Bourbaki

§ 6. The Hausdorff series

In this paragraph we assume that KK is a field of characteristic 0.

1. Exponential and logarithm in filtered algebras

Let AA be a unital associative algebra which is Hausdorff and complete under a real filtration (Aα)(A_\alpha). We write m=A0+=⋃α>0Aαm = A_0^+ = \bigcup_{\alpha > 0} A_\alpha.

For x∈mx \in m, the family (xn/n!)n∈N(x^n/n!)_{n \in \mathbf{N}} is summable. We write

ex=exp⁡x=∑n⩾0xn/n!. e^x = \exp x = \sum_{n \geqslant 0} x^n/n!.

Then exp⁡(x)∈1+m\exp(x) \in 1 + m and the mapping exp⁡:m→1+m\exp : m \to 1 + m is called the exponential mapping of AA.

For all y∈1+my \in 1 + m, the family ((−1)n−1(y−1)n/n)n⩾1((-1)^{n-1}(y-1)^n/n)_{n \geqslant 1} is summable. We write

log⁡y=∑n⩾1(−1)n−1(y−1)n/n. \log y = \sum_{n \geqslant 1} (-1)^{n-1}(y-1)^n/n.

Then log⁡y∈m\log y \in m and the mapping log⁡:1+m→m\log : 1 + m \to m is called the logarithmic mapping of AA.

Proposition 1

The exponential mapping is a homeomorphism of mm onto 1+m1 + m and the logarithmic mapping is the inverse homeomorphism.

For x∈Aαx \in A_\alpha, xnn!∈Anα\frac{x^n}{n!} \in A_{n\alpha}. It follows that the series defining the exponential converges uniformly on each of the sets AαA_\alpha for α>0\alpha > 0; as AαA_\alpha is open in mm and m=⋃α>0Aαm = \bigcup_{\alpha > 0} A_\alpha, the exponential mapping is continuous. It can be similarly shown that the logarithmic mapping is continuous.

Let ee and ll be the formal power series with no constant term

e(X)=∑n≥1Xnn!,l(X)=∑n≥1(−1)n−1Xn/n. e(X) = \sum_{n \geq 1} \frac{X^n}{n!}, \quad l(X) = \sum_{n \geq 1} (-1)^{n-1} X^n / n.

We know (Algebra, Chapter IV, § 6, no. 9) that e(l(X))=l(e(X))=Xe(l(X)) = l(e(X)) = X on A^({X})=K[[X]]\hat{A}(\{X\}) = K[[X]]. By substitution (§ 5, no. 1), we deduce that

e(l(x))=l(e(x))=x e(l(x)) = l(e(x)) = x

for x∈mx \in m; as

exp⁡x=e(x)+1,log⁡(1+x)=l(x) \exp x = e(x) + 1, \quad \log(1 + x) = l(x)

it follows immediately that

log⁡exp⁡x=x,exp⁡log⁡(1+x)=1+x \log \exp x = x, \quad \exp \log(1 + x) = 1 + x

for xx in mm, whence the proposition.

Remark

(1) If x∈m,y∈mx \in m, y \in m and xx and yy commute, then

exp⁡(x+y)=exp⁡(x)exp⁡(y), \exp(x + y) = \exp(x) \exp(y),

since the family (xii!⋅yjj!)i,j∈N\left( \frac{x^i}{i!} \cdot \frac{y^j}{j!} \right)_{i,j \in \mathbf{N}} is summable (cf. Algebra, Chapter IV, § 6, no. 9, Proposition 11).

(2) As the series ee and ll are without constant term and AαA_\alpha is a closed ideal of AA, exp⁡Aα⊂1+Aα\exp A_\alpha \subset 1 + A_\alpha and log⁡(1+Aα)⊂Aα\log(1 + A_\alpha) \subset A_\alpha whence exp⁡Aα=1+Aα\exp A_\alpha = 1 + A_\alpha and log⁡(1+Aα)=Aα\log(1 + A_\alpha) = A_\alpha for α>0\alpha > 0.

(3) Let BB be a complete Hausdorff filtered unital associative algebra and n=⋃α>0Bαn = \bigcup_{\alpha > 0} B_\alpha. Let ff be a continuous unital homomorphism of AA into BB such that f(m)⊂nf(m) \subset n. Then f(exp⁡x)=exp⁡f(x)f(\exp x) = \exp f(x) for x∈mx \in m and f(log⁡y)=log⁡f(y)f(\log y) = \log f(y) for y∈1+my \in 1 + m; we show for example the first of these formulae:

f(exp⁡x)=∑n≥0f(xn)/n!=∑n≥0f(x)n/n!=exp⁡f(x). f(\exp x) = \sum_{n \geq 0} f(x^n)/n! = \sum_{n \geq 0} f(x)^n/n! = \exp f(x).

