Bourbaki

§ 4. Passage from Lie algebras to Lie groups

Recall that, until the end of the chapter, K is assumed to be of characteristic 0.

1. Passage from Lie algebra morphisms to Lie group morphisms

Lemma 1

Let G be a Lie group germ and h\mathfrak{h} a Lie subalgebra of L(G)L(G) admitting a topological supplement. The union of the ghg\mathfrak{h} (resp. hg\mathfrak{h}g) for g∈Gg \in G is an integrable vector subbundle of T(G)T(G).

By considering the left trivialization of T(G)T(G) (§ 2, no. 3), it is seen immediately that the ghg\mathfrak{h}, for g∈Gg \in G, are the fibres of a vector subbundle E of T(G)T(G). Let g∈Gg \in G. The set of (La)g(L_a)_g, where a∈ha \in \mathfrak{h}, is equal to ghg\mathfrak{h}. Now, if aa and bb belong to h\mathfrak{h}, then [La,Lb]=L[a,b][L_a, L_b] = L_{[a, b]} and [a,b]∈h[a, b] \in \mathfrak{h}. Hence E is integrable (Differentiable and Analytic Manifolds, R, 9.3.3 (iv)). The argument is similar for the hg\mathfrak{h}g.

The integral foliation (Differentiable and Analytic Manifolds, R, 9.3.2) of the union of the ghg\mathfrak{h} (resp. hg\mathfrak{h}g) is called the left (resp. right) foliation of G associated with h\mathfrak{h}.

Theorem 1

Let G and H be Lie group germs and f a continuous morphism of L(G)L(G) into L(H)L(H). (i) There exist an open Lie subgroup germ G′G' of G and a morphism ϕ\phi of G′G' into H such that f=L(ϕ)f = L(\phi). (ii) Let G1,G2G_1, G_2 be open Lie subgroup germs of G and ϕi\phi_i a morphism of GiG_i into H such that f=L(ϕi)f = L(\phi_i) for i=1,2i = 1, 2. Then ϕ1\phi_1 and ϕ2\phi_2 coincide on a neighbourhood of e. Let p1:G×H→G,p2:G×H→Hp_1 : G \times H \to G, p_2 : G \times H \to H be the canonical projections. For all (g,h)∈G×H(g, h) \in G \times H, let fg,hf_{g, h} be the mapping ga↦hf(a)ga \mapsto hf(a) of Tg(G)=gL(G)T_g(G) = gL(G) into Th(H)=hL(H)T_h(H) = hL(H). By considering the left trivializations of T(G)T(G) and T(H)T(H), it is seen immediately that the fg,hf_{g, h} define a morphism of p1∗T(G)p_1^*T(G) into* p2∗T(H)p_2^*T(H). Let aa be the graph of ff; it is a closed Lie subalgebra of L(G)×L(H)L(G) \times L(H) which admits {0}×L(H)\{0\} \times L(H) as topological supplement. For all (g,h)∈G×H(g, h) \in G \times H, the graph of fg,hf_{g,h} is (g,h)⋅a(g, h) \cdot a. The union of these graphs is an integrable vector subbundle of T(G×H)T(G \times H) (Lemma 1). Then there exist (Differentiable and Analytic Manifolds, R, 9.3.7) an open neighbourhood UU of eGe_G in GG and an analytic mapping ϕ\phi of UU into HH such that ϕ(eG)=eH\phi(e_G) = e_H and Tg(ϕ)=fg,ϕ(g)T_g(\phi) = f_{g, \phi(g)} for all g∈Ug \in U. In particular, TeG(ϕ)=fT_{e_G}(\phi) = f.

Let VV be an open neighbourhood of eGe_G in GG such that, for (s,t)∈V×V(s, t) \in V \times V, the products stst and ϕ(s)ϕ(t)\phi(s)\phi(t) are defined and st∈Ust \in U. Consider the mappings α1,α2\alpha_1, \alpha_2 of V×VV \times V into HH defined by

α1(s,t)=ϕ(ts),α2(s,t)=ϕ(t)ϕ(s). \alpha_1(s, t) = \phi(ts), \quad \alpha_2(s, t) = \phi(t)\phi(s).

Then α1(t,e)=ϕ(t)=α2(t,e)\alpha_1(t, e) = \phi(t) = \alpha_2(t, e). On the other hand, let tt be fixed in VV and βi\beta_i be the mapping s↦αi(s,t)s \mapsto \alpha_i(s, t) of VV into HH. Then, for all s∈Vs \in V and all a∈L(G)a \in L(G),

Ts(β1)(sa)=Tts(ϕ)(tsa)=fts,ϕ(ts)(tsa)=ϕ(ts)f(a)=fs,β1(s)(sa)Ts(β2)(sa)=ϕ(t)Ts(ϕ)(sa)=ϕ(t)fs,ϕ(s)(sa)=ϕ(t)ϕ(s)f(a)=fs,β2(s)(sa). \begin{align*} T_s(\beta_1)(sa) &= T_{ts}(\phi)(tsa) = f_{ts, \phi(ts)}(tsa) \\ &= \phi(ts)f(a) = f_{s, \beta_1(s)}(sa) \\ T_s(\beta_2)(sa) &= \phi(t)T_s(\phi)(sa) = \phi(t)f_{s, \phi(s)}(sa) \\ &= \phi(t)\phi(s)f(a) = f_{s, \beta_2(s)}(sa). \end{align*}

Hence (Differentiable and Analytic Manifolds, R, 9.3.7) α1\alpha_1 and α2\alpha_2 coincide on a neighbourhood of (eG,eG)(e_G, e_G). The restriction of ϕ\phi to a sufficiently small symmetric open neighbourhood of eGe_G is therefore a morphism of Lie group germs, whence (i).

Let G1,G2,ϕ1,ϕ2G_1, G_2, \phi_1, \phi_2 be as in (ii) and let us prove that ϕ1,ϕ2\phi_1, \phi_2 coincide on a neighbourhood of eGe_G. There exists an open neighbourhood WW of eGe_G such that ϕ1(ts)=ϕ1(t)ϕ1(s)\phi_1(ts) = \phi_1(t)\phi_1(s), ϕ2(ts)=ϕ2(t)ϕ2(s)\phi_2(ts) = \phi_2(t)\phi_2(s) for all s,ts, t in WW. Then, if s∈Ws \in W and a∈L(G)a \in L(G),

Ts(ϕi)(sa)=ϕi(s)Te(ϕi)(a)=ϕi(s)f(a)=fs,ϕi(s)(sa) T_s(\phi_i)(sa) = \phi_i(s)T_e(\phi_i)(a) = \phi_i(s)f(a) = f_{s, \phi_i(s)}(sa)

for i=1,2i = 1, 2. As ϕ1(eG)=eH=ϕ2(eG)\phi_1(e_G) = e_H = \phi_2(e_G), it follows from Differentiable and Analytic Manifolds, R, 9.3.7 that ϕ1\phi_1 and ϕ2\phi_2 coincide on a neighbourhood of eGe_G.

Corollary 1

Let GG and HH be two Lie group germs. If L(G)L(G) and L(H)L(H) are isomorphic, GG and HH are locally isomorphic.

This follows from Theorem 1 and § 1, no. 10, Proposition 21.

Corollary 2

Let GG be a Lie group germ. If L(G)L(G) is commutative, GG is locally isomorphic to the additive Lie group L(G)L(G).

The Lie algebra of the additive group L(G)L(G) is isomorphic to L(G)L(G). Hence it suffices to apply Corollary 1.

Corollary 3

Let GG be a Lie group. If L(G)L(G) is commutative, GG contains a commutative open subgroup.

There exists an open Lie subgroup germ U of G which is commutative (Corollary 2). Let V be a neighbourhood of e such that V^2 ⊂ U. Then xy = yx for all x, y in V. Hence the subgroup of G generated by V is commutative; it is obviously open.

2. Passage from Lie algebras to Lie groups

We shall denote the Hausdorff series by H(X, Y) (Chapter II, § 6, no. 4, Definition 1).

Lemma 2

Let L be a complete normed Lie algebra over R\mathbf{R} or C\mathbf{C}. Let G be the set of x∈Lx \in L such that ∥x∥<13log⁡32\|x\| < \frac{1}{3} \log \frac{3}{2}. Let θ\theta be the mapping x↦−xx \mapsto -x of G into G. Let H be the restriction to G×GG \times G of the Hausdorff function of L (Chapter II, § 7, no. 2). (i) (G,0,θ,H)(G, 0, \theta, H) is a Lie group germ. (ii) Let ϕ\phi be the identity mapping of G into L. The differential of ϕ\phi at 0 is an isomorphism of the normable Lie algebra L(G)L(G) onto L. (i) follows from Chapter II, § 7, no. 2. As ϕ\phi is a chart on G, the differential ψ\psi of ϕ\phi at 0 is an isomorphism of normable spaces. On the other hand, the expansion as an integral series H=∑i,j≥0HijH = \sum_{i,j \geq 0} H_{ij} of the mapping H is such that H11(x,y)=12[x,y]H_{11}(x, y) = \frac{1}{2}[x, y]. By § 3, Proposition 24, for all a,ba, b in L(G)L(G),

ψ([a,b])=H11(ψ(a),ψ(b))−H11(ψ(b),ψ(a))=[ψ(a),ψ(b)] \psi([a, b]) = H_{11}(\psi(a), \psi(b)) - H_{11}(\psi(b), \psi(a)) = [\psi(a), \psi(b)]
which proves (ii).

