence (b).

Corollary

Suppose that K is a Q-algebra. Let P be the linear mapping of A+(X)A^+(X) into itself such that

Cites 1 · Cited by 1

Cites Proposition 1

Cited by Theorem 2, II § 6

P(x1…xn)=1n(ad⁡x1∘⋯∘ad⁡xn−1)(xn) P(x_1 \ldots x_n) = \frac{1}{n} (\operatorname{ad} x_1 \circ \cdots \circ \operatorname{ad} x_{n-1})(x_n)

for n≥1n \geq 1 and x1,…,xnx_1, \ldots, x_n in X. Then P is a projector of A+(X)A^+(X) onto L(X)L(X).

The image of P is contained in L(X)L(X). Further, for all n≥1n \geq 1 and all uu in Ln(X)L^n(X), P(u)=1nπ(u)P(u) = \frac{1}{n} \pi(u), whence P(u)=uP(u) = u by Proposition 1. As

L(X)=∑n≥1Ln(X), L(X) = \sum_{n \geq 1} L^n(X),

we see that the restriction of P to L(X)L(X) is the identity.

Remark

Suppose that K is a field of characteristic zero and let Q be the projector of A(X)=U(L(X))A(X) = U(L(X)) onto L(X)L(X) associated with the canonical graduation of U(L(X))U(L(X)), cf. § 1, no. 5. For α∈N(X)\alpha \in \mathbf{N}^{(X)}, Q(Aα(X))⊂Lα(X)Q(A^\alpha(X)) \subset L^\alpha(X). For it suffices to verify that the image and the kernel of Q are graded submodules of A(X)A(X) with the graduation of type N(X)\mathbf{N}^{(X)}. This is obvious for the image, which is equal to L(X)L(X). On the other hand, let nn be an integer ≥1\geq 1. The vector subspace of A(X)A(X) generated by the yny^n, where y∈L(X)y \in L(X), is equal to the vector subspace of A(X)A(X) generated by the ∑σ∈Snyσ(1)yσ(2)⋯yσ(n)\sum_{\sigma \in S_n} y_{\sigma(1)} y_{\sigma(2)} \cdots y_{\sigma(n)}, where y1,y2,…,yny_1, y_2, \ldots, y_n are homogeneous elements of L(X)L(X); then this subspace is a graded submodule of A(X)A(X).

(Note that, if Card(X) ≥2\geq 2, the projectors P and Q do not coincide on A+(X)A^+(X). For let x,yx, y be in X with x≠yx \neq y and write

z=x[x,y]+[x,y]x=x2y−yx2. z = x[x, y] + [x, y]x = x^2y - yx^2.

Then Q(z)=0Q(z) = 0 and P(z)=13[x,[x,y]]≠0P(z) = \frac{1}{3}[x, [x, y]] \neq 0, cf. § 2, no. 10, Example and no. 11, Theorem 1.)

Cites 2

Cites II § 1. Enveloping bigebra of a Lie algebra, II § 2. Free Lie algebras

3. Dimension of the homogeneous components of l(x)

Let X be a set, α\alpha an element of N(X)\mathbf{N}^{(X)} and d an integer > 0. We write d∣αd|\alpha if there exists β∈N(X)\beta \in \mathbf{N}^{(X)} such that α=dβ\alpha = d\beta. The element β\beta, which is unique, is then denoted by α/d\alpha/d.

Lemma 1

Let n be an integer > 0, T1,…,TnT_1, \ldots, T_n indeterminates and u1,…,unu_1, \ldots, u_n elements of Z\mathbf{Z}. Let (c(α))α∈Nn−{0}(c(\alpha))_{\alpha \in \mathbf{N}^n - \{0\}} be a family of elements of Z\mathbf{Z} such that

1−∑i=1nuiTi=∏α≠0(1−Tα)c(α). 1 - \sum_{i=1}^n u_i T_i = \prod_{\alpha \neq 0} (1 - T^\alpha)^{c(\alpha)}.
Cites 1 · Cited by 1

Cites A IV § 6. Symmetric functions

Cited by Theorem 2

For all α∈Nn−{0}\alpha \in \mathbf{N}^n - \{0\},

c(α)=1∣α∣∑d∣αμ(d)(∣α∣/d)!(α/d)!uα/d c(\alpha) = \frac{1}{|\alpha|} \sum_{d|\alpha} \mu(d) \frac{(|\alpha|/d)!}{(\alpha/d)!} u^{\alpha/d}

where μ\mu is the Möbius function (Appendix).

Formula (7) is equivalent, on taking logarithms on both sides (Algebra, Chapter IV, § 6, no. 9) to:

log⁡(1−∑i=1nuiTi)=∑α≠0c(α)log⁡(1−Tα). \log \left( 1 - \sum_{i=1}^n u_i T_i \right) = \sum_{\alpha \neq 0} c(\alpha) \log (1 - T^\alpha).