(4) Let E be a unital associative algebra. If a is a nilpotent element of E, the family (ann!)n∈N\left( \frac{a^n}{n!} \right)_{n \in \mathbf{N}} has finite support and we write exp⁡a=∑n≥0an/n!\exp a = \sum_{n \geq 0} a^n / n!. An element b is unipotent if b−1b - 1 is nilpotent; then we write

log⁡b=∑n≥1(−1)n−1(b−1)n/n. \log b = \sum_{n \geq 1} (-1)^{n-1} (b - 1)^n / n.

We deduce from the relations e(l(X))=l(e(X))=Xe(l(X)) = l(e(X)) = X that the mapping a↦exp⁡aa \mapsto \exp a is a bijection of the set of nilpotent elements of E onto the set of unipotent elements of E and that b↦log⁡bb \mapsto \log b is the inverse mapping.

2. Hausdorff group

Let X be a set. We use the notation of § 5, nos. 1 and 2. The free Lie algebra L(X) is identified with its canonical image in A(X) (§ 3, no. 1, Theorem 1). We denote by L^(X)\hat{L}(X) the closure of L(X) in A^(X)\hat{A}(X), that is the set of elements of A^(X)\hat{A}(X) of the form a=∑n≥1ana = \sum_{n \geq 1} a_n such that an∈Ln(X)a_n \in L^n(X) for all n≥0n \geq 0; this is filtered Lie subalgebra of A^(X)\hat{A}(X).

Theorem 1

The restriction of the exponential mapping of A^(X)\hat{A}(X) to L^(X)\hat{L}(X) is a bijection of L^(X)\hat{L}(X) onto a closed subgroup of the Magnus group Γ(X)\Gamma(X).

We write A(X)=A,An(X)=An,A^(X)=A^,Ln(X)=Ln,L^(X)=L^,Γ(X)=ΓA(X) = A, A^n(X) = A^n, \hat{A}(X) = \hat{A}, L^n(X) = L^n, \hat{L}(X) = \hat{L}, \Gamma(X) = \Gamma. Let B be the algebra A⊗AA \otimes A with the graduation of type N\mathbf{N} defined by Bn=∑i+j=nAi⊗AjB^n = \sum_{i+j=n} A^i \otimes A^j. Let B^=∏n≥0Bn\hat{B} = \prod_{n \geq 0} B^n be the associated complete filtered algebra (Commutative Algebra, Chapter III, § 2, no. 12, Example 1). The coproduct c:A→A⊗Ac : A \to A \otimes A defined in § 3, no. 1, Corollary 1 to Theorem 1 is graded of degree 0 and hence extends by continuity to a homomorphism c^:A^→B^\hat{c} : \hat{A} \to \hat{B} given by

c^(∑n≥0an)=∑n≥0c(an)for an∈An. \hat{c}\left( \sum_{n \geq 0} a_n \right) = \sum_{n \geq 0} c(a_n) \quad \text{for } a_n \in A^n.

We also define continuous homomorphisms δ′\delta' and δ′′\delta'' of A^\hat{A} into B^\hat{B} by

δ′(∑n≥0an)=∑n≥0(an⊗1),δ′′(∑n≥0an)=∑n≥0(1⊗an)for an∈An. \delta'\left( \sum_{n \geq 0} a_n \right) = \sum_{n \geq 0} (a_n \otimes 1), \qquad \delta''\left( \sum_{n \geq 0} a_n \right) = \sum_{n \geq 0} (1 \otimes a_n) \quad \text{for } a_n \in A^n.

By Corollary 2 to Theorem 1 of § 3, no. 1, LnL^n is the set of an∈Ana_n \in A^n such that c(an)=an⊗1+1⊗anc(a_n) = a_n \otimes 1 + 1 \otimes a_n. It follows that L^\hat{L} is the set of a∈A^a \in \hat{A} such that

c^(a)=δ′(a)+δ′′(a). \hat{c}(a) = \delta'(a) + \delta''(a).

Let Δ\Delta be the set of b∈A^b \in \hat{A} of constant term equal to 1 and satisfying the relation

c^(b)=δ′(b)⋅δ′′(b), \hat{c}(b) = \delta'(b) \cdot \delta''(b),

in other words, the set of b=∑n≥0bnb = \sum_{n \geq 0} b_n such that bn∈Anb_n \in \mathbf{A}^n for all n≥0n \geq 0, b0=1b_0 = 1 and c(bn)=∑i+j=nbi⊗bjc(b_n) = \sum_{i+j=n} b_i \otimes b_j for n≥0n \geq 0. The latter characterization shows that Δ\Delta is a closed subset of Γ\Gamma; as ϵ^\hat{\epsilon}, δ′\delta' and δ′′\delta'' are ring homomorphisms and every element of δ′(A^)\delta'(\hat{\mathbf{A}}) commutes with every element of δ′′(A^)\delta''(\hat{\mathbf{A}}), the restrictions to Γ\Gamma of the mappings cc and δ′δ′′\delta'\delta'' are group homomorphisms and Δ\Delta is a subgroup of Γ\Gamma.