G is called the Lie group germ defined by L.

Suppose that K is ultrametric. Let pp be the characteristic of the residue field of K. If p≠0p \neq 0, let λ=∣p∣1/(p−1)\lambda = |p|^{1/(p-1)}; if p=0p = 0, let λ=1\lambda = 1.

Lemma 3

Let L be a complete normed Lie algebra over K. Let G be the set of x∈Lx \in L such that ∥x∥<λ\|x\| < \lambda. Let H:G×G→GH : G \times G \to G be the Hausdorff function of L (Chapter II, § 8, no. 3). (i) With the law of composition H, G is a Lie group in which 0 is identity element and −x-x the inverse of x for all x∈Gx \in G. (ii) Let ϕ\phi be the identity mapping of G into L. The differential of ϕ\phi at 0 is an isomorphism of the normable Lie algebra L(G)L(G) onto L. (iii) For all μ∈R+∗\mu \in \mathbf{R}_+^*, let GμG_\mu be the set of x∈Lx \in L such that ∥x∥<μ\|x\| < \mu. Then the GμG_\mu, for μ<λ\mu < \lambda, form a fundamental system of open and closed neighbourhoods of 0 and are subgroups of G. Assertions (i) and (iii) follow from Chapter II, § 8, no. 3, Proposition 3 and (ii) can be proved as in Lemma 2.

G is called the Lie group defined by L.

Theorem 2

Let L be a complete normable Lie algebra. There exists a Lie group germ G such that L(G) is isomorphic to L. Two such Lie group germs are locally isomorphic.

The first assertion follows from Lemmas 2 and 3. The second assertion follows from Corollary 1 to Theorem 1 of no. 1.

Corollary 1

Let G be a Lie group. There exists a neighbourhood of e which contains no finite subgroup distinct from {e}. If K = R or C, there exists an open neighbourhood of e which contains no subgroup distinct from {e}.

We write L(G) = L. Choose a norm on L defining the topology on L and such that ∥[x,y]∥≤∥x∥∥y∥\| [x, y] \| \leq \| x \| \| y \| for all x, y in L.

Suppose that K = R or C. Let G' be the Lie group germ defined by L. There exist an open ball U' of centre 0 in G' and an isomorphism ϕ\phi of the Lie group germ U' onto an open neighbourhood U of e in G. Let V' = 12U′\frac{1}{2} U', V=ϕ(V′)V = \phi(V'), H be a subgroup of G contained in V and h∈Hh \in H. Let x=ϕ−1(h)∈V′x = \phi^{-1}(h) \in V'. If x≠0x \neq 0, there exists an integer n>0n > 0 such that x,2x,…,nxx, 2x, \ldots, nx are in V', (n+1)x∈U′(n+1)x \in U', (n+1)x∉V′(n+1)x \notin V'. Then h,h2,…,hnh, h^2, \ldots, h^n are in V,

hn+1∈U,hn+1∉V, h^{n+1} \in U, \qquad h^{n+1} \notin V,

which is absurd. Hence H = {e}.

Suppose that K is ultrametric. It suffices to prove the corollary when G is the Lie group associated with L. If g∈Gg \in G, the powers of g evaluated in G are the elements of Zg\mathbf{Z}g evaluated in L. These are distinct if g≠eg \neq e. Hence G contains no finite subgroup distinct from {e}.

Corollary 2

Let k be a non-discrete closed subfield of K, G a Lie group over k and L = L(G). Suppose that L has a normable Lie K-algebra structure L', compatible with the normable Lie k-algebra structure and invariant under the adjoint representation of G. Then there exists on G one and only one Lie K-group structure compatible with the Lie k-group structure and for which the Lie algebra is L'.

There exists a Lie group germ G1G_1 over K such that L(G1)=L′L(G_1) = L' (Theorem 2). By Corollary 1 to Theorem 1 of no. 1, G and G1G_1, considered as Lie k-group germs, are locally isomorphic. Hence there exist an open neighbourhood G' of e in G and a Lie K-group germ structure on G', with Lie algebra L, which is compatible with the Lie group germ structure over k. Let V be a symmetric open neighbourhood of e in G such that V2⊂G′V^2 \subset G'. Let g∈Gg \in G. Then ϕ=Int⁡g\phi = \operatorname{Int} g is a k-isomorphism of a sufficiently small open Lie subgroup germ of G' onto an open Lie subgroup germ of G'; and Te(ϕ)T_e(\phi) is K-linear and hence Tx(ϕ)T_x(\phi) is K-linear for x sufficiently close to e; therefore the restriction of Int g to a sufficiently small open neighbourhood of e in V is K-analytic (Differentiable and Analytic Manifolds, R, 5.14.6). By § 1, no. 9, Proposition 18, there exists on G an analytic K-manifold structure with which G is a Lie

K-group and V is an open sub-K-manifold of G. By translation, it is seen that the underlying k-manifold structure of G is the given structure. The Lie algebra of the Lie K-group G is the same as that of the open Lie K-subgroup germ V and hence is L'. Finally, the uniqueness stated in the corollary follows from § 3, no. 8, Proposition 32.

Theorem 3

Let G be a Lie group germ and h\mathfrak{h} a Lie subalgebra of L(G)L(G) admitting a topological supplement. There exists a Lie subgroup germ H of G such that L(H)=hL(H) = \mathfrak{h}. If H1H_1 and H2H_2 are Lie subgroup germs of G such that

L(H1)=L(H2)=h, L(H_1) = L(H_2) = \mathfrak{h},

then H1∩H2H_1 \cap H_2 is open in H1H_1 and H2H_2.

There exists a Lie group germ H′H' with Lie algebra isomorphic to h\mathfrak{h} (Theorem 2). Shrinking H′H' if necessary, it can be assumed that there exists a morphism ϕ\phi of H′H' into G such that L(ϕ)L(\phi) is an isomorphism of L(H′)L(H') onto h\mathfrak{h} (no. 1, Theorem 1). As h\mathfrak{h} admits a topological supplement, ϕ\phi is an immersion at e. Hence, shrinking H′H' further, it can be assumed that ϕ\phi is an isomorphism of the manifold H′H' onto a submanifold of G. This proves the existence of H. The second assertion follows from the following proposition:

Proposition 1

Let G be a Lie group germ and H and H′H' two Lie subgroup germs. In order that L(H)⊃L(H′)L(H) \supset L(H'), it is necessary and sufficient that H∩H′H \cap H' be open in H′H'.

If H∩H′H \cap H' is open in H′H', then L(H′)=L(H∩H′)⊂L(H)L(H') = L(H \cap H') \subset L(H). Suppose that L(H)⊃L(H′)L(H) \supset L(H'). Let i, i′i' be the canonical injections of H, H′H' into G. By shrinking H′H' if necessary, it can be assumed that there exists a morphism ψ\psi of H′H' into H such that L(ψ)L(\psi) is the canonical injection of L(H′)L(H') into L(H)L(H) (no. 1, Theorem 1). Then L(i∘ψ)=L(i′)L(i \circ \psi) = L(i') and hence there exists a neighbourhood V of eH′e_{H'} in H′H' such that i∘ψi \circ \psi and i′i' coincide on V (Theorem 1). Therefore V⊂HV \subset H, hence V⊂H∩H′V \subset H \cap H' and H∩H′H \cap H' is open in H′H' (§ 1, no. 10).

Proposition 2

Let G be a Lie group over K, k a non-discrete closed subfield of K and H a Lie subgroup of the Lie k-group G. Suppose that L(H)L(H) is a vector sub-K-space of L(G)L(G) which admits a topological supplement. Then H is a Lie subgroup of the Lie K-group G.

There exists a Lie subgroup germ H′H' of the Lie K-group G such that L(H′)=L(H)L(H') = L(H) (Theorem 3). Consider G, H, H′H' as Lie k-group germs; Theorem 3 then proves that H∩H′H \cap H' is open in H and H′H'. Hence there exists an open neighbourhood U of e in G such that U∩HU \cap H is a submanifold of G over K. Therefore, H is a Lie subgroup of the Lie K-group G (§ 1, no. 3, Proposition 6).

