Bourbaki

§ 1. Lie groups

1. Definition of a Lie group

Let G be a set. A group structure and an analytic K-manifold structure on G are called compatible if the following condition holds: (GL) The mapping (g,h)↦gh−1(g, h) \mapsto gh^{-1} of G×GG \times G into G is analytic.

Definition 1

A Lie group over K is a set G with a group structure and an analytic K-manifold structure such that these two structures are compatible.

A Lie group over R\mathbf{R} (resp. C\mathbf{C}, Qp\mathbf{Q}_p) is called a real (resp. complex, pp-adic) Lie group.

Let G be a group with an analytic manifold structure. For g,h,g0,h0g, h, g_0, h_0 in G,

gh−1=(g0h0−1)h0((g0−1g)(h0−1h)−1)h0−1. gh^{-1} = (g_0 h_0^{-1}) h_0 ((g_0^{-1} g)(h_0^{-1} h)^{-1}) h_0^{-1}.
It follows that G is a Lie group if and only if the following three conditions hold:

(GL1_1) for all g0∈Gg_0 \in G, the mapping g↦g0gg \mapsto g_0 g of G into G is analytic; (GL2_2) for all g0∈Gg_0 \in G, the mapping g↦g0gg0−1g \mapsto g_0 g g_0^{-1} of G into G is analytic in an open neighbourhood of ee; (GL3_3) the mapping (g,h)↦gh−1(g, h) \mapsto gh^{-1} of G×GG \times G into G is analytic in an open neighbourhood of (e,e)(e, e).

Let G be a Lie group. For all g∈Gg \in G, γ(g)\gamma(g) and δ(g)\delta(g) are automorphisms of the underlying manifold of G. It follows that this manifold is pure (Differentiable and Analytic Manifolds, R, 5.1.7). In particular, the dimension of G at gg is equal to dim⁡G\dim G for all g∈Gg \in G (recall that dim⁡G\dim G is an integer ⩾0\geqslant 0 or +∞+\infty).

Since an analytic mapping is continuous, a Lie group is a topological group with the underlying topology of its manifold structure. Let G be a set. A topological group structure and an analytic K-manifold structure on G are called compatible if the group structure and the manifold structure are compatible and the topology on G is the topology underlying the manifold structure.

Lemma 1

Let G be a Lie group, U an open neighbourhood of e, E a complete normed space and ϕ:U→E\phi : U \to E a chart of the manifold G. There exists a neighbourhood W of e contained in U such that ϕ∣W\phi | W is an isomorphism of W (with the right uniform structure) onto ϕ(W)\phi(W) (with the uniform structure induced by that on E).

It can be assumed that ϕ(e)=0\phi(e) = 0. Let U′=ϕ(U)U' = \phi(U). Let ψ:U′→U\psi : U' \to U be the inverse mapping of ϕ\phi. Let V be a symmetric open neighbourhood of e such that V2⊂UV^2 \subset U and let V′=ϕ(V)V' = \phi(V). We define mappings θ1,θ2\theta_1, \theta_2 of V′×V′V' \times V' into V′×U′V' \times U' as follows:

θ1(x,y)=(x,ϕ(ψ(x)ψ(y)−1))θ2(x,y)=(x,ϕ(ψ(y)−1ψ(x))). \begin{align*} \theta_1(x, y) &= (x, \phi(\psi(x)\psi(y)^{-1})) \\ \theta_2(x, y) &= (x, \phi(\psi(y)^{-1}\psi(x))). \end{align*}
It is immediately verified that θ2(θ1(x,y))=θ1(θ2(x,y))=(x,y)\theta_2(\theta_1(x, y)) = \theta_1(\theta_2(x, y)) = (x, y) for x,yx, y sufficiently close to 0. On the other hand, θ1\theta_1 and θ2\theta_2 are analytic and hence strictly differentiable at (0,0)(0, 0). Therefore (Differentiable and Analytic Manifolds, R, 1.2.2) there exist a neighbourhood W′W' of 0 in V′V' and constants a>0,b>0a > 0, b > 0 such that
a(∥x1−x2∥+∥ϕ(ψ(x1)ψ(y1)−1)−ϕ(ψ(x2)ψ(y2)−1)∥)≤∥x1−x2∥+∥y1−y2∥≤b(∥x1−x2∥+∥ϕ(ψ(x1)ψ(y1)−1)−ϕ(ψ(x2)ψ(y2)−1)∥) a(\|x_1 - x_2\| + \|\phi(\psi(x_1)\psi(y_1)^{-1}) - \phi(\psi(x_2)\psi(y_2)^{-1})\|) \leq \|x_1 - x_2\| + \|y_1 - y_2\| \leq b(\|x_1 - x_2\| + \|\phi(\psi(x_1)\psi(y_1)^{-1}) - \phi(\psi(x_2)\psi(y_2)^{-1})\|)

for all x1,x2,y1,y2x_1, x_2, y_1, y_2 in W′W'. Writing x1=x2=y2x_1 = x_2 = y_2, we obtain

(2)a∥ϕ(ψ(x1)ψ(y1)−1)∥≤∥x1−y1∥≤b∥ϕ(ψ(x1)ψ(y1)−1)∥. (2) \quad a \| \phi(\psi(x_1)\psi(y_1)^{-1}) \| \leq \| x_1 - y_1 \| \leq b \| \phi(\psi(x_1)\psi(y_1)^{-1}) \|.

For δ>0\delta > 0, let NδN_\delta be the set of ordered pairs (x,y)∈W′×W′(x, y) \in W' \times W' such that ∥x−y∥≤δ\| x - y \| \leq \delta. The NδN_\delta form a fundamental system of entourages in W′W'. We write W=ψ(W′)W = \psi(W'). Let MδM_\delta be the set of ordered pairs (u,v)∈W×W(u, v) \in W \times W such that ∥ϕ(uv−1)∥≤δ\| \phi(uv^{-1}) \| \leq \delta. The MδM_\delta form a fundamental system of entourages in WW with the right uniform structure. But relation (2) proves that

Nδ⊂(ϕ×ϕ)(Ma−1δ),(ϕ×ϕ)(Mδ)⊂Nbδ N_\delta \subset (\phi \times \phi)(M_{a^{-1}\delta}), \qquad (\phi \times \phi)(M_\delta) \subset N_{b\delta}

and hence WW has the property of the lemma.

Proposition 1

A Lie group is a complete metrizable topological group.

Since ee admits an open neighbourhood homeomorphic to an open ball of a normed space, ee admits a countable fundamental system of neighbourhoods whose intersection is {e}\{e\}. Hence GG is metrizable (General Topology, Chapter III, § 1, Corollary to Proposition 2 and Chapter IX, § 3, Proposition 1). By Lemma 1, there exists a neighbourhood of ee which is complete under the right uniform structure and hence GG is complete (General Topology, Chapter III, § 3, Proposition 4).

Proposition 2

Let GG be a Lie group.* (i) If K=RK = \mathbf{R} or C\mathbf{C}, GG is locally connected. (ii) If KK is distinct from R\mathbf{R} and C\mathbf{C}, GG is zero-dimensional (General Topology, Chapter IX, § 6, Definition 5). (iii) Suppose that KK is locally compact. For GG to be locally compact, it is necessary and sufficient that GG be finite-dimensional. (iv) *If GG is generated by a subspace whose topology admits a countable base, then the topology on GG admits a countable base.

Let UU be a neighbourhood of ee. There exists an open neighbourhood U1U_1 of ee contained in UU and homeomorphic to an open ball of a normed space EE over KK. If K=RK = \mathbf{R} or C\mathbf{C}, U1U_1 is connected, which proves (i). Suppose that KK is ultrametric. There exists a neighbourhood U2U_2 of ee which is closed in GG and such that U2⊂U1U_2 \subset U_1. Then there exists a neighbourhood U3U_3 of ee such that U3⊂U2U_3 \subset U_2 and U3U_3 is open and closed relative to U1U_1. Then U3U_3 is closed relative to U2U_2 and hence GG and open relative to U1U_1 and hence GG. This proves (ii). For GG to be locally compact, it is necessary and sufficient that EE be locally compact; if KK is locally compact, this amounts to saying that EE is finite-dimensional (Topological Vector Spaces, Chapter I, § 2, Theorem 3), whence (iii). Suppose that GG is generated by a subset VV and let W=V∪V−1W = V \cup V^{-1}; then

G=W∪W2∪W3∪…; G = W \cup W^2 \cup W^3 \cup \ldots;

if there exists a sequence dense in VV, we see that there exists a sequence dense in G and, as G is metrizable (Proposition 1), the topology on G admits a countable base.

Corollary

If K = R or C and G is connected and finite-dimensional, then G is locally connected and locally compact and its topology admits a countable base.