Now

−log⁡(1−∑i=1nuiTi)=∑j≥11j(∑i=1nuiTi)j -\log \left( 1 - \sum_{i=1}^n u_i T_i \right) = \sum_{j \geq 1} \frac{1}{j} \left( \sum_{i=1}^n u_i T_i \right)^j
=∑j≥11j∑∣β∣=j∣β∣!β!uβTβ = \sum_{j \geq 1} \frac{1}{j} \sum_{|\beta|=j} \frac{|\beta|!}{\beta!} u^\beta T^\beta
=∑∣β∣>01∣β∣∣β∣!β!uβTβ. = \sum_{|\beta|>0} \frac{1}{|\beta|} \frac{|\beta|!}{\beta!} u^\beta T^\beta.

On the other hand

−∑α≠0c(α)log⁡(1−Tα)=∑∣α∣>0,k≥11kc(α)Tkα -\sum_{\alpha \neq 0} c(\alpha) \log (1 - T^\alpha) = \sum_{|\alpha|>0, k \geq 1} \frac{1}{k} c(\alpha) T^{k\alpha}
=∑∣β∣>0,k∣β1kc(βk)Tβ. = \sum_{|\beta|>0, k|\beta} \frac{1}{k} c\left( \frac{\beta}{k} \right) T^\beta.

Hence (7) is equivalent to

∑k∣β∣βk∣c(βk)=∣β∣!β!uβfor all β∈Nn−{0}. \sum_{k|\beta} \left| \frac{\beta}{k} \right| c\left( \frac{\beta}{k} \right) = \frac{|\beta|!}{\beta!} u^\beta \quad \text{for all } \beta \in \mathbf{N}^n - \{0\}.

Let Λ\Lambda be the set of (λ1,λ2,…,λn)∈Nn−{0}(\lambda_1, \lambda_2, \ldots, \lambda_n) \in \mathbf{N}^n - \{0\} such that the g.c.d. of λ1,λ2,…,λn\lambda_1, \lambda_2, \ldots, \lambda_n is equal to 1. Every element of Nn−{0}\mathbf{N}^n - \{0\} can be written uniquely in the form mλm\lambda, where mm is an integer ⩾1\geqslant 1 and λ∈Λ\lambda \in \Lambda. Condition (12) is equivalent to

∑k∣m∣mλk∣c(mλk)=(m∣λ∣)!(mλ)!umλfor all λ∈Λ and all m⩾1. \sum_{k|m} \left| \frac{m\lambda}{k} \right| c\left( \frac{m\lambda}{k} \right) = \frac{(m|\lambda|)!}{(m\lambda)!} u^{m\lambda} \quad \text{for all } \lambda \in \Lambda \text{ and all } m \geqslant 1.

By the Möbius inversion formula (Appendix), condition (13) is equivalent to

∣mλ∣c(mλ)=∑d∣mμ(d)∣mλd∣!(mλd)!umλd |m\lambda| c(m\lambda) = \sum_{d|m} \mu(d) \frac{\left| \frac{m\lambda}{d} \right|!}{\left( \frac{m\lambda}{d} \right)!} u^{\frac{m\lambda}{d}}

for all λ∈Λ\lambda \in \Lambda and all m⩾1m \geqslant 1.

Theorem 2

Let XX be a finite set and n=Card(X)n = \mathrm{Card}(X).

Cites 6 · Cited by 1

Cites II § 2. Free Lie algebras, A III § 3. Graded algebras, Corollary, II § 2, Corollary 3, I § 2, Theorem 1, Lemma 1

Cited by Remark

(a) For every integer r⩾1r \geqslant 1, the KK-module Lr(X)L^r(X) is free of rank

c(r)=1r∑d∣rμ(d)nr/d, c(r) = \frac{1}{r} \sum_{d|r} \mu(d) n^{r/d},

where μ\mu is the Möbius function.

(b) For all α∈NX−{0}\alpha \in \mathbf{N}^X - \{0\}, the KK-module Lα(X)L^\alpha(X) (§ 2, no. 6) is free of rank

c(α)=1∣α∣∑d∣αμ(d)(∣α∣/d)!(α/d)!. c(\alpha) = \frac{1}{|\alpha|} \sum_{d|\alpha} \mu(d) \frac{(|\alpha|/d)!}{(\alpha/d)!}.

We already know that the modules Lr(X)L^r(X), where r∈Nr \in \mathbf{N}, and Lα(X)L^\alpha(X), where α∈NX\alpha \in \mathbf{N}^X, are free (§ 2, no. 11, Corollary to Theorem 1). Consider the multi-gradation (Aα(X))α∈NX(A^\alpha(X))_{\alpha \in \mathbf{N}^X} of A(X)A(X) defined by the canonical homomorphism ϕ\phi of Mo(X)Mo(X) into NX\mathbf{N}^X (Algebra, Chapter III, § 3, no. 1, Example 3); then Aα(X)∩L(X)=Lα(X)A^\alpha(X) \cap L(X) = L^\alpha(X) by Remark 3 of no. 1. For α∈NX\alpha \in \mathbf{N}^X, the KK-module Aα(X)A^\alpha(X) admits as basis the set of words in which each letter xx of XX appears α(x)\alpha(x) times. Let d(α)d(\alpha) be the number of these words, that is the rank of Aα(X)A^\alpha(X); we shall calculate in two different ways the formal power series

P((Tx)x∈X)∈Z[[((Tx)x∈X)]] P((T_x)_{x \in X}) \in \mathbf{Z}[[((T_x)_{x \in X})]]

defined by

P(T)=∑α∈NXd(α)Tα. P(T) = \sum_{\alpha \in \mathbf{N}^X} d(\alpha) T^\alpha.