3. Exponential mappings

Theorem 4

Let G be a Lie group germ, L its Lie algebra, V an open neighbourhood of 0 in L, φ an analytic mapping of V into G such that φ(0) = 0 and T_0(φ) = Id_L. The following conditions are equivalent: (i) For all b ∈ L, φ((λ + λ')b) = φ(λb)φ(λ'b) for |λ| and |λ'| sufficiently small. (ii) For all b ∈ L and every integer n > 0, φ_*(b^n) is homogeneous of degree n in U(G) (T_0^{(∞)}(L) is identified with TS(L) and b^n is evaluated in TS(L)). (iii) The mapping φ_* of TS(L) into U(G) is compatible with the graduations of TS(L) and U(G). (iv) The mapping φ_* of TS(L) into U(G) is the canonical mapping of TS(L) into the enveloping algebra of L. (v) There exist a norm on L defining the topology of L and such that

∥[x,y]∥≤∥x∥∥y∥ \|[x, y]\| \leq \|x\|\|y\|
for all x, y in L and an open subgroup germ W ⊂ V of the Lie group germ defined by L (no. 2) such that φ|W is an isomorphism of W onto an open Lie subgroup germ of G.

(v) ⇒ (i): obvious, for (λb).(λ'b) = (λ + λ')b in W for |λ| and |λ'| sufficiently small. (i) ⇒ (ii): suppose that condition (i) is satisfied. Let b ∈ L. Let ψ be the restriction of φ to V ∩ Kb. By hypothesis, there exists a symmetric neighbourhood T of 0 in the additive Lie group Kb such that ψ|T is a morphism of the Lie group germ T into G. Hence

φ∗(bn)=(ψ∣T)∗(bn)=((ψ∣T)∗(b))n=(φ∗(b))n, \varphi_*(b^n) = (\psi|T)_*(b^n) = ((\psi|T)_*(b))^n = (\varphi_*(b))^n,
so that φ_*(b^n) is homogeneous of degree n in U(G). (ii) ⇒ (iii): this follows from the fact that TS^n(L) is the vector subspace of TS(L) generated by the n-th powers of the elements of L (Algebra, Chapter IV, § 5, Proposition 5). (iii) ⇒ (iv): the canonical mapping of TS(L) into the enveloping algebra of L is the unique morphism of graded cogebras mapping 1 to 1 and extending Id_L (Chapter II, § 1, no. 5, Remark 3). Now φ_* is a cogebra morphism and φ_*|L = Id_L by hypothesis. If condition (iii) holds, it is seen that condition (iv) also holds. (iv) ⇒ (v): suppose that condition (iv) is satisfied. Choose a norm on L defining the topology of L and such that \|[x, y]\| \leq \|x\|\|y\| for all x, y in L. Let H be the Lie group germ defined by the normed Lie algebra L. By Theorem 1, there exist an open subgroup germ S ⊂ V of H and an isomorphism φ' of S onto an open subgroup germ of G. As we know already that (v) ⇒ (iv), the mapping φ_* of TS(L) into U(G) is the canonical mapping of TS(L) into the enveloping algebra of L. Thus φ_*(t) = φ'_*(t) for all t ∈ T_0^{(∞)}(L). As φ and φ' are analytic, φ and φ' coincide on a neighbourhood of 0.

Definition 1

Let G be a Lie group germ and L its Lie algebra. An exponential mapping of G is any analytic mapping ϕ\phi defined on an open neighbourhood of 0 in L, with values in G and satisfying the conditions of Theorem 4.

Theorem 4 implies immediately that, for every Lie group germ G, there exists an exponential mapping of G and that two exponential mappings of G coincide on a neighbourhood of 0.

Example

(1) Let G be the additive group of a complete normable space E. The canonical isomorphism of L(G) onto E satisfies condition (i) of Theorem 4 and is therefore an exponential mapping of G.

(2) Let A be a complete normed unital associative algebra. Let A* be the Lie group consisting of the invertible elements of A. We identify L(A*) with A (§ 3, no. 9, Corollary to Proposition 33). If K = \mathbf{R} or \mathbf{C}, we know that the mapping exp of A into A* defined in Chapter II, § 7, no. 3 satisfies condition (i) of Theorem 4 and hence is an exponential mapping. Now let K be ultrametric. Let pp be the characteristic of the residue field of K. If p≠0p \neq 0, let λ=∣p∣1/(p−1)\lambda = |p|^{1/(p-1)}; if p=0p = 0, let λ=1\lambda = 1. Let U be the set of x∈Ax \in A such that ∥x∥<λ\|x\| < \lambda. We know (Chapter II, § 8, no. 4) that the mapping exp of U into A* satisfies condition (i) of Theorem 4 and hence is an exponential mapping. Note that U is an additive subgroup of A.

This example explains the terminology adopted in Definition 1.

Let G be a Lie group germ and ϕ\phi an exponential mapping of G. Then ϕ\phi is étale at 0 and hence there exists an open neighbourhood U of 0 in L(G) such that ϕ(U)\phi(U) is open in G and ϕ∣U\phi|U is an isomorphism of the analytic manifold U onto the analytic manifold ϕ(U)\phi(U).

A canonical chart (of the first species) on G is a chart ψ\psi on the analytic manifold G whose inverse mapping is an exponential mapping. If further G is finite-dimensional and a basis of L(G) is chosen, the coordinate system defined by ψ\psi and this basis in the domain of ψ\psi is called a canonical coordinate system (of the first species).

Proposition 3

Let G be a Lie group germ, L its Lie algebra and ϕ\phi an exponential mapping of G. Let L1,…,LnL_1, \ldots, L_n be vector subspaces of L such that L is the topological direct sum of L1,…,LnL_1, \ldots, L_n. The mapping

(b1,b2,…,bn)↦θ(b1,b2,…,bn)=ϕ(b1)ϕ(b2)⋯ϕ(bn), (b_1, b_2, \ldots, b_n) \mapsto \theta(b_1, b_2, \ldots, b_n) = \phi(b_1)\phi(b_2)\cdots\phi(b_n),

defined on an open subset of L1×L2×⋯×LnL_1 \times L_2 \times \cdots \times L_n, is analytic. The tangent mapping at (0,0,…,0)(0, 0, \ldots, 0) to θ\theta is the canonical mapping of L1×⋯×LnL_1 \times \cdots \times L_n into L.

Let kik_i be the canonical injection of LiL_i into L1×L2×⋯×LnL_1 \times L_2 \times \cdots \times L_n. Then, for all b∈Lib \in L_i, (T(0,…,0)θ)(T0ki)(b)=(T0ϕ)(b)=b(T_{(0, \ldots, 0)}\theta)(T_0k_i)(b) = (T_0\phi)(b) = b and hence (T(0,…,0)θ)∣Li(T_{(0, \ldots, 0)}\theta)|L_i is the canonical injection of LiL_i into L.

In particular, θ\theta is étale at (0,0,…,0)(0, 0, \ldots, 0). Its restriction to a sufficiently small

LIE GROUPS

open neighbourhood U of (0,0,…,0)(0, 0, \ldots, 0) has an open image in G and is an isomorphism of the manifold U onto the manifold θ(U)\theta(U). The inverse mapping η\eta of θ(U)\theta(U) onto U is called a canonical chart of the second species of G, associated with the given decomposition of L as a direct sum. If further G is finite-dimensional and each LiL_i is generated by a non-zero vector eie_i, the coordinate system on θ(U)\theta(U) defined by η\eta and the eie_i is called a canonical coordinate system of the second species.

Proposition 4

Let G be a Lie group germ and ϕ\phi an injective exponential mapping of G. For all x, y in L(G)L(G),

(1)

x+y=lim⁡λ∈K∗,λ→0λ−1ϕ−1(ϕ(λx)ϕ(λy)) x + y = \lim_{\lambda \in K^*, \lambda \to 0} \lambda^{-1} \phi^{-1}(\phi(\lambda x) \phi(\lambda y))
(2)
[x,y]=lim⁡λ∈K∗,λ→0λ−2ϕ−1(ϕ(λx)ϕ(λy)ϕ(−λx)ϕ(−λy)) [x, y] = \lim_{\lambda \in K^*, \lambda \to 0} \lambda^{-2} \phi^{-1}(\phi(\lambda x) \phi(\lambda y) \phi(-\lambda x) \phi(-\lambda y))
(note that ϕ−1(ϕ(λx)ϕ(λy))\phi^{-1}(\phi(\lambda x) \phi(\lambda y)) and ϕ−1(ϕ(λx)ϕ(λy)ϕ(−λx)ϕ(−λy))\phi^{-1}(\phi(\lambda x) \phi(\lambda y) \phi(-\lambda x) \phi(-\lambda y)) are defined for ∣λ∣|\lambda| sufficiently small).