Lemma 2

Let X be a manifold of class C^r, e a point of X, U and V open neighbourhoods of e and m a mapping of class C^r of U × U into X satisfying the following conditions: (a) m(e, x) = m(x, e) = x for all x ∈ U; (b) V ⊂ U, m(V × V) ⊂ U and m(m(x, y), z) = m(x, m(y, z)) for all x, y, z in V.

Then there exists an open neighbourhood W of e in V and an automorphism θ of the manifold W such that θ(e) = e, θ(θ(x)) = x and m(x, θ(x)) = m(θ(x), x) = e for all x ∈ W.

m(e, y) = y for all y ∈ U and hence, by the implicit function theorem, there exists an open neighbourhood W_1 of e in V and a mapping θ_1 of class C^r of W_1 into V such that θ_1(e) = e, m(x, θ_1(x)) = e for all x ∈ W_1. Similarly, there exists an open neighbourhood W_2 of e in V and a mapping θ_2 of class C^r of W_2 into V such that θ_2(e) = e, m(θ_2(x), x) = e for all x ∈ W_2. For x ∈ W_1 ∩ W_2,

θ2(x)=m(θ2(x),e)=m(θ2(x),m(x,θ1(x)))=m(m(θ2(x),x),θ1(x))=m(e,θ1(x))=θ1(x). \begin{align*} \theta_2(x) &= m(\theta_2(x), e) = m(\theta_2(x), m(x, \theta_1(x))) \\ &= m(m(\theta_2(x), x), \theta_1(x)) = m(e, \theta_1(x)) = \theta_1(x). \end{align*}

Let θ(x) be the common value of θ_1(x) and θ_2(x) for x ∈ W_1 ∩ W_2. Let W be the set of x ∈ W_1 ∩ W_2 such that θ(x) ∈ W_1 ∩ W_2. The set W is open. For x ∈ W,

θ(θ(x))=m(m(x,θ(x)),θ(θ(x)))=m(x,m(θ(x),θ(θ(x))))=m(x,e)=x \theta(\theta(x)) = m(m(x, \theta(x)), \theta(\theta(x))) = m(x, m(\theta(x), \theta(\theta(x)))) = m(x, e) = x

and hence θ(x) ∈ W. We see that θ | W defines an automorphism of the manifold W.

Proposition 3

Let X be an analytic manifold and m an analytic associative law of composition on X admitting an identity element. The set G of invertible elements of X is open in X and G is a Lie group with m | (G × G) and the manifold structure induced by that on X.

By Lemma 2, G is a neighbourhood of the identity element. For all g ∈ G, the mapping x ↦ m(g, x) is an automorphism of the manifold X. Hence the image of G under this mapping is a neighbourhood of g, obviously contained in G. Therefore G is open in X. Clearly conditions (GL_1) and (GL_2) hold. Condition (GL_3) holds by Lemma 2.

Examples of Lie groups

(1) Let E be a complete normable space over K. The mapping (x, y) ↦ x − y of E × E into E is continuous and linear and hence analytic.

Hence E, with its additive group and analytic manifold structures, is a Lie group.

In particular, K is a Lie group.

(2) Let A be a complete normable unital associative algebra over K. The multiplication (x,y)↦xy(x, y) \mapsto xy of A×AA \times A into A is bilinear and continuous and hence analytic. Proposition 3 shows that the group A∗A^* of invertible elements of A is open in A (which also follows from General Topology, Chapter IX, § 3, Proposition 13) and that A∗A^* is a Lie group.

For example, let E be a complete normable space over K and let A=L(E)A = \mathscr{L}(E) (General Topology, Chapter IX, § 3, Proposition 5). Then A∗A^* is the automorphism group GL(E)\mathbf{GL}(E) of E. This group therefore has canonically a Lie group structure over K. More particularly, GL(n,K)\mathbf{GL}(n, K), with the manifold structure induced by that on Mn(K)\mathbf{M}_n(K), is a Lie group. For n=1n = 1, we see that the multiplicative group K∗K^* is a Lie group with the manifold structure induced by that on K.

(3) Let G be a Lie group over K. Let K′=RK' = \mathbf{R} or C\mathbf{C} or a non-discrete complete ultrametric field and σ\sigma an isomorphism of the valued field K′K' onto a valued subfield of K. Then the group G, with the K′K'-manifold structure obtained by restriction of scalars, is a Lie group over K′K', which is said to be derived from the Lie group G by restriction of scalars (from K to K′K' by means of σ\sigma). For example, every complex Lie group has canonically a real Lie group structure. Again, with every complex Lie group G is associated a complex Lie group called the conjugate of G, derived from G by means of the automorphism z↦z‾z \mapsto \overline{z} of C\mathbf{C}.

2. Morphisms of Lie groups

Definition 2

Let G and H be Lie groups. A Lie group morphism of G into H (or simply a morphism of G into H if no confusion can arise) is a mapping of G into H which is a group homomorphism and is analytic. The automorphism group of G is denoted by Aut⁡(G)\operatorname{Aut}(G).*.

The identity mapping of G is a morphism. The composition of two morphisms is a morphism. If f:G→Hf : G \to H and f′:H→Gf' : H \to G are two inverse morphisms, ff and f′f' are Lie group isomorphisms.

Example

(1) Let G be a Lie group. For all x∈Gx \in G, Int⁡(x)\operatorname{Int}(x) is an automorphism of the Lie group G.

(2) Let G be a Lie group. Let G∨G^\vee denote the opposite group to G, with the same manifold structure as G. It is immediate that G∨G^\vee is a Lie group (called the opposite Lie group to G) and that the mapping g↦g−1g \mapsto g^{-1} is an isomorphism of the Lie group G onto the Lie group G∨G^\vee.

(3) Let G be a Lie group and E a complete normable space. An analytic linear representation of G on E (or simply a linear representation of G on E when no confusion can arise) is a morphism of the Lie group G into the Lie group GL(E)\mathbf{GL}(E), in other words an analytic mapping π\pi of G into GL(E)\mathbf{GL}(E) such that π(gg′)=π(g)π(g′)\pi(gg') = \pi(g)\pi(g') for g,g′g, g' in G. Suppose that E admits a finite basis (e1,e2,…,en)(e_1, e_2, \ldots, e_n) over K; let (e1∗,e2∗,…,en∗)(e_1^*, e_2^*, \ldots, e_n^*) be the dual basis; let ρ\rho be a homomorphism of the group G into the group GL(E)\mathbf{GL}(E); then the following conditions are equivalent: (i) ρ\rho is an analytic linear representation; (ii) for all x∈Ex \in E and x′∈E′x' \in E', the function g↦⟨ρ(g)x,x′⟩g \mapsto \langle \rho(g)x, x' \rangle on G is analytic; (iii) for all i and j, the function g↦⟨ρ(g)ei,ej∗⟩g \mapsto \langle \rho(g)e_i, e_j^* \rangle on G is analytic. For the implications (i) ⇒\Rightarrow (ii) ⇒\Rightarrow (iii) are clear. On the other hand, the functions u↦⟨uei,ej∗⟩u \mapsto \langle ue_i, e_j^* \rangle form a coordinate system on L(E)\mathscr{L}(E); hence their restrictions to GL(E)\mathbf{GL}(E) form a coordinate system on GL(E)\mathbf{GL}(E), whence the implication (iii) ⇒\Rightarrow (i).

Let G be a real Lie group, E a real complete normable space and ρ\rho a homomorphism of the group G into the group GL(E)\mathbf{GL}(E). We shall see in § 8, Theorem 1 that, if ρ\rho is continuous (when GL(E)\mathbf{GL}(E) has the topology derived from the norm on L(E)\mathbf{L}(E)), then ρ\rho is analytic. But note that this notion of continuity is different from that considered in Integration, Chapter VIII, § 2, Definition 1 (iii) (Exercise 1).

(4) Let G be a real Lie group and E a complex complete normable space. An analytic linear representation of G on E is a morphism of G into the underlying real Lie group of GL(E)\mathbf{GL}(E).

Proposition 4

Let G and H be Lie groups and f a homomorphism of the group G into the group H. For f to be analytic, it is necessary and sufficient that there exist a non-empty open subset U of G such that f∣Uf|_U is analytic.

The condition is obviously necessary. Suppose that it holds. For all x0∈Gx_0 \in G, f(x0x)=f(x0)f(x)f(x_0x) = f(x_0)f(x) for all x∈Ux \in U and hence f∣x0Uf|_{x_0U} is analytic. But the sets x0Ux_0U, where x0∈Gx_0 \in G, form an open covering of G.

Remark

If f is an immersion at e (resp. a submersion at e), clearly f is an immersion (resp. a submersion).

3. Lie subgroups

Let G be a Lie group and H a subgroup of G which is at the same time a submanifold of G. Then the mapping (x,y)↦xy−1(x, y) \mapsto xy^{-1} of H×HH \times H into G is analytic and hence the mapping (x,y)↦xy−1(x, y) \mapsto xy^{-1} of H×HH \times H into H is analytic. (Differentiable and Analytic Manifolds, R, 5.8.5). Thus H, with the group and manifold structures induced by those on G, is a Lie group.