By Proposition 1 of no. 1, the exponential mapping of A^\hat{\mathbf{A}} is a bijection of the set A^+\hat{\mathbf{A}}^+ of elements of A^\hat{\mathbf{A}} with no constant term onto Γ\Gamma. Let a∈A^+a \in \hat{\mathbf{A}}^+ and b=exp⁡ab = \exp a. As ϵ^\hat{\epsilon} is a continuous ring homomorphism,

ϵ^(b)=ϵ^(∑n≥0an/n!)=∑n≥0ϵ^(a)n/n!=exp⁡ϵ^(a). \hat{\epsilon}(b) = \hat{\epsilon}\left( \sum_{n \geq 0} a^n/n! \right) = \sum_{n \geq 0} \hat{\epsilon}(a)^n/n! = \exp \hat{\epsilon}(a).

The relations

δ′(b)=exp⁡δ′(a),δ′′(b)=exp⁡δ′′(a) \delta'(b) = \exp \delta'(a), \quad \delta''(b) = \exp \delta''(a)

are proved similarly and, as δ′(a)\delta'(a) commutes with δ′′(a)\delta''(a) (no. 1, Remark 1),

δ′(b)δ′′(b)=exp⁡(δ′(a)+δ′′(a)). \delta'(b) \delta''(b) = \exp (\delta'(a) + \delta''(a)).

Therefore aa satisfies (3) if and only if bb satisfies (4), which proves the theorem.

Remark

The above proof shows that exp⁡(L^)\exp(\hat{\mathbf{L}}) is the subgroup Δ\Delta of Γ\Gamma consisting of the bb satisfying (4).

Hence the group law of Δ\Delta can be transported by the exponential mapping to L^\hat{\mathbf{L}}. In other words, L^\hat{\mathbf{L}} is a complete topological group with the law of composition (a,b)↦ahb(a, b) \mapsto a \mathbin{\mathbf{h}} b given by

ahb=log⁡(exp⁡a⋅exp⁡b). a \mathbin{\mathbf{h}} b = \log(\exp a \cdot \exp b).

The topological group thus obtained is called the Hausdorff group (derived from X relative to K).

Let gg be the homomorphism of the free group F=F(X)F = F(X) into Γ\Gamma such that g(x)=exp⁡xg(x) = \exp x for x∈Xx \in X. As exp⁡x−1−x=∑n≥2xn/n!\exp x - 1 - x = \sum_{n \geq 2} x^n/n! is of order ≥2\geq 2, gg is injective by Theorem 1 of § 5, no. 3. Therefore the mapping log⁡∘g\log \circ g is an injective homomorphism of FF into the Hausdorff group which extends the canonical injection X→L^X \to \hat{\mathbf{L}}.

For every integer m≥1m \geq 1 we denote by L^m\hat{\mathbf{L}}_m the set of elements of order ≥m\geq m in L^\hat{\mathbf{L}} and by Γm\Gamma_m the set of u∈Γu \in \Gamma such that u−1u - 1 is of order ≥m\geq m. Then L^m=exp⁡(Γm)\hat{\mathbf{L}}_m = \exp(\Gamma_m) by Remark 2 of no. 1; as (Γm)m≥1(\Gamma_m)_{m \geq 1} is an integral central filtration on Γ\Gamma (§ 4, no. 5 Proposition 2), (L^m)m≥1(\hat{\mathbf{L}}_m)_{m \geq 1} is an integral central filtration on the group L^\hat{\mathbf{L}}.

3. Lie formal power series

Lemma 1

Let g\mathfrak{g} be a filtered Lie algebra (§ 4, no. 1), (gα)α∈R(\mathfrak{g}_\alpha)_{\alpha \in \mathbf{R}} its filtration and let α∈R\alpha \in \mathbf{R}. Let P be a homogeneous Lie polynomial of degree n in the indeterminates (Γi)i∈I(\Gamma_i)_{i \in I} (§ 2, no. 4). Then P((ai))∈gnαP((a_i)) \in \mathfrak{g}_{n\alpha} for every family (ai)i∈I(a_i)_{i \in I} of elements of gα\mathfrak{g}_\alpha*.

Every Lie polynomial of degree n⩾2n \geqslant 2 is a finite sum of terms of the form [Q,R][Q, R] where QQ and RR are of degree <n< n and the sum their degrees is equal to nn (§ 2, no. 7, Proposition 7). The lemma follows by induction on nn.