Let L=L(G)L = L(G) be given a norm defining the topology of L and such that ∥[x,y]∥≤∥x∥∥y∥\| [x, y] \| \leq \| x \| \| y \| for all x,yx, y in L. Using Theorems 2 and 4, it can be assumed that G is the Lie group germ defined by L and that ϕ=IdG\phi = \mathrm{Id}_G. Let (x,y)↦x⋅y(x, y) \mapsto x \cdot y denote the product in the group G. The formulae to be proved can then be written (3)

x+y=lim⁡λ∈K∗,λ→0λ−1((λx)⋅(λy)) x + y = \lim_{\lambda \in K^*, \lambda \to 0} \lambda^{-1}((\lambda x) \cdot (\lambda y))
(4)
[x,y]=lim⁡λ∈K∗,λ→0λ−2((λx)⋅(λy)⋅(−λx)⋅(−λy)). [x, y] = \lim_{\lambda \in K^*, \lambda \to 0} \lambda^{-2}((\lambda x) \cdot (\lambda y) \cdot (-\lambda x) \cdot (-\lambda y)).

There exists an open neighbourhood V of 0 in K such that the function

λ↦f(λ)=(λx)⋅(λy) \lambda \mapsto f(\lambda) = (\lambda x) \cdot (\lambda y)
is defined and analytic on V. By Chapter II, § 6, no. 4, Remark 2, the expansion of ff as an integral series about the origin is
λ(x+y)+12λ2[x,y]+⋯ \lambda(x + y) + \frac{1}{2} \lambda^2 [x, y] + \cdots
and this proves (3). On the other hand, for u,vu, v in G and ∥u∥,∥v∥\|u\|, \|v\| sufficiently small, u⋅vu \cdot v is an analytic function of (u,v)(u, v) and the terms of degrees 1 and 2 in the expansion of this function as an integral series about the origin are u+v+12[u,v]u + v + \frac{1}{2}[u, v]. By Differentiable and Analytic Manifolds, R, 3.2.7 and 4.2.3, the terms of degrees 1 and 2 in the expansion of the function f(λ)⋅f(−λ)f(\lambda) \cdot f(-\lambda) as an integral series about the origin are the terms of degrees 1 and 2 in
f(λ)+f(−λ)+12[f(λ),f(−λ)] f(\lambda) + f(-\lambda) + \frac{1}{2}[f(\lambda), f(-\lambda)]
or also in
λ(x+y)+12λ2[x,y]−λ(x+y)+12λ2[x,y]+12[λ(x+y),−λ(x+y)]=λ2[x,y] \lambda(x + y) + \frac{1}{2} \lambda^2 [x, y] - \lambda(x + y) + \frac{1}{2} \lambda^2 [x, y] \\ + \frac{1}{2} [\lambda(x + y), -\lambda(x + y)] = \lambda^2 [x, y]
and this proves (4).

EXponential mappings

§ 4.3

Proposition 5

Let G be a Lie group, k a non-discrete closed subfield of K, G' the group G considered as a Lie group over k and ϕ\phi (resp. ϕ′\phi') an exponential mapping of G (resp. G'). Then ϕ\phi and ϕ′\phi' coincide on a neighbourhood of 0.

ϕ\phi satisfies hypothesis (i) of Theorem 4 relative to G' and is therefore an exponential mapping of G'.

Proposition 6

Let G be a Lie group germ, L its Lie algebra and ϕ:V→G\phi : V \to G an exponential mapping of G. For all x∈Vx \in V, let Tx(L)T_x(L) be identified with L, so that the right differential ϖ(x)\varpi(x) of ϕ\phi at x is a linear mapping of L into L. For x sufficiently close to 0,

ϖ(x)=∑n≥01(n+1)!(ad⁡x)n. \varpi(x) = \sum_{n \geq 0} \frac{1}{(n+1)!} (\operatorname{ad} x)^n.

Let L be given a norm compatible with its topology and such that ∥[x,y]∥≤∥x∥∥y∥\| [x, y] \| \leq \| x \| \| y \| for all x,y∈Lx, y \in L. It suffices to consider the case where G is the Lie group germ defined by L and ϕ=Id⁡G\phi = \operatorname{Id}_G. By definition, ϖ(x)\varpi(x) is then the tangent mapping at x to the mapping y↦y.x−1y \mapsto y . x^{-1} of G into G. If H(X,Y)H(X, Y) denotes the Hausdorff series, ϖ(x)\varpi(x) is therefore, for ∥x∥\| x \| sufficiently small, the tangent mapping at 0 to the mapping y↦H(x,+y,−x)y \mapsto H(x, +y, -x) of G into G. In H(X+Y,−X)H(X + Y, -X), the sum of terms of the first degree in Y is

∑m≥01(m+1)!(ad⁡X)mY \sum_{m \geq 0} \frac{1}{(m+1)!} (\operatorname{ad} X)^m Y

(Chapter II, § 6, no. 5, Proposition 5). The proposition then follows from Differentiable and Analytic Manifolds, R, 3.2.4 and 4.2.3.

Let G be a Lie group germ and t∈Kt \in K. A t-th power mapping of G is any mapping, defined and analytic on an open neighbourhood of e, with values in G and coinciding on a neighbourhood of e with a mapping

g↦ϕ(tϕ−1(g)) g \mapsto \phi(t \phi^{-1}(g))

where ϕ\phi is an injective exponential mapping of G.

Proposition 7

(i) *If t∈Zt \in \mathbf{Z}, a t-th power mapping coincides on a neighbourhood of e with the mapping g↦gtg \mapsto g^t.

(ii) The tangent mapping at e to a t-th power mapping is the homothety of ratio t.

(iii) If h is a t-th power mapping and h' a t'-th power mapping of G, h∘h′h \circ h' is a (tt')-th power mapping and g↦h(g)h′(g)g \mapsto h(g)h'(g) is a (t + t')-th power mapping.

(iv) If h is a t-th power mapping and u∈Un(G)u \in U^n(G), then h∗(u)=tnuh_*(u) = t^n u.

It suffices to prove the proposition when G is the Lie group germ defined by a complete normed Lie algebra and when the t-th power mappings considered are constructed using the exponential mapping ϕ=Id⁡G\phi = \operatorname{Id}_G. But in that case everything is obvious.

4. Functorization of exponential mappings

Proposition 8

Let G and H be Lie group germs, h a morphism of G into H and ϕG\phi_G and ϕH\phi_H exponential mappings of G and H. There exists a neighbourhood V of 0 in L(G)L(G) such that h∘ϕGh \circ \phi_G and ϕH∘L(h)\phi_H \circ L(h) coincide on V.

Let L(G)L(G) and L(H)L(H) be given norms defining their topologies and such that ∥[x,y]∥≤∥x∥∥y∥\| [x, y] \| \leq \| x \| \| y \| for all x and y. It can be assumed that G (resp. H) is the Lie group germ defined by L(G)L(G) (resp. L(H)L(H)), so that ϕG\phi_G (resp. ϕH\phi_H) coincides with IdG\mathrm{Id}_G (resp. IdH\mathrm{Id}_H) in a neighbourhood of 0. On the other hand, there exists a symmetric open neighbourhood W of 0 in L(G)L(G) such that L(h)L(h) is a morphism of the Lie group germ W into H. By Theorem 1, L(h)L(h) coincides with h on a neighbourhood of 0, whence the proposition.

Loosely speaking, if G and H are identified in a neighbourhood of the identity element with L(G)L(G) and L(H)L(H) by means of exponential mappings, every morphism of G into H is linear in a neighbourhood of 0.

Corollary 1

Let G be a Lie group germ, G' a Lie subgroup germ of G and ϕ\phi an exponential mapping of G.

(i) There exists an open neighbourhood V of 0 in L(G′)L(G') such that ϕ∣V\phi|V is an isomorphism of the manifold V onto an open neighbourhood of e in G'.

(ii) Let x∈L(G)x \in L(G). The following conditions are equivalent: (a) x∈L(G′)x \in L(G'); (b) ϕ(λx)∈G′\phi(\lambda x) \in G' for ∣λ∣|\lambda| sufficiently small.

(i) is obtained by applying Proposition 8 to the canonical injection of G' into G and (ii) follows from (i).

Corollary 2

Let G be a Lie group, ρ\rho an analytic linear representation of G and ϕ\phi an exponential mapping of G. There exists a neighbourhood V of 0 in L(G)L(G) such that

ρ(ϕ(x))=exp⁡(L(ρ)x) \rho(\phi(x)) = \exp(L(\rho)x)
for all x∈Vx \in V.

This follows from Proposition 8 and Example 2 of no. 3.

Corollary 3

Let G be a Lie group and ϕ\phi an exponential mapping of G.

(i) There exists a neighbourhood V of 0 in L(G)L(G) such that

Ad(ϕ(x))=exp⁡ad x. \mathrm{Ad}(\phi(x)) = \exp \mathrm{ad}\ x.
for all x∈Vx \in V.

(ii) If g∈Gg \in G, there exists a neighbourhood W of 0 in L(G)L(G) such that

gϕ(x)g−1=(Ad g.x) g \phi(x) g^{-1} = (\mathrm{Ad}\ g.x)
for all x∈Wx \in W.

(i) follows from Corollary 2 and § 3, no. 12, Proposition 44. (ii) follows from Proposition 8 applied to Int g.