Definition 3

Let G be a Lie group. A subset H of G is called a Lie subgroup if H is a subgroup and a submanifold of G.

An open subgroup of G is a Lie subgroup of G. In particular, if G is a real or complex Lie group, its identity component is a Lie subgroup of G.

Proposition 5

Let G be a Lie group and H a Lie subgroup of G. (i) H is closed in G. (ii) The canonical injection of H into G is a Lie group morphism. (iii) Let L be a Lie group and f a mapping of L into G such that f(L) ⊂ H. For f to be a morphism of L into H, it is necessary and sufficient that f be a morphism of L into G.

By Differentiable and Analytic Manifolds, R, 5.8.3, H is locally closed. Hence H is closed (General Topology, Chapter III, § 2, Proposition 4). Assertion (ii) is obvious. Assertion (iii) follows from Differentiable and Analytic Manifolds, R, 5.8.5.

Proposition 6

Let G be a Lie group and H a subgroup of G. For H to be a Lie subgroup of G, it is necessary and sufficient that there exist a point h ∈ H and an open neighbourhood U of h in G such that H ∩ U is a submanifold of G.

The condition is obviously necessary. Suppose that it holds. For all h' ∈ H, the translation γ(h'h^{-1}) is an automorphism of the manifold G and maps the submanifold H ∩ U of U into the submanifold (h'h^{-1}H) ∩ (h'h^{-1}U) of h'h^{-1}U. As h'h^{-1}H = H and h'h^{-1}U is an open neighbourhood of h' in G, we see that every point of H has an open neighbourhood V such that V ∩ H is a submanifold of G. Hence H is a submanifold of G.

Let G be a Lie group and H a Lie subgroup of G. If L is a Lie subgroup of H, L is a Lie subgroup of G by Differentiable and Analytic Manifolds, R, 5.8.6. Let M be a Lie subgroup of G such that M ⊂ H. Then M is a Lie subgroup of H, for the canonical injection of M into H is obviously an immersion.

Let k be a non-discrete closed subfield of K. A Lie k-subgroup of G is a Lie subgroup of the underlying Lie k-group of G.

Remark

If "submanifold" is replaced by "quasi-submanifold" in Definition 3, we obtain the definition of Lie quasi-subgroups of G. (For finite-dimensional G, the Lie quasi-subgroups are just the Lie subgroups.) Suppose that K is of characteristic 0. Proposition 5 remains valid with the same proof for Lie quasi-subgroups. Proposition 6 remains valid with the same proof, replacing "Lie subgroup" by "Lie quasi-subgroup" and "submanifold" by "quasi-submanifold".

4. Semi-direct products of Lie groups

Let I be a finite set and (L_i)_{i \in I} a family of Lie groups. The group and manifold structures on L = \prod_{i \in I} L_i are compatible and L thus has a Lie group structure. L is called the product Lie group of the family of Lie groups (L_i)_{i \in I}.

Let L and M be Lie groups and σ a homomorphism of L into the automorphism group of the group M. Let S be the external semi-direct product of L by M relative to σ (Algebra, Chapter I, § 6, no. 1, Definition 2).

Proposition 7

If the mapping* (m,l)↦σ(l)m(m, l) \mapsto \sigma(l)m of M×LM \times L into MM is analytic, the group SS, with the product manifold structure of MM and LL, is a Lie group.

For l,l′l, l' in LL and m,m′m, m' in MM,

(m,l)(m′,l′)−1=mll′−1m′−1=m(σ(ll′−1)m′−1)ll′−1=(m(σ(ll′−1)m′−1),ll′−1) (m, l)(m', l')^{-1} = ml{l'}^{-1}{m'}^{-1} = m(\sigma(l{l'}^{-1}){m'}^{-1})l{l'}^{-1} = (m(\sigma(l{l'}^{-1}){m'}^{-1}), l{l'}^{-1})

whence the proposition.

If the conditions of Proposition 7 hold, the Lie group SS is called the (external) semi-direct product Lie group of LL by MM relative to σ\sigma.

Clearly the canonical injection of LL (resp. MM) into SS is an isomorphism of LL (resp. MM) onto a Lie subgroup of SS which we identify with LL (resp. MM). The canonical mapping of SS onto LL is a Lie group morphism.

Conversely, let GG be a Lie group and LL, MM two Lie subgroups such that the group GG is (algebraically) the semi-direct product of LL by MM (Algebra, Chapter I, § 6, no. 1). We write σ(l)m=lml−1\sigma(l)m = lml^{-1} for l∈Ll \in L and m∈Mm \in M. Then σ\sigma satisfies the conditions of Proposition 7. We can therefore form the semi-direct product Lie group SS of LL by MM relative to σ\sigma. The mapping j:(m,l)↦mlj: (m, l) \mapsto ml of SS onto GG is a group isomorphism and is analytic. If jj is a Lie group isomorphism, the Lie group GG is called the (internal) semi-direct product of LL by MM and SS and GG are identified. For all g∈Gg \in G, we write g=p(g)q(g)g = p(g)q(g), where p(g)∈Mp(g) \in M and q(g)∈Lq(g) \in L. For the Lie group GG to be the semi-direct product of LL by MM, it is necessary and sufficient that one of the mappings p:G→Mp: G \to M and q:G→Lq: G \to L be analytic, in which case both are analytic; or alternatively, it is necessary and sufficient that Te(G)T_e(G) be the topological direct sum of Te(M)T_e(M) and Te(L)T_e(L) (for, if this condition holds, jj is étale at eSe_S).

Example

Let EE be a normable space, G=GL(E)G = \mathbf{GL}(E), TT the translation group of EE and AA the permutation group of EE generated by GG and TT. The group AA is algebraically the semi-direct product of GG by TT. (If EE is finite-dimensional, AA is the affine group of EE, cf. Algebra, Chapter II, § 9, no. 4). Let σ\sigma be the identity linear representation of GG on EE and SS the external semi-direct product of GG by EE relative to σ\sigma. For all x∈Ex \in E, let txt_x be the translation of EE defined by xx. The mapping (x,u)↦tx∘u(x, u) \mapsto t_x \circ u is isomorphism Φ\Phi of the group SS onto the group AA. The mapping (x,u)↦σ(u)x=u(x)(x, u) \mapsto \sigma(u)x = u(x) of E×L(E)E \times \mathscr{L}(E) into EE is continuous and bilinear and hence analytic; its restriction to E×GE \times G is therefore analytic. Thus the group SS, with the product manifold structure of EE and GG, is a Lie group. We transport this structure to AA by means of Φ\Phi. Then AA becomes a Lie group, the internal semi-direct product of GG by TT as a Lie group.

Proposition 8

Let* GG and HH be Lie groups, p:G→Hp: G \to H and s:H→Gs: H \to G Lie group morphisms such that p∘s=idHp \circ s = \mathrm{id}_H and N=Ker pN = \mathrm{Ker}\ p. Then NN is a Lie subgroup of GG, s is an isomorphism of HH onto a Lie subgroup of GG and the Lie group GG is the internal semi-direct product of s(H)s(H) by NN.

Te(p)∘Te(s)=idTe(H)T_e(p) \circ T_e(s) = \mathrm{id}_{T_{e(H)}} and hence pp (resp. ss) is a submersion (resp. an immersion). By Differentiable and Analytic Manifolds, R, 5.10.5, N is a Lie subgroup of G. On the other hand, ss is a homeomorphism of H onto s(H)s(H) and hence ss is an isomorphism of H onto a Lie subgroup of G (Differentiable and Analytic Manifolds, R, 5.8.3). Finally, for all g∈Gg \in G, g=(s∘p)(g)⋅ng = (s \circ p)(g) \cdot n for some n∈Nn \in N; as s∘ps \circ p is analytic, the Lie group G is the semi-direct product of s(H)s(H) by N.

5. Quotient of a manifold by a Lie group

Let G be a Lie group, X a manifold of class CrC^r and (g,x)↦gx(g, x) \mapsto gx a law of left operation (Algebra, Chapter I, § 5, no. 1) of class CrC^r of G on X. For all g∈Gg \in G, let τ(g)\tau(g) denote the automorphism x↦gxx \mapsto gx of X defined by g. For all x∈Xx \in X, let ρ(x)\rho(x) denote the orbital mapping g↦gxg \mapsto gx of G into X defined by x. Then

ρ(x)=ρ(gx)∘δ(g),ρ(x)=τ(g)∘ρ(x)∘γ(g−1) \rho(x) = \rho(gx) \circ \delta(g), \qquad \rho(x) = \tau(g) \circ \rho(x) \circ \gamma(g^{-1})

for all g∈Gg \in G and x∈Xx \in X. Hence

Tg(ρ(x))=Te(ρ(gx))∘Tg(δ(g)) T_g(\rho(x)) = T_e(\rho(gx)) \circ T_g(\delta(g))
Tg(ρ(x))=Tx(τ(g))∘Te(ρ(x))∘Tg(γ(g−1)) T_g(\rho(x)) = T_x(\tau(g)) \circ T_e(\rho(x)) \circ T_g(\gamma(g^{-1}))

Proposition 9

Let x∈Xx \in X and g0∈Gg_0 \in G.