A Lie formal power series† (with coefficients in KK) in the indeterminates (Ti)i∈I(T_i)_{i \in I} is any element of the Lie algebra L^((Ti)i∈I)=L^(I)\hat{L}((T_i)_{i \in I}) = \hat{L}(I). Such an element uu can be written uniquely as the sum of a summable family (uν)ν∈N(I)(u_\nu)_{\nu \in \mathbf{N}^{(I)}} where uν∈Lν(I)u_\nu \in L^\nu(I).

Suppose that II is finite. Let gg be a complete Hausdorff filtered Lie algebra such that g=⋃α>0gαg = \bigcup_{\alpha > 0} g_\alpha; let t=(ti)i∈It = (t_i)_{i \in I} be a family of elements of gg.

Proposition 2

The homomorphism* ft:L(I)→gf_t : L(I) \to g such that ft(Ti)=tif_t(T_i) = t_i (§ 2, no. 4) can be extended by continuity to one and only one continuous homomorphism f^t\hat{f}_t of L^(I)\hat{L}(I) into gg.

There exists α>0\alpha > 0 such that ti∈gαt_i \in g_\alpha for all i∈Ii \in I; hence ft(Lν(I))⊂g∣ν∣∧αf_t(L^\nu(I)) \subset g_{|\nu| \wedge \alpha} for all ν\nu (Lemma 1), which implies the continuity of f^t\hat{f}_t.

If u∈L^(I)u \in \hat{L}(I), we write u((ti))=f^t(u)u((t_i)) = \hat{f}_t(u). In particular, taking g=L^(I)g = \hat{L}(I), u=u((Ti))u = u((T_i)); in the general case, u((ti))u((t_i)) is called the result of substituting the tit_i for the TiT_i in the Lie formal power series u((Ti))u((T_i)). If u=∑ν∈N(I)uνu = \sum_{\nu \in \mathbf{N}^{(I)}} u_\nu, where uν∈Lν(X)u_\nu \in L^\nu(X), the family (uν((ti)))ν∈N(I)(u_\nu((t_i)))_{\nu \in \mathbf{N}^{(I)}} is summable and

u((ti))=∑ν∈NIuν((ti)). u((t_i)) = \sum_{\nu \in \mathbf{N}^{I}} u_\nu((t_i)).

Let σ\sigma be a continuous homomorphism of gg into a complete Hausdorff filtered Lie algebra g′g' such that g′=⋃α>0gα′g' = \bigcup_{\alpha > 0} g'_\alpha. For every finite family t=(ti)i∈It = (t_i)_{i \in I} of elements of gg and all u∈L^(I)u \in \hat{L}(I),

σ(u((ti)))=u((σ(ti))), \sigma(u((t_i))) = u((\sigma(t_i))),
for the homomorphism σ∘f^t\sigma \circ \hat{f}_t is continuous and maps TiT_i to σ(ti)\sigma(t_i) for i∈Ii \in I.

Let u=(uj)j∈Ju = (u_j)_{j \in J} be a finite family of elements of L^(I)\hat{L}(I) and let v∈L^(J)v \in \hat{L}(J); by substituting the uju_j for the TjT_j in vv, we obtain an element w=v((uj)j∈J)w = v((u_j)_{j \in J}) of L^(I)\hat{L}(I) denoted by v∘uv \circ u. Then

w((ti)i∈I)=v((uj((ti)i∈I))j∈J) w((t_i)_{i \in I}) = v((u_j((t_i)_{i \in I}))_{j \in J})
for every finite family t=(ti)i∈It = (t_i)_{i \in I} of elements of gg, as is seen by operating with the continuous homomorphism f^t\hat{f}_t on the equation w=v((uj)j∈J)w = v((u_j)_{j \in J}).

Let u=∑ν∈NIuν∈L^(I)u = \sum_{\nu \in \mathbf{N}^{I}} u_\nu \in \hat{L}(I), where uν∈Lν(I)u_\nu \in L^\nu(I). The mapping u~\tilde{u} (ti)↦u((ti))(t_i) \mapsto u((t_i)) of gIg^I into gg is continuous: for in each of the open sets gαg_\alpha with α>0\alpha > 0 the family of u~α\tilde{u}_\alpha is uniformly summable and it suffices to prove that each u~ν\tilde{u}_\nu is continuous, which is immediate by induction on ∣ν∣|\nu|.

† A Lie formal power series is not in general a formal power series in the sense of Algebra, Chapter IV, § 6.

4. The Hausdorff series

Let {U, V} be a set with two elements.

Definition 1

The element* H=U⊛V=log⁡(exp⁡U⋅exp⁡V)H = U \circledast V = \log(\exp U \cdot \exp V) (no. 2) of the Lie algebra L^Q({U,V})\hat{L}_\mathbf{Q}(\{U, V\}) is called the Hausdorff series in the indeterminates U and V.