5. Structure induced on a subgroup

Lemma 4

Let G be a finite-dimensional Lie group, Ω\Omega a symmetric open neighbourhood of e in G and H a subset of Ω\Omega containing e such that the conditions x∈H,y∈H,xy−1∈Ωx \in H, y \in H, xy^{-1} \in \Omega imply xy−1∈Hxy^{-1} \in H. Let r∈NKr \in N_K. For all x∈Hx \in H, let hxh_x be the set of a∈Tx(G)a \in T_x(G) with the following property: there exist an open neighbourhood I of 0 in K and a mapping f of class CrC^r of I into G such that f(0)=x,f(I)⊂H,(T0f)(1)=af(0) = x, f(I) \subset H, (T_0f)(1) = a.

(i) Let he=hh_e = h. Then h is a Lie subalgebra of L(G)L(G) which is invariant under AdL(G)(H)\mathrm{Ad}_{L(G)}(H).

(ii) hx=xh=hxh_x = xh = hx for all x∈Hx \in H (where xhxh and hxhx are evaluated in T(G)T(G)).

(iii) Let V be a manifold of class CrC^r, v0v_0 a point of V and f a mapping of class CrC^r of V into G such that f(v0)=ef(v_0) = e and f(V)⊂Hf(V) \subset H. For every Lie subgroup germ H′H' of G with Lie algebra h, f(v)∈H′f(v) \in H' for v sufficiently close to v0v_0.

(iv) For every Lie subgroup germ H′H' of G with Lie algebra h, H′∩HH' \cap H is a neighbourhood of e in H′H'.

(v) For all x∈Hx \in H and all a∈hxa \in h_x, there exist an open neighbourhood I of 0 in K and a mapping f of class CaC^a of I into G such that f(0)=x,f(I)⊂H,(T0f)(1)=af(0) = x, f(I) \subset H, (T_0f)(1) = a. Clearly Kh=hKh = h and xhyz=hxyzxh_yz = h_{xyz} for x,y,xy,xyzx, y, xy, xyz in H. This implies (ii) and the fact that h is invariant under AdL(G)(H)\mathrm{Ad}_{L(G)}(H).

Let a1,a2a_1, a_2 be in h. Let I be an open neighbourhood of 0 in K and f1,f2f_1, f_2 mappings of class CrC^r of I into G such that fj(0)=e,fj(I)⊂H,(T0fj)(1)=ajf_j(0) = e, f_j(I) \subset H, (T_0f_j)(1) = a_j (j=1,2j = 1, 2). We define f:I→Gf : I \to G by f(λ)=f1(λ)f2(λ)f(\lambda) = f_1(\lambda)f_2(\lambda). Then f is of class CrC^r and f(0)=ef(0) = e. By shrinking I if necessary, we have f(I)⊂Hf(I) \subset H. On the other hand, the mapping of Te(G)×Te(G)T_e(G) \times T_e(G) into Te(G)T_e(G) tangent to the mapping (g,g′)→gg′(g, g') \to gg' is addition; hence (T0f)=a1+a2(T_0f) = a_1 + a_2. Hence a1+a2∈ha_1 + a_2 \in h and h is a vector subspace of L(G)L(G). Since xhx−1=hxhx^{-1} = h for all x∈Hx \in H, (Adf1(λ)).a2∈h(\mathrm{Ad}_{f_1(\lambda)}) . a_2 \in h for all λ∈I\lambda \in I. The tangent mapping at 0 to the mapping λ↦Adf1(λ)\lambda \mapsto \mathrm{Ad}f_1(\lambda) is, by Proposition 44 of § 3, no. 12, the mapping λ↦ad(a1λ)\lambda \mapsto \mathrm{ad}(a_1\lambda); hence

[a1,a2]=(ad a1).a2∈h [a_1, a_2] = (\mathrm{ad}\, a_1) . a_2 \in h

since h is closed in L(G)L(G). Hence we have proved (i). In the rest of the proof we fix a Lie subgroup germ H′H' of G with Lie algebra h.

Let V, v0,fv_0, f be as in (iii). Let Y be the left foliation of G associated with h (no. 1). For all y∈H′y \in H', Ty(H′)=yhT_y(H') = yh. On the other hand, for all v∈Vv \in V, the image of Tv(V)T_v(V) under Tv(f)T_v(f) is contained in hf(v)=f(v)hh_{f(v)} = f(v)h (by definition of hf(v)h_{f(v)}). By Differentiable and Analytic Manifolds, R, 9.3.2, f is a morphism of V into Y. As H′H' is a leaf of Y (Differentiable and Analytic Manifolds, R, 9.2.8), f(v)∈H′f(v) \in H' for v sufficiently close to v0v_0.

Let (a1,…,as)(a_1, \ldots, a_s) be a basis of h. There exist an open neighbourhood I of 0 in K and mappings f1,…,fsf_1, \ldots, f_s of class CrC^r of I into G such that fj(0)=e,fj(I)⊂H,(Tfj)1=ajf_j(0) = e, f_j(I) \subset H, (Tf_j)1 = a_j for all j. By (iii), fj(λ)∈H′f_j(\lambda) \in H' for ∣λ∣|\lambda| sufficiently small. Hence the f1(λ1)f2(λ2)…fs(λs)f_1(\lambda_1)f_2(\lambda_2)\ldots f_s(\lambda_s) constitute, for ∣λ1∣,∣λ2∣,…,∣λs∣|\lambda_1|, |\lambda_2|, \ldots, |\lambda_s| sufficiently small, a neighbourhood of ee in H′H'; and this neighbourhood is contained in HH. Hence (iv).

If a∈ha \in \mathfrak{h}, there exist an open neighbourhood II of 00 in KK and a mapping ff of class CωC^\omega of II into GG such that f(0)=e,f(I)⊂H′,(T0f)1=af(0) = e, f(I) \subset H', (T_0 f)1 = a. This, with (iv), implies (v).

Definition 2

h\mathfrak{h} is called the subalgebra tangent to HH at ee.

Proposition 9

Let GG be a finite-dimensional Lie group and HH a subgroup of GG.

(i) There exists on HH one and only one analytic manifold structure with the following property: for all rr between 1 and ω\omega, for every manifold VV of class CrC^r and for every mapping ff of VV into HH, ff is of class CrC^r as a mapping of VV into HH if and only if ff is of class CrC^r as a mapping of VV into GG.

(ii) With this structure, HH is a Lie group, the canonical injection ii of HH into GG is an immersion and L(i)(L(H))L(i)(L(H)) is the Lie subalgebra tangent at ee to HH.

In (i) the uniqueness is obvious. We prove the existence. Let h\mathfrak{h} be the Lie algebra tangent at ee to HH. Let H′H' be a Lie subgroup germ of GG with Lie algebra h\mathfrak{h}. By replacing H′H' by an open subgroup germ of H′H', it can be assumed that H′⊂HH' \subset H (Lemma 4 (iv)). For all x∈Hx \in H, xH′x−1xH'x^{-1} is a Lie subgroup germ of GG with Lie algebra xhx−1=hx\mathfrak{h}x^{-1} = \mathfrak{h}. Hence H′∩(xH′x−1)H' \cap (xH'x^{-1}) is open in H′H' (no. 2, Theorem 3) and the mapping y↦xyx−1y \mapsto xyx^{-1} is an isomorphism of H′∩x−1H′xH' \cap x^{-1}H'x onto xH′x−1∩H′xH'x^{-1} \cap H'. Using Proposition 18 of § 1, no. 9, there exist an open Lie subgroup germ WW of H′H' and a Lie group structure on HH with the following properties: WW is open in HH and the manifold structures on HH and H′H' induce the same structure on WW. It then follows that the canonical injection ii of HH into GG is an immersion and that L(i)(L(H))=L(H′)=hL(i)(L(H)) = L(H') = \mathfrak{h}. Moreover, let VV and ff be as in (i). If f:V→Hf : V \to H is of class CrC^r, i∘f:V→Gi \circ f : V \to G is of class CrC^r. Suppose that i∘f:V→Gi \circ f : V \to G is of class CrC^r; then we prove that f:V→Hf : V \to H is of class CrC^r. By translation, it suffices to consider the case where there exists v0∈Vv_0 \in V such that f(v0)=ef(v_0) = e and to prove that f:V→Hf : V \to H is of class CrC^r in an open neighbourhood of v0v_0. Now, by Lemma 4 (iii), f(v)∈H′f(v) \in H' for vv sufficiently close to v0v_0, whence our assertion. Thus we have proved (i) and (ii) has been obtained on the way.

Definition 3

The Lie group structure on HH defined in Proposition 9 is called the structure induced on HH by the Lie group structure on GG.

If HH is a Lie subgroup of GG, its Lie group structure is induced by that on GG (Differentiable and Analytic Manifolds, R, 5.8.5).

If G=RG = \mathbf{R} and H=QH = \mathbf{Q}, then h={0}\mathfrak{h} = \{0\} and hence the structure induced on HH is the discrete Lie group structure. Similarly if G=CG = \mathbf{C} (considered as a complex Lie group) and H=RH = \mathbf{R}.