(i) If ρ(x)\rho(x) is an immersion (resp. a submersion, a subimmersion) at g0g_0, then, for all g∈Gg \in G, ρ(gx)\rho(gx) is an immersion (resp. a submersion, a subimmersion).

(ii) If ρ(x)\rho(x) is of rank k at g0g_0, then for all g∈Gg \in G, ρ(gx)\rho(gx) is of constant rank equal to k.

This follows immediately from formulae (4) and (5) since Tg(δ(g)),Tx(τ(g))T_g(\delta(g)), T_x(\tau(g)) and Tg(γ(g−1))T_g(\gamma(g^{-1})) are isomorphisms.

Corollary

Let x∈Xx \in X. If K is of characteristic 0 and X is finite-dimensional, ρ(x)\rho(x) is a subimmersion. If further ρ(x)\rho(x) is injective, ρ(x)\rho(x) is an immersion.

This follows from Proposition 9 and Differentiable and Analytic Manifolds, R, 5.10.6.

Observe that, if η\eta denotes the mapping (g,x)↦gx(g, x) \mapsto gx of G×XG \times X into X, then, for g∈G,x∈X,u∈Tg(G),v∈Tx(X)g \in G, x \in X, u \in T_g(G), v \in T_x(X),

T(g,x)(η)(u,v)=T(g,x)(η)(u,0)+T(g,x)(η)(0,x) T_{(g, x)}(\eta)(u, v) = T_{(g, x)}(\eta)(u, 0) + T_{(g, x)}(\eta)(0, x)

that is

T(g,x)(η)(u,v)=Tg(ρ(x))u+Tx(τ(g))v. T_{(g, x)}(\eta)(u, v) = T_g(\rho(x))u + T_x(\tau(g))v.

Proposition 10

Let G be a Lie group and X a manifold of class CrC^r with a law of left operation of class CrC^r of G on X. Suppose that:

(a) the group G operates properly and freely on X;

(b) for all x∈Xx \in X, ρ(x)\rho(x) is an immersion (which is a consequence of (a) if K is of characteristic 0 and X is finite-dimensional).

Then the equivalence relation defined by G on X is regular (Differentiable and Analytic Manifolds, R, 5.9.5). There exists on the quotient set X/G one and only one manifold structure such that the canonical mapping π:X→X/G\pi : X \to X/G is a submersion. The underlying topology of this manifold structure is the quotient topology of that on X; it is Hausdorff. Finally, (X,G,X/G,π)(X, G, X/G, \pi) is a principal left fibre bundle.\footnote{† The principal fibre bundles defined in Differentiable and Analytic Manifolds, R, 6.2.1 are principal right fibre bundles. The definition of principal left fibre bundles can be deduced from this in an obvious way.}

Let θ\theta be the mapping (g,x)↦(x,gx)(g, x) \mapsto (x, gx) of G×XG \times X into X×XX \times X. This mapping is of class CτC^\tau. We show that it is an immersion. For u∈Tg(G)u \in T_g(G) and v∈Tx(X)v \in T_x(X), by (6),

T(g,x)(θ)(u,v)=(v,Tg(ρ(x))u+Tx(τ(g))v). T_{(g, x)}(\theta)(u, v) = (v, T_g(\rho(x))u + T_x(\tau(g))v).

But Tg(ρ(x))T_g(\rho(x)) is injective by hypothesis (b) and hence T(g,x)(θ)T_{(g, x)}(\theta) is injective. Its image is the topological direct sum of the subspace Hg,xH_{g, x} consisting of the vectors (v,Tx(τ(g))v)(v, T_x(\tau(g))v) for v∈Tx(X)v \in T_x(X) and the subspace

Ig,x={0}×Tg(ρ(x))(Tg(G)). I_{g, x} = \{0\} \times T_g(\rho(x))(T_g(G)).

By hypothesis (b), Tg(ρ(x))(Tg(G))T_g(\rho(x))(T_g(G)) admits a topological supplement Jg,xJ_{g, x} in Tgx(X)T_{gx}(X). Hence the image of T(g,x)(θ)T_{(g, x)}(\theta) admits the topological supplement {0}×Jg,x\{0\} \times J_{g, x}. Hence we have proved that θ\theta is an immersion of G×XG \times X in X×XX \times X.

As G operates freely on X, θ\theta is injective. Let C be the graph of the equivalence relation R defined by G on X. As G operates properly, θ\theta is a homeomorphism of G×XG \times X onto C (General Topology, Chapter I, § 10, Proposition 2). By Differentiable and Analytic Manifolds, R, 5.8.3, it is a submanifold of X×XX \times X and θ\theta is an isomorphism of the manifold G×XG \times X onto the manifold C. The tangent space T(x,gx)(C)T_{(x, gx)}(C) is identified with

T(g,x)(θ)(T(g,x)(G×X))=Hg,x⊕Ig,x⊂T(x,gx)(X×X). T_{(g, x)}(\theta)(T_{(g, x)}(G \times X)) = H_{g, x} \oplus I_{g, x} \subset T_{(x, gx)}(X \times X).

Let pr1_1 and pr2_2 be the canonical projections of X×XX \times X onto the two factors. It is immediate that T(x,gx)(pr1)T_{(x, gx)}(\mathrm{pr}_1) maps Hg,xH_{g, x} onto Tx(X)T_x(X) and that the kernel of T(x,gx)(pr1)∣T(x,gx)(C)T_{(x, gx)}(\mathrm{pr}_1) | T_{(x, gx)}(C) is Ig,xI_{g, x}. Thus pr1∣C\mathrm{pr}_1 | C is a submersion of C onto X. By Differentiable and Analytic Manifolds, R, 5.9.5, R is regular. By definition, there therefore exists on the quotient set X/G one and only one manifold structure such that π\pi is a submersion. The underlying topology of X/G is the quotient topology of that of X (Differentiable and Analytic Manifolds, R, 5.9.4). This topology is Hausdorff (General Topology, Chapter III, § 4, Proposition 3).

For all b∈X/Gb \in X/G, there exist an open neighbourhood W of b and a morphism σ:W→X\sigma : W \to X such that π∘σ=idW\pi \circ \sigma = \mathrm{id}_W (Differentiable and Analytic Manifolds, R, 5.9.1). Let ϕ\phi be the bijection (g,w)↦gσ(w)(g, w) \mapsto g\sigma(w) of G×WG \times W onto π−1(W)\pi^{-1}(W). It is of class CτC^\tau. Then π(gσ(w))=w\pi(g\sigma(w)) = w and

θ−1(σ(w),gσ(w))=(g,σ(w)) \theta^{-1}(\sigma(w), g\sigma(w)) = (g, \sigma(w))

and hence the inverse bijection of ϕ\phi is of class CrC^r. Clearly ϕ(gg′,w)=gϕ(g′,w)\phi(gg', w) = g\phi(g', w) for w∈W, g∈G, g′∈Gw \in W,\ g \in G,\ g' \in G. Hence (X,G,X/G,π)(X, G, X/G, \pi) is a principal left fibre bundle.

Remark

With the above hypotheses, further let HH be a manifold of class CrC^r and (x,h)↦m(x,h)(x, h) \mapsto m(x, h) a mapping of class CrC^r of X×HX \times H into XX such that m(gx,h)=gm(x,h)m(gx, h) = gm(x, h) for x∈X, g∈G, h∈Hx \in X,\ g \in G,\ h \in H. Let nn be the mapping of (X/G)×H(X/G) \times H into X/GX/G derived from mm by taking quotients. We show that nn is of class CrC^r. Consider the diagram

X×H→mXπ×1↓↓π(X/G)×H→nX/G \begin{array}{ccc} X \times H & \xrightarrow{m} & X \\ \pi \times 1 \downarrow & & \downarrow \pi \\ (X/G) \times H & \xrightarrow{n} & X/G \end{array}

It is commutative, π∘m\pi \circ m is of class CrC^r and π×1\pi \times 1 is a surjective submersion; it then suffices to apply Differentiable and Analytic Manifolds, R, 5.9.5.