We denote by HnH_n (resp. Hr,sH_{r,s}) the homogeneous component of H of total degree n (resp. multidegree (r,s)(r, s)). Then

H=∑n≥0Hn=∑r,s≥0Hr,s,Hn=∑r+s=nr,s≥0Hr,s. H = \sum_{n \geq 0} H_n = \sum_{r, s \geq 0} H_{r,s}, \quad H_n = \sum_{r+s=n \atop r, s \geq 0} H_{r,s}.

Theorem 2

If r and s are two positive integers such that* r+s≥1r + s \geq 1, then Hr,s=Hr,s′+Hr,s′′H_{r,s} = H'_{r,s} + H''_{r,s}, *where

(r+s)Hr,s′=∑m≥1(−1)m−1m∑r1+⋯+rm=rs1+⋯+sm−1=s−11+s1≥1,…,rm−1+sm−1≥1(∏i=1m−1(ad U)riri!(ad V)sisi!)(ad U)rmrm!(V) (r + s)H'_{r,s} = \sum_{m \geq 1} \frac{(-1)^{m-1}}{m} \sum_{\substack{r_1 + \cdots + r_m = r \\ s_1 + \cdots + s_{m-1} = s-1 \\ 1 + s_1 \geq 1, \ldots, r_{m-1} + s_{m-1} \geq 1}} \left( \prod_{i=1}^{m-1} \frac{(\mathrm{ad}\, U)^{r_i}}{r_i!} \frac{(\mathrm{ad}\, V)^{s_i}}{s_i!} \right) \frac{(\mathrm{ad}\, U)^{r_m}}{r_m!} (V)
(r+s)Hr,s′′=∑m≥1(−1)m−1m∑r1+⋯+rm−1=r−1s1+⋯+sm−1=sr1+s1≥1,…,rm−1+sm−1≥1(∏i=1m−1(ad U)riri!(ad V)sisi!)(U). (r + s)H''_{r,s} = \sum_{m \geq 1} \frac{(-1)^{m-1}}{m} \sum_{\substack{r_1 + \cdots + r_{m-1} = r-1 \\ s_1 + \cdots + s_{m-1} = s \\ r_1 + s_1 \geq 1, \ldots, r_{m-1} + s_{m-1} \geq 1}} \left( \prod_{i=1}^{m-1} \frac{(\mathrm{ad}\, U)^{r_i}}{r_i!} \frac{(\mathrm{ad}\, V)^{s_i}}{s_i!} \right) (U).

In A^Q({U,V})\hat{A}_\mathbf{Q}(\{U, V\}), exp⁡U⋅exp⁡V=1+W\exp U \cdot \exp V = 1 + W, where W=∑r+s≥1UrVsr!s!W = \sum_{r+s \geq 1} \frac{U^r V^s}{r! s!}, whence

H=∑m≥1(−1)m−1Wm/m (no. 2), that is: H = \sum_{m \geq 1} (-1)^{m-1} W^m / m \text{ (no. 2)}, \text{ that is:}
Hr,s=∑m≥1(−1)m−1m∑r1+⋯+rm=rs1+⋯+sm=sr1+s1≥1,…,rm+sm≥1∏i=1mUriVsiri!si!. H_{r,s} = \sum_{m \geq 1} \frac{(-1)^{m-1}}{m} \sum_{\substack{r_1 + \cdots + r_m = r \\ s_1 + \cdots + s_m = s \\ r_1 + s_1 \geq 1, \ldots, r_m + s_m \geq 1}} \prod_{i=1}^m \frac{U^{r_i} V^{s_i}}{r_i! s_i!}.

The linear mapping PnP_n, defined by Pn(x1,…,xn)=1n(∏i=1n−1(ad xi))(xn)P_n(x_1, \ldots, x_n) = \frac{1}{n} \left( \prod_{i=1}^{n-1} (\mathrm{ad}\, x_i) \right)(x_n) for n≥1n \geq 1 and x1,…,xnx_1, \ldots, x_n in {U,V}\{U, V\}, is a projector of AQn({U,V})A^n_\mathbf{Q}(\{U, V\}) onto LQn({U,V})L^n_\mathbf{Q}(\{U, V\}) (§ 3, no. 2, Corollary to Proposition 1); as Hr,sH_{r,s} belongs to LQr+s({U,V})L^{r+s}_\mathbf{Q}(\{U, V\}), Hr,s=Pr+s(Hr,s)H_{r,s} = P_{r+s}(H_{r,s}). Now