6. Primitives of differential forms with values in a Lie algebra

Lemma 5

Let XX be a manifold of class CrC^r, FF and F′F' vector bundles of class CrC^r with base space XX and ϕ\phi a morphism of FF into F′F'. For all x∈Xx \in X, let SxS_x be the set of

(a,ϕ(a))∈Fx⊕Fx′ (a, \phi(a)) \in F_x \oplus F'_x
for a∈Fxa \in F_x. Then the union SS of the SxS_x is a vector subbundle of F⊕F′F \oplus F'.

Let θ\theta and θ′\theta' be the mappings of F⊕F′F \oplus F' into itself defined as follows: if (u,v)∈Fx⊕Fx′(u, v) \in F_x \oplus F'_x, then

θ(u,v)=(u,v+ϕ(u)),θ′(u,v)=(u,v−ϕ(u)). \theta(u, v) = (u, v + \phi(u)), \qquad \theta'(u, v) = (u, v - \phi(u)).
By Differentiable and Analytic Manifolds, R, 7.7.1, θ\theta and θ′\theta' are morphisms of F⊕F′F \oplus F' into itself. Clearly θ∘θ′=θ′∘θ=IdF⊕F′\theta \circ \theta' = \theta' \circ \theta = \mathrm{Id}_{F \oplus F'}. Hence θ\theta and θ′\theta' are automorphisms of F⊕F′F \oplus F'. Therefore, S=θ(F⊕{0})S = \theta(F \oplus \{0\}) is a vector subbundle of F⊕F′F \oplus F'.

Lemma 6

Let GG be a Lie group germ, ω\omega the canonical left differential form of GG (§ 3, no. 18.9), MM a manifold of class CrC^r (r≥2r \geq 2) and α\alpha a differential form of class Cr−1C^{r-1} and degree 1 on MM with values in L(G)L(G).

(i) The elements of T(M×G)T(M \times G) at which the differential form

θ=pr1∗α−pr2∗ω \theta = \mathrm{pr}_1^*\alpha - \mathrm{pr}_2^*\omega
is zero constitute a vector subbundle SS of T(M×G)T(M \times G) of class Cr−1C^{r-1}.

(ii) For all (x,g)∈M×G(x, g) \in M \times G, T(pr1)∣S(x,g)T(\mathrm{pr}_1)|S_{(x, g)} is an isomorphism of S(x,g)S_{(x, g)} onto Tx(M)T_x(M).

(iii) If dα+[α]2=0d\alpha + [\alpha]^2 = 0 (cf. § 3, no. 14) the vector subbundle SS is integrable.

If (x,g)∈M×G(x, g) \in M \times G and (u,v)∈Tx(M)×Tg(G)(u, v) \in T_x(M) \times T_g(G), then

θ(x,g)(u,v)=α(u)−g−1v. \theta_{(x, g)}(u, v) = \alpha(u) - g^{-1}v.
Hence the kernel of θ(x,g)\theta_{(x, g)} is the set S(x,g)S_{(x, g)} of (u,gα(u))(u, g\alpha(u)) for u∈Tx(M)u \in T_x(M), whence (ii). We consider T(M×G)T(M \times G) as the direct sum of two vector bundles FF and F′F' with F(x,g)=Tx(M)×{0}F_{(x, g)} = T_x(M) \times \{0\} and F(x,g)′={0}×Tg(G)F'_{(x, g)} = \{0\} \times T_g(G) for all
(x,g)∈M×G. (x, g) \in M \times G.
For u∈Tx(M)×{0}u \in T_x(M) \times \{0\}, we write ϕ(u)=(0,gα(u))\phi(u) = (0, g\alpha(u)). Using the left trivialization of T(G)T(G), it is seen that ϕ\phi is a morphism of FF into F′F', whence (i) (Lemma 5). Finally, if dα+[α]2=0d\alpha + [\alpha]^2 = 0, then
dθ=pr1∗(dα)−pr2∗(dω)=−12(pr1∗α∧pr1∗α−pr2∗ω∧pr2∗ω)=−12(pr1∗α−pr2∗ω)∧(pr1∗α+pr2∗ω)=−12θ∧(pr1∗α+pr2∗ω) \begin{align*} d\theta &= \mathrm{pr}_1^*(d\alpha) - \mathrm{pr}_2^*(d\omega) \\ &= -\frac{1}{2}(\mathrm{pr}_1^*\alpha \wedge \mathrm{pr}_1^*\alpha - \mathrm{pr}_2^*\omega \wedge \mathrm{pr}_2^*\omega) \\ &= -\frac{1}{2}(\mathrm{pr}_1^*\alpha - \mathrm{pr}_2^*\omega) \wedge (\mathrm{pr}_1^*\alpha + \mathrm{pr}_2^*\omega) \\ &= -\frac{1}{2}\theta \wedge (\mathrm{pr}_1^*\alpha + \mathrm{pr}_2^*\omega) \end{align*}
and hence SS is integrable (Differentiable and Analytic Manifolds, R, 9.3.6).

Theorem 5

Let G be a Lie group germ, M a manifold of class CrC^r (r≥2r \geq 2) and α\alpha a differential form of class Cr−1C^{r-1} and degree 1 on M with values in L(G)L(G), such that dα+[α]2=0d\alpha + [\alpha]^2 = 0. For all x∈Mx \in M and all g∈Gg \in G, there exists a mapping ff, defined and of class Cr−1C^{r-1} on an open neighbourhood of xx, with values in GG, such that f(x)=gf(x) = g and f−1.df=αf^{-1}.df = \alpha. Two mappings which satisfy these conditions coincide on a neighbourhood of xx.

Let x∈Mx \in M and g∈Gg \in G. By Lemma 6 (whose notation we adopt) and Differentiable and Analytic Manifolds, R, 9.3.7, there exist an open neighbourhood U of xx in M and a mapping m↦ϕ(m)=(m,f(m))m \mapsto \phi(m) = (m, f(m)) of class Cr−1C^{r-1} of U into M×GM \times G such that f(x)=gf(x) = g and ϕ∗(0)=0\phi^*(0) = 0. Then

f−1.df=f∗(ω)(§3,no. 18.9)=(pr2∘ϕ)∗(ω)(for f=pr2∘ϕ)=ϕ∗(pr1∗α−θ)(Lemma 6)=ϕ∗(pr1∗α)(for ϕ∗(θ)=0)=α(for pr1∘ϕ=IdU). f^{-1}.df = f^*(\omega) \quad (§ 3, \text{no. 18.9}) \\ = (\mathrm{pr}_2 \circ \phi)^*(\omega) \quad (\text{for } f = \mathrm{pr}_2 \circ \phi) \\ = \phi^*(\mathrm{pr}_1^*\alpha - \theta) \quad (\text{Lemma 6}) \\ = \phi^*(\mathrm{pr}_1^*\alpha) \quad (\text{for } \phi^*(\theta) = 0) \\ = \alpha \quad (\text{for } \mathrm{pr}_1 \circ \phi = \mathrm{Id}_U).

Let f′f' be a mapping of class Cr−1C^{r-1} of U into G such that f′(x)=gf'(x) = g and f′−1df′=α{f'}^{-1}df' = \alpha. By § 3, 18.9, f′−1{f'}^{-1} is locally constant and hence f′=ff' = f in a neighbourhood of xx.

Proposition 10

Let M be an analytic manifold, g\mathfrak{g} a complete normable Lie algebra and α\alpha an analytic differential form of degree 1 on M, with values in g\mathfrak{g}, with the following properties: (a) for all m∈Mm \in M, αm\alpha_m is an isomorphism of Tm(M)T_m(M) onto g\mathfrak{g}; (b) dα+[α]2=0d\alpha + [\alpha]^2 = 0.

Then, for all m0∈Mm_0 \in M, there exist an open neighbourhood M′M' of m0m_0 in M and a Lie group germ structure on M′M', compatible with the manifold structure on M′M', with identity element m0m_0 and with the following properties: (i) αm0\alpha_{m_0} is an isomorphism of L(M′)L(M') onto g\mathfrak{g}; (ii) the differential form m↦αm0−1∘αmm \mapsto \alpha_{m_0}^{-1} \circ \alpha_m is the left canonical differential form of M′M'. If M1′M'_1 and M2′M'_2 are two such group germs, M1′M'_1 and M2′M'_2 have a common open subgroup germ.