Let GG be a Lie group, XX a manifold of class CrC^r and (g,x)↦xg(g, x) \mapsto xg a law of right operation of class CrC^r of GG on XX. Let τ(g)x=ρ(x)g=xg\tau(g)x = \rho(x)g = xg for g∈G, x∈Xg \in G,\ x \in X. Then this time

(3′)(3')
ρ(x)=ρ(xg)∘γ(g−1),ρ(x)=τ(g)∘ρ(x)∘δ(g) \rho(x) = \rho(xg) \circ \gamma(g^{-1}), \qquad \rho(x) = \tau(g) \circ \rho(x) \circ \delta(g)
and hence
(4′)(4')
Tg(ρ(x))=Te(ρ(xg))∘Tg(γ(g−1)) T_g(\rho(x)) = T_e(\rho(xg)) \circ T_g(\gamma(g^{-1}))
(5′)(5')
Tg(ρ(x))=Tx(τ(g))∘Te(ρ(x))∘Tg(δ(g)). T_g(\rho(x)) = T_x(\tau(g)) \circ T_e(\rho(x)) \circ T_g(\delta(g)).

On the other hand, if η\eta denotes the mapping (g,x)↦xg(g, x) \mapsto xg of G×XG \times X into XX, formula (6) remains valid. Proposition 9, its Corollary and Proposition 10 remain equally true (with "principal left fibre bundle" replaced by "principal right fibre bundle" in the last).

6. Homogeneous spaces and quotient groups

Proposition 11

Let XX be a Lie group and GG a Lie subgroup of XX.

(i) There exists on the homogeneous set X/GX/G one and only one analytic manifold structure such that the canonical projection π\pi of XX onto X/GX/G is a submersion. The law of operation of XX on X/GX/G is analytic. For all x∈Xx \in X, the kernel of Tx(π)T_x(\pi) is obtained from Te(G)T_e(G) by Te(γ(x))T_e(\gamma(x)).

(ii) If GG is normal in XX, X/GX/G is a Lie group with its group structure and the manifold structure defined in (i). The mapping π\pi is a Lie group morphism.

By General Topology, Chapter III, § 4, no. 1, Example 1, GG operates properly and freely on XX by right translation. Hence the first assertion of (i) follows from

Proposition 10 of no. 5. The second follows from the Remark of no. 5. Since π\pi is a submersion, the kernel of Tx(π)T_x(\pi) is the tangent space at xx to

π−1(π(x))=xG=γ(x)(G) \pi^{-1}(\pi(x)) = xG = \gamma(x)(G)
and hence is obtained from Te(G)T_e(G) by Te(γ(x))T_e(\gamma(x)).

Suppose that GG is normal. Let mm be the mapping (x,y)↦xy−1(x, y) \mapsto xy^{-1} of (X/G)×(X/G)(X/G) \times (X/G) into X/GX/G. Then (m∘(π×π))(x,y)=π(xy−1)(m \circ (\pi \times \pi))(x, y) = \pi(xy^{-1}) for all x,yx, y in XX. Hence m∘(π×π)m \circ (\pi \times \pi) is analytic. As π×π\pi \times \pi is a surjective submersion, mm is analytic (Differentiable and Analytic Manifolds, R. 5.9.5), whence (ii).

The homogeneous set X/GX/G with the manifold structure defined in (i) is called the quotient (left) Lie homogeneous space of XX by GG. The (right) Lie homogeneous space G∖XG \setminus X is defined analogously. When GG is normal, the Lie group X/GX/G defined in (ii) is called the quotient Lie group of XX by GG.

Proposition 12

Let XX be a Lie group and YY a non-empty analytic manifold with a law of analytic left operation of XX on YY. For all y∈Yy \in Y, let φ(y)\varphi(y) be the orbital mapping by yy and XyX_y the stabilizer of yy in XX. The following conditions are equivalent: (i) there exists y∈Yy \in Y such that φ(y)\varphi(y) is a surjective submersion; (i') for all y∈Yy \in Y, φ(y)\varphi(y) is a surjective submersion; (ii) there exists y∈Yy \in Y such that XyX_y is a Lie subgroup of XX and the canonical mapping of X/XyX/X_y into YY is a manifold isomorphism; (ii') for all y∈Yy \in Y, XyX_y is a Lie subgroup of XX and the canonical mapping of X/XyX/X_y into YY is a manifold isomorphism; (iii) the mapping (x,y)↦(y,xy)(x, y) \mapsto (y, xy) of X×YX \times Y into Y×YY \times Y is a surjective submersion.

As the canonical mapping of XX onto X/XyX/X_y is a submersion, the equivalences (i) ⇔\Leftrightarrow (ii), (i') ⇔\Leftrightarrow (ii') are immediate. (i) ⇔\Leftrightarrow (i') by Proposition 9 of no. 5. The equivalence (i') ⇔\Leftrightarrow (iii) follows from formula (7) of no. 5.

Under the conditions of Proposition 12, YY is called a (left) Lie homogeneous space of XX. Right Lie homogeneous spaces of XX are defined analogously.

Example

Let GG be a Lie group. We make G×GG \times G operate on GG on the left by (g1,g2)x=g1xg2−1(g_1, g_2)x = g_1xg_2^{-1}. Let ρ\rho be the orbital mapping of ee. Then the restrictions of T(e,e)(ρ)T_{(e, e)}(\rho) to T(e,e)(G×{e})=Te(G)×{0}T_{(e, e)}(G \times \{e\}) = T_e(G) \times \{0\} and to

T(e,e)({e}×G)={0}×Te(G) T_{(e, e)}(\{e\} \times G) = \{0\} \times T_e(G)
are isomorphisms of these spaces onto Te(G)T_e(G). Hence T(e,e)(ρ)T_{(e, e)}(\rho) is surjective and Ker⁡T(e,e)(ρ)\operatorname{Ker} T_{(e, e)}(\rho) admits for example the topological supplement Te(G)×{0}T_e(G) \times \{0\} in T(e,e)(G×G)T_{(e, e)}(G \times G). Thus ρ\rho is a submersion at (e,e)(e, e). Hence GG is a left Lie homogeneous space of G×GG \times G.

Proposition 13

Let GG be a Lie group, HH a normal Lie subgroup of GG, XX a manifold of class CrC^r and (g,x)↦gx(g, x) \mapsto gx a law of left operation of class CrC^r of GG on XX. Suppose that conditions (a) and (b) of Proposition 10 hold.

(i) The law of left operation (h,x)↦hx(h, x) \mapsto hx of HH on XX satisfies conditions (a) and (b) of Proposition 10 (so that we can consider the quotient manifolds X/GX/G and X/HX/H).

(ii) The law of left operation of GG on XX defines on taking quotients a law of left operation of class CrC^r of G/HG/H on X/HX/H; this law satisfies conditions (a) and (b) of Proposition 10 (so that we can consider the quotient manifold (X/H)/(G/H)(X/H)/(G/H)).

(iii) The canonical mapping of XX onto X/HX/H defines on taking quotients a bijection of X/GX/G onto (X/H)/(G/H)(X/H)/(G/H). This bijection is an isomorphism of manifolds of class CrC^r.

Clearly HH operates freely on XX; it operates properly by General Topology, Chapter III, § 4, no. 1, Example 1. The orbital mappings of HH on XX are immersions since the canonical injection of HH into GG is an immersion. This proves (i).

The law of left operation of GG on XX obviously defines, on taking quotients, a law of left operation of G/HG/H on X/HX/H. This law is of class CrC^r by Differentiable and Analytic Manifolds, R, 5.9.6. Let g∈Gg \in G and x∈Xx \in X be such that (Hg)(Hx)=Hx(Hg)(Hx) = Hx; then H(gx)=HxH(gx) = Hx and hence gx∈Hxgx \in Hx and g∈Hg \in H; this proves that G/HG/H operates freely on X/HX/H. The mapping θ:(g,x)↦(x,gx)\theta : (g, x) \mapsto (x, gx) of G×XG \times X into X×XX \times X is closed; on the other hand, θ(Hg×Hx)=Hx×H(gx)\theta(Hg \times Hx) = Hx \times H(gx); it follows immediately that the mapping

(Hg,Hx)↦(Hx,H(gx))(Hg, Hx) \mapsto (Hx, H(gx))

of (G/H)×(X/H)(G/H) \times (X/H) into (X/H)×(X/H)(X/H) \times (X/H) is closed; as moreover G/HG/H operates freely on X/HX/H, Theorem 1 (c) of General Topology, Chapter I, § 10, no. 2 proves that G/HG/H operates properly on X/HX/H.

Let π\pi be the canonical mapping of XX onto X/HX/H, σ\sigma the canonical mapping of GG onto G/HG/H, xx an element of XX and y=π(x)y = \pi(x).

G→ρ(x)X↓σ↓πG/H→ρ(y)X/H \begin{array}{ccc} G & \xrightarrow{\rho(x)} & X \\ \downarrow \sigma & & \downarrow \pi \\ G/H & \xrightarrow{\rho(y)} & X/H \end{array}

Then π∘ρ(x)=ρ(y)∘σ\pi \circ \rho(x) = \rho(y) \circ \sigma and hence

Tx(π)∘Te(ρ(x))=Te(ρ(y))∘Te(σ). T_x(\pi) \circ T_e(\rho(x)) = T_e(\rho(y)) \circ T_e(\sigma).