(1) We have

P(T)=∑m∈Mo(X)Tϕ(m)=∑r=0∞∑x1,…,xrTx1⋯Txr=∑r=0∞(∑x∈XTx)r P(T) = \sum_{m \in Mo(X)} T^{\phi(m)} = \sum_{r=0}^\infty \sum_{x_1, \ldots, x_r} T_{x_1} \cdots T_{x_r} = \sum_{r=0}^\infty \left( \sum_{x \in X} T_x \right)^r

whence

P(T)=(1−∑x∈XTx)−1. P(T) = \left( 1 - \sum_{x \in X} T_x \right)^{-1}.

(2) For all α∈NX−{0}\alpha \in \mathbf{N}^X - \{0\}, let (eα,j)1≤j≤c(α)(e_{\alpha,j})_{1 \leq j \leq c(\alpha)} be a basis of Lα(X)L^\alpha(X) and give the set II of ordered pairs (α,j)(\alpha, j) such that α∈NX−{0}\alpha \in \mathbf{N}^X - \{0\} and 1≤j≤c(α)1 \leq j \leq c(\alpha) a total ordering. By Theorem 1 of no. 1 and the Poincaré–Birkhoff–Witt Theorem (Chapter I, § 2, no. 7, Corollary 3 to Theorem 1), the elements

ym=∏(α,j)∈I(eα,j)m(α,j), y_m = \prod_{(\alpha, j) \in I} (e_{\alpha, j})^{m(\alpha, j)},
where the index mm runs through N(I)\mathbf{N}^{(I)}, form a basis of A(X)A(X). Each ymy_m is of multidegree ∑(α,j)∈Im(α,j)α\sum_{(\alpha, j) \in I} m(\alpha, j) \alpha. Let u(m)u(m) denote this multidegree. It follows that
P(T)=∑m∈N(I)Tu(m)=∑m∈N(I)∏(α,j)∈ITm(α,j)α=∏(α,j)∈I∑r=0∞Trα=∏(α,j)∈I(1−Tα)−1, P(T) = \sum_{m \in \mathbf{N}^{(I)}} T^{u(m)} = \sum_{m \in \mathbf{N}^{(I)}} \prod_{(\alpha, j) \in I} T^{m(\alpha, j)\alpha} = \prod_{(\alpha, j) \in I} \sum_{r=0}^\infty T^{r\alpha} = \prod_{(\alpha, j) \in I} (1 - T^\alpha)^{-1},
whence finally
P(T)=∏α∈NX−{0}(1−Tα)−c(α). P(T) = \prod_{\alpha \in \mathbf{N}^X - \{0\}} (1 - T^\alpha)^{-c(\alpha)}.
Comparing (18) and (19), we obtain
1−∑x∈XTx=∏α∈NX−{0}(1−Tα)c(α). 1 - \sum_{x \in X} T_x = \prod_{\alpha \in \mathbf{N}^X - \{0\}} (1 - T^\alpha)^{c(\alpha)}.
Lemma 1 then gives (b).

If we now substitute the same indeterminate UU for the TxT_x for x∈Xx \in X in formula (20), we obtain

1−nU=∏α∈NX−{0}(1−U∣α∣)c(α)=∏r>0(1−Ur)c(r). 1 - nU = \prod_{\alpha \in \mathbf{N}^X - \{0\}} (1 - U^{|\alpha|})^{c(\alpha)} = \prod_{r > 0} (1 - U^r)^{c(r)}.
Applying Lemma 1 again, we deduce (a).

Example

We have

c(1)=n,c(2)=12(n2−n),c(3)=13(n3−n),c(4)=14(n4−n2),c(5)=15(n5−n),c(6)=16(n6−n3−n2+n). \begin{align*} c(1) &= n, & c(2) &= \frac{1}{2}(n^2 - n), & c(3) &= \frac{1}{3}(n^3 - n), \\ c(4) &= \frac{1}{4}(n^4 - n^2), & c(5) &= \frac{1}{5}(n^5 - n), & c(6) &= \frac{1}{6}(n^6 - n^3 - n^2 + n). \end{align*}

Remark

Let XX be a set and let α∈N(X)\alpha \in \mathbf{N}^{(X)}; the rank of the free KK-module Lα(X)L^\alpha(X) is also given by formula (16). This follows immediately from Theorem 2 (b) and Proposition 4 of § 2, no. 6.

Cites 2

Cites Proposition 4, II § 2, Theorem 2

Exercises

The letter XX denotes a set.

See the exercises for § 3.

Transcribed from the printed volume. Not translated.