Pr+s(∏i=1mUriVsiri!si!)=1r+s(∏i=1m−1(ad U)riri!(ad V)sisi!)(ad U)rmrm!(ad V)sm−1sm−1!(V) P_{r+s} \left( \prod_{i=1}^m \frac{U^{r_i} V^{s_i}}{r_i! s_i!} \right) = \frac{1}{r+s} \left( \prod_{i=1}^{m-1} \frac{(\mathrm{ad}\, U)^{r_i}}{r_i!} \frac{(\mathrm{ad}\, V)^{s_i}}{s_i!} \right) \frac{(\mathrm{ad}\, U)^{r_m}}{r_m!} \frac{(\mathrm{ad}\, V)^{s_{m-1}}}{s_{m-1}!} (V)

when sm⩾1s_m \geqslant 1 and

(13)Pr+s(∏i=1mUriri!Vsisi!)=1r+s((∏i=1m−1(ad U)riri!(ad V)sisi!)(ad U)rm−1rm!)(U) (13)\quad P_{r+s}\left(\prod_{i=1}^{m} \frac{U r_i}{r_i!} \frac{V s_i}{s_i!}\right) = \frac{1}{r+s} \left( \left( \prod_{i=1}^{m-1} \frac{(\mathrm{ad}\ U)^{r_i}}{r_i!} \frac{(\mathrm{ad}\ V)^{s_i}}{s_i!} \right) \frac{(\mathrm{ad}\ U)^{r_m-1}}{r_m!} \right)(U)

when rm⩾1r_m \geqslant 1 and sm=0s_m = 0. Moreover, obviously (ad t)p−1.t=0(\mathrm{ad}\ t)^{p-1}.t = 0 if p⩾2p \geqslant 2 and (ad t)0.t=t(\mathrm{ad}\ t)^0.t = t. It follows that the two sides of (12) are zero when sm⩾2s_m \geqslant 2 and those of (13) are zero when rm⩾2r_m \geqslant 2. The theorem then follows since Hr,s′H'_{r,s} is the sum of the terms of type (12) and Hr,s′′H''_{r,s} is the sum of the terms of type (13).

Remark

(1) We have defined (§ 3, no. 2, Remark) a projector QQ of A(X)A(X) onto L(X)L(X) such that Q(am)=0Q(a^m) = 0 for a∈L(X)a \in L(X) and m⩾2m \geqslant 2 and Q(1)=0Q(1) = 0. Then H=Q(exp⁡H)=Q(exp⁡U.exp⁡V)H = Q(\exp H) = Q(\exp U.\exp V), whence immediately

(14)Hr,s=Q(UrVsr!s!)for r+s⩾1. (14)\quad H_{r,s} = Q \left( \frac{U^r V^s}{r! s!} \right) \quad \text{for } r + s \geqslant 1.

(2) We have

(15)H(U,V)≡U+V+12[U,V]+112[U,[U,V]]+112[V,[V,U]]−124[U,[V,[U,V]]] (15)\quad H(U, V) \equiv U + V + \frac{1}{2}[U, V] + \frac{1}{12}[U, [U, V]] \\ + \frac{1}{12}[V, [V, U]] - \frac{1}{24}[U, [V, [U, V]]]
modulo ∑n⩾5Ln({U,V})\sum_{n \geqslant 5} L^n(\{U, V\}).

(3) H0,n=Hn,0=0H_{0,n} = H_{n,0} = 0 for every integer n≠1n \neq 1, whence

(16)H(U,0)=H(0,U)=U. (16)\quad H(U, 0) = H(0, U) = U.

On the other hand, as [U,−U]=0[U, -U] = 0,

(17)H(U,−U)=0. (17)\quad H(U, -U) = 0.

5. Substitutions in the Hausdorff series

As KK is a field containing Q\mathbf{Q}, the Hausdorff series can be considered as a Lie formal power series with coefficients in KK. Therefore, if gg is a complete Hausdorff filtered Lie algebra with g=⋃α>0gαg = \bigcup_{\alpha > 0} g_\alpha, then, for a,ba, b in gg, aa and bb can be substituted for UU and VV in HH (cf. no. 3 and § 2, no. 5, Remark).

In particular, let AA be a complete Hausdorff filtered unital associative algebra. We write m=⋃α>0Aαm = \bigcup_{\alpha > 0} A_\alpha and mα=Aα∩mm_\alpha = A_\alpha \cap m for α∈R\alpha \in \mathbf{R}; hence mα=Aαm_\alpha = A_\alpha for α>0\alpha > 0 and mα=mm_\alpha = m for α⩽0\alpha \leqslant 0. With the bracket [a,b]=ab−ba[a, b] = ab - ba, mm is a complete Hausdorff filtered Lie algebra, to which the above can be applied. With this notation, we have the following result which completes Proposition 1 of no. 1.

Proposition 3

If a∈m,b∈ma \in m, b \in m, then exp⁡H(a,b)=exp⁡a.exp⁡b\exp H(a, b) = \exp a . \exp b.