Let u∈Te(G/H)u \in T_e(G/H) be such that Te(ρ(y))u=0T_e(\rho(y))u = 0. There exists v∈Te(G)v \in T_e(G) such that u=Te(σ)vu = T_e(\sigma)v. Then Tx(π)(Te(ρ(x))v)=0T_x(\pi)(T_e(\rho(x))v) = 0, hence Te(ρ(x))vT_e(\rho(x))v is tangent to HxHx (Differentiable and Analytic Manifolds, R, 5.10.5) and therefore is of the form Te(ρ(x)∣H)v′T_e(\rho(x) | H)v' for some v′∈Te(H)v' \in T_e(H). As Te(ρ(x))T_e(\rho(x)) is injective, it follows that v=v′v = v', whence v∈Te(H)v \in T_e(H) and therefore u=0u = 0. Thus Te(ρ(y))T_e(\rho(y)) is injective. The image of Te(ρ(y))T_e(\rho(y)) is equal to that of Tx(π)∘Te(ρ(x))T_x(\pi) \circ T_e(\rho(x)); now the image of Te(ρ(x))T_e(\rho(x)) admits a topological supplement in Tx(X)T_x(X) and contains the kernel of Tx(π)T_x(\pi). It is therefore seen that ρ(y)\rho(y) is an immersion, which completes the proof of (ii).

Assertion (iii) follows from the above and Differentiable and Analytic Manifolds, R, 5.9.7.

Corollary

Let G be a Lie group and H and L normal Lie subgroups of G with L ⊂ H. Then H/L is a normal Lie subgroup of G/L and the canonical bijection of G/H onto (G/L)/(H/L) is a Lie group isomorphism.

7. Orbits

Proposition 14

Let G be a Lie group, X an analytic manifold and (g, x) ↦ gx a law of analytic left operation of G on X. Let x ∈ X. Suppose that the corresponding orbital mapping ρ(x) is a subimmersion (which is always the case if K is of characteristic 0 and X is finite-dimensional (Corollary to Proposition 9)). Let G_x be the stabilizer of x in G. (i) G_x is a Lie subgroup and T_e(G_x) = Ker T_e(ρ(x)). (ii) The canonical mapping i_x of the homogeneous space G/G_x into X is an immersion with image Gx. (iii) If further the orbit Gx is locally closed and the topology on G admits a countable base, then Gx is a submanifold of X, i_x is an isomorphism of the manifold G/G_x onto the manifold Gx and T_x(Gx) = Im T_e(ρ(x)).

The inverse image of x under ρ(x) is G_x. As ρ(x) is a subimmersion, G_x is a submanifold and, for all g ∈ G, the tangent space J to gG_x = ρ_{(x)}^{-1}(gx) at g is Ker T_g(ρ(x)) (Differentiable and Analytic Manifolds, R, 5.10.5), whence (i). Let π: G → G/G_x be the canonical mapping. Then i_x ∘ π = ρ(x). As G/G_x is a quotient manifold of G, this equality proves that i_x is analytic. Further, the kernels of T_g(ρ(x)) and T_g(π) are both equal to J. Hence T_{π(g)}(i_x) is injective. The image of T_{π(g)}(i_x) is equal to the image of T_g(ρ(x)) and hence admits a topological supplement. This proves (ii).

Suppose that Gx is locally closed. Every point of Gx then has a neighbourhood in Gx which is homeomorphic to a closed subspace of a complete metric space and hence is a Baire space. Hence Gx is a Baire space (General Topology, Chapter IX, § 5, Proposition 4). If G has a countable base, i_x is therefore a homeomorphism of G/G_x onto Gx (General Topology, Chapter IX, § 5). Then by (ii) and Differentiable and Analytic Manifolds, R, 5.8.3, i_x is an isomorphism of the manifolds G/G_x onto the manifold Gx and

Tx(Gx)=Im⁡Tπ(e)(ix)=Im⁡Te(ρ(x)). T_x(Gx) = \operatorname{Im} T_{\pi(e)}(i_x) = \operatorname{Im} T_e(\rho(x)).

Remark

Let G be a finite-dimensional Lie group, X a manifold of class C^r and (g, x) ↦ gx a law of left operation of class C^r of G on X. Then Proposition 14 remains valid. The only point which needs a different proof is the fact that G_x is a Lie subgroup. But if r ≠ ω, K = \mathbf{R}; as clearly G_x is closed, G_x is a Lie subgroup by § 8, Theorem 2.

Corollary

Let G be a Lie group whose topology admits a countable base and X a non-empty finite-dimensional analytic manifold with a law of analytic left operation of G on X. Suppose that G operates transitivity on X and that K is of characteristic 0. Then X is a Lie homogeneous space for G.

Let x ∈ X. The orbit of x, equal to X, is closed and we can therefore apply Proposition 14 (iii).

8. Vector bundles with operators

Let G be a Lie group, X a manifold of class C^r and (g, x) ↦ gx a law of left operation of class C^r of G on X. Let E be a vector bundle of class C^r, with base space X and π : E → X the projection of E onto X. For all x ∈ X, let E_x be the fibre of E at x. Let (g, u) ↦ gu be a law of left operation of G on E such that π is compatible with the operations of G on X and on E. For all g ∈ G and all x ∈ X, the restriction to E_x of the mapping u ↦ gu is a bijection ψ_{g, x} of E_x onto E_{gx}. We shall assume that, for all g ∈ G and all x ∈ X, ψ_{g, x} is continuous and linear and hence is an isomorphism of the normable space E_x onto the normable space E_{gx}.

Let φ be the automorphism (g, x) ↦ (g, gx) of the manifold G × X. Let p be the canonical projection of G × X onto X and E' the inverse image of E under p. Let ψ : E' → E' be the mapping the sum of the ψ_{g, x} : E'_{(g, x)} → E'_{(g, gx)}.

Definition 4

If ψ is a φ-morphism of vector bundles of class C^r, E is called a vector G-bundle of class C^r.

In other words, E is a vector G-bundle of class C^r if for all (g_0, x_0) ∈ G × X the following condition holds: there exists an open neighbourhood U of (g_0, x_0) in G × X such that, if E' | U (resp. E' | φ(U)) is identified with a trivial vector bundle of fibre M (resp. N) by means of a vector chart, the mapping (g, x) ↦ ψ_{g, x} of U into L(M, N) is of class C^r.

The mapping ψ is obviously bijective and it follows from the above local criterion that ψ^{-1} is a φ^{-1}-morphism of vector bundles so that ψ is a φ-isomorphism of vector bundles.

A trivial vector G-bundle of base X is a vector bundle X × F (where F is a complete normable space) with the law of operation (g, (x, f)) ↦ (gx, f) of G on X × F.

We again assume the hypotheses and notation preceding Definition 4 and further take τ to be a vector functor of class C^r for isomorphisms (Differentiable and Analytic Manifolds, R, 7.6.6). Then τE is a vector bundle with base space X. For all x ∈ X, its fibre (τE)_x is equal to τ(E_x). For all normable spaces N_1, N_2, let Isom(N_1, N_2) denote the set of isomorphisms of N_1 onto N_2. If g ∈ G, then

τ(ψg,x)∈Isom((τE)x,(τE)gx). \tau(\psi_{g, x}) \in \mathrm{Isom}((\tau E)_x, (\tau E)_{gx}).

The τ(ψ_{g, x}) define a law of left operation (g, u) ↦ gu of G on τE and the canonical projection of τE onto X is compatible with the operations of G on X and τE.

Proposition 15

If E is a vector G-bundle of class C^r, τE is a vector G-bundle of class C^r*.

Let g_0, x_0, U, M, N be as in the paragraph following Definition 4. Then the mapping (g, x) ↦ τ(ψ_{g,x}) of U into L(τM,τN)\mathscr{L}(\tau M, \tau N) is the composition of the mapping (g, x) ↦ ψ_{g,x} of U into L(M,N)\mathscr{L}(M, N) and the mapping f ↦ τ(f) of Isom(M, N) into Isom(τM, τN); these two mappings are of class C^r and hence so is their composition, whence the proposition.

Proposition 16

Let G be a Lie group, X a manifold of class C^r (r ≥ 2) and (g, x) ↦ gx a law of left operation of class C^r of G on X, whence, by transporting the structure, there is a law of left operation of G on TX. Under this law, TX is a vector G-bundle of class C^{r-1}.

Let pr_1 (resp. pr_2) be the canonical projection of G × X onto G (resp. X) and let E_1 (resp. E_2) be the inverse image of TG (resp. TX) relative to pr_1 (resp. pr_2). Then the vector bundle T(G × X) is the direct sum of E_1 and E_2. Let i: E_2 → T(G × X) and q: T(G × X) → E_2 be the canonical vector bundle morphisms defined by this decomposition into a direct sum. Let φ be the mapping (g, x) ↦ (g, gx) of G × X into G × X. Then the mapping denoted by ψ in Definition 4 (where we put E = TX) is just q ∘ T(φ) ∘ i. But T(φ) is a φ-morphism of vector bundles of class C^{r-1} (Differentiable and Analytic Manifolds, R, 8.1.2).