There exists a Lie group germ G such that L(G)=gL(G) = \mathfrak{g}. Let m0∈Mm_0 \in M. By Theorem 5, there exist an open neighbourhood M′M' of m0m_0 in M and an analytic mapping ff of M′M' into G such that f(m0)=ef(m_0) = e and f−1.df=αf^{-1}.df = \alpha. Then Tm0(f)=αm0T_{m_0}(f) = \alpha_{m_0} is an isomorphism of Tm0(M)T_{m_0}(M) onto g\mathfrak{g}; hence, shrinking M′M' and G, it can be assumed that ff is an isomorphism of the manifold M′M' onto the manifold G. We transport to M′M' the Lie group germ structure on G by means of f−1f^{-1}. Then Tm0(f)T_{m_0}(f) becomes an isomorphism of L(M′)L(M') onto L(G)=gL(G) = \mathfrak{g}, whence (i). On the other hand, if ω\omega denotes the left canonical differential form of GG, then

αm0−1∘αm=(Tm0f)−1∘(f−1.df)(m)=(Tmf)−1∘ω(f(m))∘Tmf \alpha_{m_0}^{-1} \circ \alpha_m = (\mathrm{T}_{m_0} f)^{-1} \circ (f^{-1}.df)(m) = (\mathrm{T}_m f)^{-1} \circ \omega(f(m)) \circ \mathrm{T}_m f
and hence m↦αm0−1∘αmm \mapsto \alpha_{m_0}^{-1} \circ \alpha_m is the left canonical differential form of M′M'.

Let M′′M'' be an open neighbourhood of m0m_0, with a Lie group germ structure, with identity element m0m_0 and with the analogous properties to properties (i) and (ii). Then αm0\alpha_{m_0} is an isomorphism of L(M′)L(M') onto gg and also of L(M′′)L(M'') onto gg and hence L(M′)=L(M′′)L(M') = L(M''). Therefore, shrinking M′M' and M′′M'', it can be assumed that there exists an isomorphism ϕ\phi of the group germ M′M' onto the group germ M′′M'' (no. 1, Corollary 1 to Theorem 1). Then ϕ−1.dϕ\phi^{-1}.d\phi is the canonical left differential of M′M'. On the other hand, let ψ\psi be the canonical injection of the manifold M′∩M′′M' \cap M'' into the Lie group germ M′′M''; clearly ψ−1.dψ\psi^{-1}.d\psi is a restriction of the canonical left differential of M′′M''. Hence (ψ−1.dψ)(m)=αm0−1∘αm=(ϕ−1.dϕ)(m)(\psi^{-1}.d\psi)(m) = \alpha_{m_0}^{-1} \circ \alpha_m = (\phi^{-1}.d\phi)(m) for all m∈M′∩M′′m \in M' \cap M''. Therefore ϕ\phi and ψ\psi coincide on a neighbourhood of m0m_0 (§ 3, 18.9). This proves the last assertion of the proposition.

Corollary

Let MM be an analytic manifold of finite dimension nn. Let ω1,…,ωn\omega_1, \ldots, \omega_n be analytic differential forms of degree 1 on MM, with scalar values, which are linearly independent at each point of MM and such that, for all k=1,…,nk = 1, \ldots, n, dωkd\omega_k is a linear combination with constant coefficients of the ωi∧ωj\omega_i \wedge \omega_j. Then, for all m0∈Mm_0 \in M, there exists an open neighbourhood M′M' of m0m_0 in MM and a Lie group germ structure on M′M' compatible with the manifold structure on M′M', with identity element m0m_0 and such that ω1∣M′,…,ωn∣M′\omega_1|_{M'}, \ldots, \omega_n|_{M'} form a basis of the space of left invariant differential forms on M′M' of degree 1 with scalar values.

If M1′M'_1 and M2′M'_2 are two such group germs, M1′M'_1 and M2′M'_2 have a common open subgroup germ.

Let X1,…,XnX_1, \ldots, X_n be the vector fields on MM such that, at each point mm of MM, the (Xt)m(X_t)_m constitute the basis of Tm(M)T_m(M) dual to ((ω1)m,…,(ωn)m)((\omega_1)_m, \ldots, (\omega_n)_m). These fields are analytic. By hypothesis, there exist cijk∈Kc_{ijk} \in \mathbf{K} (1≤i,j,k≤n1 \leq i, j, k \leq n) such that cijk=−cjikc_{ijk} = -c_{jik} and dωk=∑i<jcijkωi∧ωjd\omega_k = \sum_{i < j} c_{ijk} \omega_i \wedge \omega_j. By Differentiable and Analytic Manifolds, R, 8.5.7, formula (11),

⟨[Xi,Xj],ωk⟩=−(dωk)(Xi,Xj)=−(∑r<scrskωr∧ωs)(Xi,Xj)=−cijk \langle [X_i, X_j], \omega_k \rangle = -(d\omega_k)(X_i, X_j) = -\left( \sum_{r < s} c_{rsk} \omega_r \wedge \omega_s \right)(X_i, X_j) = -c_{ijk}
and hence [Xi,Xj]=−∑kcijkXk[X_i, X_j] = -\sum_k c_{ijk} X_k. It then follows that the −cijk-c_{ijk} are the constants of structure of a Lie algebra gg relative to a basis (e1,…,en)(e_1, \ldots, e_n). For all m∈Mm \in M, let αm\alpha_m be the linear mapping of Tm(M)T_m(M) into gg which maps (X1)m(X_1)_m to e1,…,(Xn)me_1, \ldots, (X_n)_m to ene_n. Then α\alpha is an analytic differential form on MM of degree 1 with values in gg and αm\alpha_m is an isomorphism of Tm(M)T_m(M) onto gg. On the other hand, α=∑k=1nωkek\alpha = \sum_{k=1}^n \omega_k e_k and hence
dα=∑k=1n(dωk)ek=∑k=1n(∑i<jcijkωi∧ωj)ek d\alpha = \sum_{k=1}^n (d\omega_k)e_k = \sum_{k=1}^n \left( \sum_{i<j} c_{ijk}\omega_i \wedge \omega_j \right)e_k
and
[α]2=∑k=1n[ωkek]2+∑i<j(ωiei)∧(ωjej)(§3,formula (30))=∑i<j(ωi∧ωj)[ei,ej]=−∑k=1n∑i<j(cijkωi∧ωj)ek=−dα. \begin{align*} [\alpha]^2 &= \sum_{k=1}^n [\omega_k e_k]^2 + \sum_{i<j} (\omega_i e_i) \wedge (\omega_j e_j) \quad (§ 3, \text{formula (30)}) \\ &= \sum_{i<j} (\omega_i \wedge \omega_j)[e_i, e_j] \\ &= -\sum_{k=1}^n \sum_{i<j} (c_{ijk}\omega_i \wedge \omega_j)e_k \\ &= -d\alpha. \end{align*}
It then suffices to apply Proposition 10.

7. Passage from laws of infinitesimal operation to laws of operation

Proposition 11

Let G1G_1 and G2G_2 be Lie group germs and X1X_1 and X2X_2 manifolds of class CrC^r (r≥2r \geq 2). For i=1,2i = 1, 2, let ψi\psi_i be a law chunk of left operation of class CrC^r of GiG_i on XiX_i and DiD_i the associated law of infinitesimal operation. Let μ:G1→G2\mu : G_1 \to G_2 be a morphism and ϕ:X1→X2\phi : X_1 \to X_2 a mapping of class CrC^r. Suppose that, for all a∈L(G)a \in L(G), the vector fields (D1)a(D_1)_a and (D2)L(μ)a(D_2)_{L(\mu)a} are ϕ\phi-related (Differentiable and Analytic Manifolds, R, 8.2.6). Then there exists a neighbourhood Ω\Omega of {e}×X1\{e\} \times X_1 in G1×X1G_1 \times X_1 such that ϕ(ψ1(g,x))=ψ2(μ(g),ϕ(x))\phi(\psi_1(g, x)) = \psi_2(\mu(g), \phi(x)) for all (g,x)∈Ω(g, x) \in \Omega.

Let p1:G1×X2→G1,p2:G1×X2→X2p_1 : G_1 \times X_2 \to G_1, p_2 : G_1 \times X_2 \to X_2 be the canonical projections. For all (g1,x2)∈G1×X2(g_1, x_2) \in G_1 \times X_2, let fg1,x2f_{g_1, x_2} be the mapping g1a↦(D2)L(μ)a(x2)g_1 a \mapsto (D_2)_{L(\mu)a}(x_2) of Tg1(G1)=g1L(G1)T_{g_1}(G_1) = g_1 L(G_1) into Tx2(X2)T_{x_2}(X_2). The fg1,x2f_{g_1, x_2} define a morphism of p1∗T(G1)p_1^*T(G_1) into p2∗T(X2)p_2^*T(X_2).