Let a,ba, b be in mm; there exists α>0\alpha > 0 such that a∈Aαa \in A_\alpha and b∈Aαb \in A_\alpha. Then there exists a continuous homomorphism θ\theta of the Magnus algebra A^({U,V})\hat{A}(\{U, V\}) into AA mapping UU to aa and VV to bb (§ 5, no. 1, Proposition 1).

The restriction of θ\theta to L^({U,V})\hat{L}(\{U, V\}) is a continuous homomorphism of Lie algebras of L({U,V})L(\{U, V\}) into mm which maps UU (resp. VV) to aa (resp. bb). By formula (6) of no. 3, therefore θ(H)=H(a,b)\theta(H) = H(a, b). It then suffices to apply the continuous homomorphism θ\theta to the two sides of the relation

exp⁡H(U,V)=exp⁡U.exp⁡V \exp H(U, V) = \exp U . \exp V
taking account of Remark 3 of no. 1.

Remark 1

If aa and bb commute, then Hr,s(a,b)=0H_{r,s}(a, b) = 0 for r+s≥2r + s \geq 2, for every homogeneous Lie polynomial of degree ≥2\geq 2 is zero at (a,b)(a, b). Then H(a,b)=a+bH(a, b) = a + b and Proposition 3 recovers the formula

exp⁡(a+b)=exp⁡a.exp⁡b. \exp(a + b) = \exp a . \exp b.

Proposition 4

Let gg be a complete Hausdorff filtered Lie algebra such that g=⋃α>0gαg = \bigcup_{\alpha > 0} g_\alpha. The mapping (a,b)↦H(a,b)(a, b) \mapsto H(a, b) is a group law on gg compatible with the topology on gg under which 00 is the identity element and −a-a is the inverse of aa for all a∈ga \in g.

The mapping (a,b)↦H(a,b)(a, b) \mapsto H(a, b) of g×gg \times g into gg is continuous (no. 3); as the mapping a↦−aa \mapsto -a is obviously continuous, it suffices to prove the relations

(18)H(H(a,b),c)=H(a,H(b,c))(19)H(a,−a)=0(20)H(a,0)=H(0,a)=a \begin{align*} (18) \quad & H(H(a, b), c) = H(a, H(b, c)) \\ (19) \quad & H(a, -a) = 0 \\ (20) \quad & H(a, 0) = H(0, a) = a \end{align*}
for a,b,ca, b, c in gg. By formula (7) of no. 3, it suffices to prove these formulae when a,b,ca, b, c are three indeterminates and g=L^({a,b,c})g = \hat{L}(\{a, b, c\}). Now the restriction of the exponential mapping to L^({a,b,c})\hat{L}(\{a, b, c\}) is an injection into the Magnus algebra A^({a,b,c})\hat{A}(\{a, b, c\}) and by Proposition 3:
exp⁡H(H(a,b),c)=exp⁡H(a,b).exp⁡c=exp⁡a.exp⁡b.exp⁡cexp⁡H(a,H(b,c))=exp⁡a.exp⁡H(b,c)=exp⁡a.exp⁡b.exp⁡cexp⁡H(a,−a)=exp⁡a.exp⁡(−a)=exp⁡(a−a)=exp⁡0exp⁡H(a,0)=exp⁡a.exp⁡0=exp⁡aexp⁡H(0,a)=exp⁡0.exp⁡a=exp⁡a. \begin{align*} \exp H(H(a, b), c) &= \exp H(a, b) . \exp c = \exp a . \exp b . \exp c \\ \exp H(a, H(b, c)) &= \exp a . \exp H(b, c) = \exp a . \exp b . \exp c \\ \exp H(a, -a) &= \exp a . \exp(-a) = \exp(a - a) = \exp 0 \\ \exp H(a, 0) &= \exp a . \exp 0 = \exp a \\ \exp H(0, a) &= \exp 0 . \exp a = \exp a. \end{align*}
This establishes relations (18) to (20).

Remark

(2) Take gg to be the Lie algebra L^(X)\hat{L}(X). The group law introduced in the above proposition coincides with the law defined in no. 2. In other words,

ahb=H(a,b)for a,b in L^(X); a \mathbin{\mathbf{h}} b = H(a, b) \quad \text{for } a, b \text{ in } \hat{L}(X);
thus the Hausdorff group law is given by the Hausdorff series.