Corollary

If τ is a vector functor of class C^r for isomorphisms, τ(TX) is a vector G-bundle of class C^{r-1}.

This follows from Propositions 15 and 16.

Remark 1

If τ is a vector functor of class C^r for isomorphisms in finite dimension and E is of finite rank, τE is defined similarly and Proposition 15 remains valid; the Corollary to Proposition 16 remains valid provided X is finite-dimensional.

Example

With the hypotheses and notation of Proposition 16, let F be a complete normable space. Then L((TX)p;F)\mathscr{L}((TX)^p; F) is a vector G-bundle of class C^{r-1}; so is Alt^p(TX; F) if K is of characteristic zero or X is finite-dimensional (Differentiable and Analytic Manifolds, R, 7.7, 7.8). If X is finite-dimensional, ⨂p(TX)⊗⨂q(TX)∗\bigotimes^p (TX) \otimes \bigotimes^q (TX)^* is a vector G-bundle of class C^{r-1}.

Proposition 17

Let G be a Lie group, X a left Lie homogeneous space of G, x_0 a point of X, G_0 the stabilizer of x_0 in G, E and E' left vector G-bundles of class C^r and base space X, E_0 (resp. E'_0) the fibre of E (resp. E') at x_0 and f an element of L(E0,E0′)\mathscr{L}(E_0, E'_0) such that f(gu) = gf(u) for all u ∈ E_0 and g ∈ G_0. Then there exists one and only one morphism of E into E' compatible with the operations of G and extending f.

The uniqueness of this morphism is obvious. We prove its existence. Let g, g' elements of G and u ∈ E_0 be such that gu = g'u. Then g'^{-1}g ∈ G_0 and g'^{-1}gu = u and hence g'^{-1}gf(u) = f(u), that is gf(u) = g'f(u). Hence a mapping φ is defined of E into E' by writing φ(gu) = gf(u). Clearly this mapping extends f and it is compatible with the operations of G. We show that φ is a vector bundle morphism of class C^r. Let x_1 ∈ X. There exists an open neighbourhood V of x_1 in X and a submanifold W of G such that the mapping g ↦ gx_0 is an isomorphism θ of class C^r of W onto V. By shrinking V and W it can be assumed that: (1) E | V (resp. E' | V) is identified with a trivial vector bundle of fibre M (resp. M'); (2) if ψ_g (resp. ψ'_g) denotes the mapping u ↦ gu of E_0 (resp. E'_0) into E_{gx_0} (resp. E'_{gx_0}), then the mappings g ↦ ψ_g and g ↦ ψ_g^{-1} (resp. g ↦ ψ'_g and g ↦ ψ'_g^{-1}) of W into L(E0,M)\mathscr{L}(E_0, M) and L(M,E0)\mathscr{L}(M, E_0) (resp. L(E0′,M′)\mathscr{L}(E'_0, M') and L(M′,E0′)\mathscr{L}(M', E'_0)) are of class C^r.

For x ∈ V, let φ_x : M → N be the restriction of φ to E_x = M. Then φ_x is obtained by composing the following mappings: (1) the mapping (ψ_θ^{-1,x})^{-1} of M into E_0; (2) the mapping f of E_0 into E_0; (3) the mapping ψ'_θ^{-1,x} of E_0 into M'. Hence we see that the mapping x ↦ φ_x of V into L(M,M′)\mathscr{L}(M, M') is of class C^r.

Corollary 1

Let E_0^{G_0} be the set of elements of E_0 which are invariant under G_0. For all u ∈ E_0^{G_0}, let σ_u be the mapping of X into E defined by σ_u(gx_0) = gu for all g ∈ G. (i) The G-invariant sections† of E are of class C^r. (ii) u ↦ σ_u is a bijection of E_0^{G_0} onto the set of G-invariant sections of E.

Assertion (ii) is obvious. To prove (i) it is sufficient to prove that each section σ_u is of class C^r. Let E' be the trivial G-bundle of base X and fibre E_0^{G_0}. Let f be the canonical injection of E_0^{G_0} into E_0. By Proposition 17 there exists a morphism φ of E' into E compatible with the operations of G and extending f. If u ∈ E_0^{G_0} and g ∈ G, then

σu(gx0)=gu=gf(u)=φ(gu)=φ((u,gx0)) \sigma_u(gx_0) = gu = gf(u) = \varphi(gu) = \varphi((u, gx_0))

and hence σ_u(x) = φ((u, x)) for all x ∈ X, which proves our assertion.

For example, let G be a finite-dimensional real Lie group, G0 a compact Lie subgroup of G and X the homogeneous space G/G_0. Let x_0 denote the canonical image of e in X. There exists a positive definite symmetric bilinear form on T_{x_0}(X) invariant under G_0 (Integration, Chapter VII, § 3, Proposition 1). Applying the above to (TX)* ⊗ (TX)*, we see that there exists on X an analytic Riemannian metric invariant under G_*.

† By a section of E we here mean a mapping σ (not necessarily of class C^r) of X into E such that p ∘ σ = Id_X, where p denotes the projection of E onto X.

Corollary 2

Suppose that G0G_0 operates trivially on E0E_0. Let E′E' be the trivial GG-bundle with base space XX and fibre E0E_0. There exists one and only one isomorphism of EE onto E′E' compatible with the operations of GG and extending IdE0\mathrm{Id}_{E_0}.

This follows immediately from Proposition 17.

Remark 2

In this no., the laws of left operation can be replaced throughout by laws of right operation.

9. Local definition of a Lie group

Proposition 18

Let GG be a group and UU and VV two subsets of GG containing ee. Suppose that UU has an analytic manifold structure satisfying the following conditions: (i) V=V−1,V2⊂U,VV = V^{-1}, V^2 \subset U, V is open in UU; (ii) the mapping (x,y)↦xy−1(x, y) \mapsto xy^{-1} of V×VV \times V into UU is analytic; (iii) for all g∈Gg \in G, there exists an open neighbourhood V′V' of ee in VV such that gV′g−1⊂UgV'g^{-1} \subset U and such that the mapping x↦gxg−1x \mapsto gxg^{-1} of V′V' into UU is analytic.

Then there exists one and only one analytic manifold structure on GG with the following properties: (α) GG with this structure is a Lie group; (β) VV is open in GG; (γ) the manifold structures on GG and UU induce the same structure on VV.

(a) Let AA be an open subset of VV and v0v_0 an element of VV such that v0A⊂Vv_0A \subset V. Then v0Av_0A is the set of v∈Vv \in V such that v0−1v∈Av_0^{-1}v \in A and hence is an open subset of VV (taking account of (ii)). Moreover, (ii) implies that the mappings v↦v0vv \mapsto v_0v of AA onto v0Av_0A and v↦v0−1vv \mapsto v_0^{-1}v of v0Av_0A onto AA are inverse analytic bijections and hence analytic isomorphisms.

(b) We choose an open neighbourhood WW of ee in VV such that W=W−1W = W^{-1}, W3⊂VW^3 \subset V and there exists a chart (W,ϕ,E)(W, \phi, E) of the manifold UU with domain WW. For all g∈Gg \in G, let ϕg\phi_g be the mapping h↦ϕ(g−1h)h \mapsto \phi(g^{-1}h) of gWgW into EE. We show that the charts ϕg\phi_g of GG are analytically compatible. Let g1,g2g_1, g_2 be elements of GG such that g1W∩g2W≠∅g_1W \cap g_2W \neq \emptyset, so that g2−1g1g_2^{-1}g_1 and g1−1g2g_1^{-1}g_2 belong to W2W^2. By (a), W∩g1−1g2WW \cap g_1^{-1}g_2W is an open subset of WW and hence

ϕg1(g1W∩g2W)=ϕ(W∩g1−1g2W) \phi_{g_1}(g_1W \cap g_2W) = \phi(W \cap g_1^{-1}g_2W)
is an open subset DD of EE. For d∈Dd \in D,
(ϕg2∘ϕg1−1)(d)=ϕ(g2−1g1ϕ−1(d)); (\phi_{g_2} \circ \phi_{g_1}^{-1})(d) = \phi(g_2^{-1}g_1\phi^{-1}(d));
by (a) we see that ϕg2∘ϕg1−1\phi_{g_2} \circ \phi_{g_1}^{-1} is analytic.

(c) By (b) there exists on GG an analytic manifold structure such that (ϕg)g∈G(\phi_g)_{g \in G} is an atlas on GG. For all g0∈Gg_0 \in G, the mapping g↦g0gg \mapsto g_0g (g∈Gg \in G) leaves this atlas invariant and hence is an automorphism of GG with this manifold structure. In particular condition (GL₁) is satisfied.