Let x0∈X1x_0 \in X_1. There exist an open neighbourhood GG of ee in G1G_1 and an open neighbourhood XX of x0x_0 in X1X_1 such that ψ1(g,x)\psi_1(g, x) and ψ2(μ(g),ϕ(x))\psi_2(\mu(g), \phi(x)) are defined for (g,x)∈G×X(g, x) \in G \times X. We write, for (g,x)∈G×X(g, x) \in G \times X,

α(g,x)=ϕ(ψ1(g,x))∈X2,β(g,x)=ψ2(μ(g),ϕ(x))∈X2. \alpha(g, x) = \phi(\psi_1(g, x)) \in X_2, \qquad \beta(g, x) = \psi_2(\mu(g), \phi(x)) \in X_2.
If GG and XX are sufficiently small, then, for all (a,g,x)∈L(G1)×G×X(a, g, x) \in L(G_1) \times G \times X,
(Tα)(ag,0x)=(Tϕ)((D1)a(ψ1(g,x)))=(D2)L(μ)a(ϕ(ψ1(g,x)))=(D2)L(μ)aα(g,x),(Tβ)(ag,0x)=(Tψ2)(L(μ)a.μ(g),ϕ(x))=(D2)L(μ)a(ψ2(μ(g),ϕ(x)))=(D2)L(μ)aβ(g,x). \begin{align*} (T\alpha)(ag, 0_x) &= (T\phi)((D_1)_a(\psi_1(g, x))) \\ &= (D_2)_{L(\mu)a}(\phi(\psi_1(g, x))) \\ &= (D_2)_{L(\mu)a}\alpha(g, x), \\ (T\beta)(ag, 0_x) &= (T\psi_2)(L(\mu)a.\mu(g), \phi(x)) \\ &= (D_2)_{L(\mu)a}(\psi_2(\mu(g), \phi(x))) \\ &= (D_2)_{L(\mu)a}\beta(g, x). \end{align*}

Hence for x∈Xx \in X, the morphisms g↦α(g,x)g \mapsto \alpha(g, x) and g↦β(g,x)g \mapsto \beta(g, x) are integrals of ff; as

β(e,x)=ϕ(x)=α(e,x) \beta(e, x) = \phi(x) = \alpha(e, x)
for all x∈Xx \in X, it follows from Differentiable and Analytic Manifolds, R, 9.3.7, that α\alpha and β\beta coincide on a neighbourhood of (e,x0)(e, x_0). Hence the proposition.

Corollary

Let G be a Lie group germ and X a manifold of class CrC^r. Consider two law chunks of left operation of class CrC^r of G on X. Suppose that, for all a∈L(G)a \in L(G), the corresponding vector field DaD_a on X is the same for both law chunks. Then these two law chunks coincide on a neighbourhood of {e}×X\{e\} \times X.

Theorem 6

Let G be a Lie group germ, X a manifold of class CrC^r (r≥2r \geq 2) and x0x_0 a point of X. Let a↦Daa \mapsto D_a be a law of left infinitesimal operation of class Cr−1C^{r-1} of L(G)L(G) on X.

(i) There exists an open neighbourhood X′X' of x0x_0 in X and a law chunk of left operation of class Cr−1C^{r-1} of G on X′X' such that the associated law of infinitesimal operation is a↦Da∣X′a \mapsto D_a|X'.

(ii) Let there be two law chunks of left operation of class Cr−1C^{r-1} of G on an open neighbourhood X′′X'' of x0x_0; if they admit a↦Da∣X′′a \mapsto D_a|X'' as associated law of infinitesimal operation, they coincide on a neighbourhood of (e,x0)(e, x_0).

Assertion (ii) follows from the Corollary to Proposition 11. We prove (i). For all (g,x)∈G×X(g, x) \in G \times X and all a∈L(G)a \in L(G), we write

Qa(g,x)=(ag,Da(x))∈Tg(G)×Tx(X). Q_a(g, x) = (ag, D_a(x)) \in T_g(G) \times T_x(X).
Let S(g,x)S_{(g, x)} be the set of Qa(g,x)Q_a(g, x) for a∈L(G)a \in L(G). By Lemma 5 of no. 6, the S(g,x)S_{(g, x)} are the fibres of a vector subbundle S of T(G)×T(X)T(G) \times T(X). Let a,ba, b be in L(G)L(G); then
[Qa,Qb](g,x)=([Ra,Rb](g),[Da,Db](x))=(−R[a,b](g),−D[a,b](x))(§3,18.6)=Q−[a,b](g,x) \begin{align*} [Q_a, Q_b](g, x) &= ([R_a, R_b](g), [D_a, D_b](x)) \\ &= (-R_{[a, b]}(g), -D_{[a, b]}(x)) \quad (§ 3, 18.6) \\ &= Q_{-[a, b]}(g, x) \end{align*}
and hence S is integrable (Differentiable and Analytic Manifolds, R, 9.3.3, (iv)).

By Differentiable and Analytic Manifolds, R, 9.3.7, there exist an open neighbourhood G1G_1 of e in G, an open neighbourhood X1X_1 of x0x_0 in X and a mapping (g,x)↦gx(g, x) \mapsto gx of class Cr−1C^{r-1} of G1×X1G_1 \times X_1 into X such that ex=xex = x for all x∈X1x \in X_1 and

(ag)x=Da(gx)for a∈L(G),g∈G1,x∈X1. (ag)x = D_a(gx) \quad \text{for } a \in L(G), g \in G_1, x \in X_1.
In particular
ax=Da(x). ax = D_a(x).
Let G2G_2 be an open neighbourhood of e in G1G_1 and X2X_2 an open neighbourhood of x0x_0 in X1X_1 such that gg′gg' is defined and belongs to G1G_1 for g,g′g, g' in G2G_2 and gxgx is defined and belongs to X1X_1 for (g,x)∈G2×X2(g, x) \in G_2 \times X_2. Consider the mappings α1,α2\alpha_1, \alpha_2 of G2×(G2×X2)G_2 \times (G_2 \times X_2) into XX defined by
α1(g,(h,x))=g(hx),α2(g,(h,x))=(gh)x. \alpha_1(g, (h, x)) = g(hx), \quad \alpha_2(g, (h, x)) = (gh)x.
They are of class Cr−1C^{r-1}. Then
α1(e,(h,x))=hx=α2(e,(h,x)). \alpha_1(e, (h, x)) = hx = \alpha_2(e, (h, x)).
On the other hand
T(α1)(ag,0(h,x))=(ag)(hx)=Da(g(hx))by (6)=Da(α1(g,(h,x))),T(α2)(ag,0(h,x))=(agh)x=Da((gh)x)by (6)=Da(α2(g,(h,x))). \begin{align*} T(\alpha_1)(ag, 0_{(h, x)}) &= (ag)(hx) \\ &= D_a(g(hx)) \qquad \text{by (6)} \\ &= D_a(\alpha_1(g, (h, x))), \\ T(\alpha_2)(ag, 0_{(h, x)}) &= (agh)x \\ &= D_a((gh)x) \qquad \text{by (6)} \\ &= D_a(\alpha_2(g, (h, x))). \end{align*}
By Differentiable and Analytic Manifolds, R, 9.3.7, α1\alpha_1 and α2\alpha_2 coincide on a neighbourhood of (e,(e,x0))(e, (e, x_0)). Then (i) follows from (7) and Proposition 23 of § 1, no. 11.

Corollary 1

Let GG be a Lie group germ and XX a paracompact manifold of class CrC^r (r≥2r \geq 2). Let a↦Daa \mapsto D_a be a law of left infinitesimal operation of class Cr−1C^{r-1} of L(G)L(G) on XX.

(i) There exists a law chunk of left operation of class Cr−1C^{r-1} of GG on XX such that the associated law of infinitesimal operation is a↦Daa \mapsto D_a.

(ii) Two laws of left operation of class Cr−1C^{r-1} of GG on XX which admit a↦Daa \mapsto D_a as associated law of infinitesimal operation coincide on a neighbourhood of {e}×X\{e\} \times X.

Assertion (ii) follows from the Corollary to Proposition 11. By Theorem 6 (i), there exist an open covering (Xi)i∈I(X_i)_{i \in I} of XX and, for all i∈Ii \in I, a law chunk of left operation ψi\psi_i of Class Cr−1C^{r-1} of GG on XiX_i such that the associated law of infinitesimal operation is a↦Da∣Xia \mapsto D_a|_{X_i}. As XX is paracompact, the covering (Xi)i∈I(X_i)_{i \in I} can be assumed to be locally finite. For all (i,j)∈I×I(i, j) \in I \times I and all x∈Xi∩Xjx \in X_i \cap X_j, ψi\psi_i and ψj\psi_j coincide on a neighbourhood of (e,x)(e, x) (Corollary to Proposition 11). As XX is normal, we can apply Proposition 24 of § 1, no. 11, which proves (i).

Corollary 2

Let XX be a paracompact manifold of class CrC^r (r≥2r \geq 2) and ξ\xi a vector field of class Cr−1C^{r-1} on XX. There exists a law chunk of operation ψ\psi of class Cr−1C^{r-1} of KK on XX such that for all x∈Xx \in X, ξ(x)\xi(x) is the image under t↦(t,x)t \mapsto (t, x) of the tangent vector 1 to KK at 0. Two law chunks of operation with the above property coincide on a neighbourhood of {0}×X\{0\} \times X.

This is a special case of Corollary 1.

§ 5.1

Remark

Laws of left operation can of course be replaced, throughout this no., by laws of right operation.

Exercises

See the exercises for § 4.

Transcribed from the printed volume. Not translated.