(3) Let gg be a Lie algebra with the integral filtration (Cng\mathscr{C}^n g) defined by the lower central series. Suppose that there exists m≥1m \geq 1 such that Cmg={0}\mathscr{C}^m g = \{0\}. With the topology derived from the filtration (Cng)n≥1(\mathscr{C}^n g)_{n \geq 1}, the Lie algebra gg is Hausdorff, complete and even discrete. Then P(a1,…,ar)=0P(a_1, \ldots, a_r) = 0 for a1,…,ara_1, \ldots, a_r in gg and for every homogeneous Lie polynomial PP of degree ≥m\geq m; in particular, Hr,s(a,b)=0H_{r,s}(a, b) = 0 for r+s≥mr + s \geq m and the series H(a,b)=∑r,sHr,s(a,b)H(a, b) = \sum_{r,s} H_{r,s}(a, b) has only a finite number of non-zero terms. The group law (a,b)↦H(a,b)(a, b) \mapsto H(a, b) on gg is then a polynomial mapping (§ 2, no. 4).

Proposition 5

Let Kr,sK_{r,s} be the component of H(U+V,−U)H(U + V, -U) of multidegree (r,s)(r, s). Then*

Kn,1(U,V)=1(n+1)!(ad⁡U)n(V)for n≥0. K_{n,1}(U, V) = \frac{1}{(n+1)!} (\operatorname{ad} U)^n(V) \quad \text{for } n \geq 0.

We write K(U,V)=H(U+V,−U)K(U, V) = H(U + V, -U), K1(U,V)=∑n≥0Kn,1(U,V)K_1(U, V) = \sum_{n \geq 0} K_{n,1}(U, V). We denote by L (resp. R) left (resp. right) multiplication by UU on A^({U,V})\hat{A}(\{U, V\}).

We can write

eUVe−U=∑p,qUpp!V(−U)qq!=∑n≥01n!(∑p+q=nn!p!q!(Lp(−R)q)⋅V)=∑n≥01n!(L−R)nV e^{U}V e^{-U} = \sum_{p, q} \frac{U^p}{p!} V \frac{(-U)^q}{q!} = \sum_{n \geq 0} \frac{1}{n!} \left( \sum_{p+q=n} \frac{n!}{p!q!} (L^p(-R)^q) \cdot V \right) = \sum_{n \geq 0} \frac{1}{n!} (L - R)^n V
and therefore
eUVe−U=∑n≥01n!(ad⁡U)nV. e^{U}V e^{-U} = \sum_{n \geq 0} \frac{1}{n!} (\operatorname{ad} U)^n V.
We now calculate modulo the ideal ∑m≥0∑n≥2Am,n({U,V})\sum_{m \geq 0} \sum_{n \geq 2} A^{m,n}(\{U, V\}) of A({U,V})A(\{U, V\}). For all n≥1n \geq 1,
(U+V)n≡Un+∑i=1n−1UiVUn−1−i (U + V)^n \equiv U^n + \sum_{i=1}^{n-1} U^i V U^{n-1-i}
whence
(ad⁡U)(U+V)n≡((L−R)∑i=1n−1LiRn−i)⋅V≡(Ln−Rn)⋅V≡UnV−VUn. (\operatorname{ad} U)(U + V)^n \equiv ((L - R) \sum_{i=1}^{n-1} L^i R^{n-i}) \cdot V \\ \equiv (L^n - R^n) \cdot V \\ \equiv U^n V - V U^n.
Therefore
(ad⁡U)⋅eU+V≡eUV−VeU (\operatorname{ad} U) \cdot e^{U+V} \equiv e^{U}V - V e^{U}
summing over nn.

On the other hand, K1(U,V)≡K(U,V)K_1(U, V) \equiv K(U, V) and eK1(U,V)≡1+K1(U,V)e^{K_1(U, V)} \equiv 1 + K_1(U, V) and hence

K1≡eK−1≡eU+Ve−U−1 K_1 \equiv e^K - 1 \equiv e^{U+V}e^{-U} - 1
by Proposition 3. We deduce that
(ad U)K1≡UeU+Ve−U−eU+Ve−UU≡(eUV−VeU)e−U(by (23))≡eUVe−U−V≡∑n≥11n!(ad U)nV(by (22))≡(ad U)(∑n≥0(ad U)n(n+1)!V). \begin{align*} (\mathrm{ad}\ U)K_1 &\equiv Ue^{U+V}e^{-U} - e^{U+V}e^{-U}U \\ &\equiv (e^{UV} - Ve^U)e^{-U} \quad \text{(by (23))} \\ &\equiv e^{UV}e^{-U} - V \\ &\equiv \sum_{n \geq 1} \frac{1}{n!} (\mathrm{ad}\ U)^n V \quad \text{(by (22))} \\ &\equiv (\mathrm{ad}\ U)\left( \sum_{n \geq 0} \frac{(\mathrm{ad}\ U)^n}{(n+1)!} V \right). \end{align*}
It then suffices to apply the Remark of § 2, no. 11.

Exercises

See the exercises for § 6.

Transcribed from the printed volume. Not translated.