(d) Let v0∈Vv_0 \in V. By (ii) there exists an open neighbourhood AA of ee in WW such that v0A⊂Vv_0A \subset V. This proves first that VV is open in GG. By (a) the mapping v↦v0vv \mapsto v_0v of AA onto v0Av_0A is an analytic isomorphism with the structures induced by U. By (c) we see that the manifold structures on G and U induce the same structure on v_0A and hence finally on V.

(e) by (d), (ii) and (iii), we see that conditions (GL_2) and (GL_3) hold. Hence G is a Lie group.

(f) If a manifold structure on G is compatible with the group structure of G and is such that V is an open submanifold of G, then (\phi_g)_{g\in G} is an atlas on G. Hence the uniqueness assertion of the proposition.

Proposition 19

Let G be a topological group, H a Lie group and f a homomorphism of the group G into the group H. Suppose that there exist an open neighbourhood of e_G in G, a chart (V, \phi, E) of the manifold H at e_H and a closed vector subspace F of E admitting a topological supplement, such that f(U) \subset V and \phi \circ f)|U is a homeomorphism of U onto \phi(V) \cap F. Then there exists a unique manifold structure on G such that f is an immersion; this structure is the inverse image under f of the manifold structure on H. With this structure G is a Lie group.

As translations of G (resp. H) are homeomorphisms (resp. analytic isomorphisms), f satisfies condition (R) of Differentiable and Analytic Manifolds, R, 5.8.1. The first two assertions of the proposition then follow from Differentiable and Analytic Manifolds, R, loc. cit. Consider the commutative diagram

G×G→mGf×f↓↓fH×H→nH \begin{array}{ccc} G \times G & \xrightarrow{m} & G \\ f \times f \downarrow & & \downarrow f \\ H \times H & \xrightarrow{n} & H \end{array}

where m(x, y) = xy^{-1} (resp. n(x, y) = xy^{-1}) for x, y in G (resp. H). Then n \circ (f \times f) is analytic, hence f \circ m is analytic and hence m is analytic since f is an immersion. Therefore G is a Lie group.

The Lie group structure on G is called the inverse image of the Lie group structure on H under f.

Corollary

Let G be a topological group, N a discrete normal subgroup of G and \pi the canonical mapping of G onto G/N. Suppose that an analytic manifold structure is given on G/N, compatible with the topological group structure on G/N. Then there exists a unique manifold structure on G such that \pi is an immersion; this structure is the inverse image under \pi of the manifold structure on G/N. With this structure, \pi is étale, G is a Lie group and G/N is the quotient Lie group of G by N.

Remark

Let H be a connected real or complex Lie group, \tilde{H} its universal covering† and \pi the canonical mapping of \tilde{H} onto H. When we speak of \tilde{H}

† Cf. General Topology, Chapter XI; whilst awaiting the publication of this chapter, see for example L. S. Pontrjagin, Topological groups, 2nd edition translated from Russian, Gordon and Breach, 1966; or G. Hochschild, The structure of Lie groups, Holden-Day, 1965.

LIE GROUPS

as a Lie group, we shall always mean with the inverse image structure of that on H under π\pi.

10. Group germs

Definition 5

A Lie group germ over K* is a system (G,e,θ,m)(G, e, \theta, m) satisfying the following conditions: (i) GG is an analytic manifold over KK; (ii) e∈Ge \in G; (iii) θ\theta is an analytic mapping of GG into GG; (iv) mm is an analytic mapping of an open subset Ω\Omega of G×GG \times G into GG; (v) for all g∈Gg \in G, (e,g)∈Ω(e, g) \in \Omega, (g,e)∈Ω(g, e) \in \Omega, m(e,g)=m(g,e)=gm(e, g) = m(g, e) = g; (vi) for all g∈Gg \in G, (g,θ(g))∈Ω(g, \theta(g)) \in \Omega, (θ(g),g)∈Ω(\theta(g), g) \in \Omega, m(g,θ(g))=m(θ(g),g)=em(g, \theta(g)) = m(\theta(g), g) = e; (vii) if g,h,kg, h, k are elements of GG such that (g,h)∈Ω(g, h) \in \Omega, (h,k)∈Ω(h, k) \in \Omega, (m(g,h),k)∈Ω(m(g, h), k) \in \Omega, (g,m(h,k))∈Ω(g, m(h, k)) \in \Omega, then m(m(g,h),k)=m(g,m(h,k))m(m(g, h), k) = m(g, m(h, k)).

ee is called the identity element of the group germ. We often write ghgh instead of m(g,h)m(g, h) and (by an abuse of notation) g−1g^{-1} instead of θ(g)\theta(g).

A Lie group GG is a Lie group germ with the obvious choice of e,θ,me, \theta, m.

Let GG be a Lie group germ. Then ee−1=eee^{-1} = e, that is

e−1=e. e^{-1} = e.

For all g∈Gg \in G,

g=eg=((g−1)−1g−1)g=(g−1)−1(g−1g)=(g−1)−1e, g = eg = ((g^{-1})^{-1}g^{-1})g = (g^{-1})^{-1}(g^{-1}g) = (g^{-1})^{-1}e,

that is

(g−1)−1=g. (g^{-1})^{-1} = g.

A subset of GG invariant under the mapping g↦g−1g \mapsto g^{-1} is called symmetric.

The manifold GG, with the point ee, the mapping g↦g−1g \mapsto g^{-1} and the mapping (g,h)↦hg(g, h) \mapsto hg is a Lie Group germ G∨G^\vee called the opposite of GG.

The Lie group germ GG is called commutative if, for all (g,h)∈G×G(g, h) \in G \times G such that ghgh is defined, hghg is defined and equal to ghgh.

Let GG be a Lie group germ. The set of (g,h)∈G×G(g, h) \in G \times G such that ghgh is defined is a neighbourhood of (e,e)(e, e). On the other hand, the mappings (g,h)↦gh(g, h) \mapsto gh and g↦g−1g \mapsto g^{-1} are continuous. Hence (gh)k=g(hk)(gh)k = g(hk) for g,h,kg, h, k sufficiently close to ee. Similarly, (h−1g−1)(gh)=h−1(eh)=h−1h=e(h^{-1}g^{-1})(gh) = h^{-1}(eh) = h^{-1}h = e for g,hg, h sufficiently close to ee, whence multiplying on the right by (gh)−1(gh)^{-1},

(gh)−1=h−1g−1for g,h sufficiently close to e. (gh)^{-1} = h^{-1}g^{-1} \quad \text{for } g, h \text{ sufficiently close to } e.

Proposition 20

Let GG be a Lie group germ and g∈Gg \in G. There exist an open neighbourhood UU of ee and an open neighbourhood VV of gg with the following properties:* (a) ugug is defined for all u∈Uu \in U; (b) vg−1vg^{-1} is defined for all v∈Vv \in V; (c) the mappings u↦ug, v↦vg−1u \mapsto ug,\ v \mapsto vg^{-1} are inverse analytic isomorphisms of one another of UU onto VV and of VV onto UU.

§ 1.10

As the set of definition of the product is open in G×GG \times G, there exist an open neighbourhood UU of ee and an open neighbourhood VV of gg with properties (a) and (b). Let η(u)=ug\eta(u) = ug for u∈Uu \in U, η′(v)=vg−1\eta'(v) = vg^{-1} for v∈Vv \in V. By shrinking UU and VV, it can be assumed that (ug)g−1=u(ug)g^{-1} = u and (vg−1)g=v(vg^{-1})g = v for u∈Uu \in U and v∈Vv \in V. Then η\eta and η′\eta' are injections. By shrinking UU further, it can be assumed that η(U)⊂V\eta(U) \subset V. Then η′(V)⊃U\eta'(V) \supset U and η(U)\eta(U) is the inverse image of UU under η′\eta' and hence is an open neighbourhood of gg in VV. Replacing VV by η(U)\eta(U), we finally arrive at the situation where η\eta and η′\eta' are analytic inverse bijections.

Let G1,G2G_1, G_2 be two Lie group germs with identity elements e1,e2e_1, e_2. A mapping ff of G1G_1 into G2G_2 is called a morphism if ff satisfies the following conditions: (i) ff is analytic; (ii) f(e1)=e2f(e_1) = e_2; (iii) if g,hg, h are elements of G1G_1 such that ghgh is defined, then f(g)f(h)f(g)f(h) is defined and is equal to f(gh)f(gh).

Let g∈G1g \in G_1. As gg−1gg^{-1} is defined and is equal to e1e_1, f(g)f(g−1)f(g)f(g^{-1}) is defined and is equal to e2e_2 and hence

f(g)−1=f(g)−1(f(g)f(g−1))=(f(g)−1f(g))f(g−1) f(g)^{-1} = f(g)^{-1}(f(g)f(g^{-1})) = (f(g)^{-1}f(g))f(g^{-1})
that is
f(g)−1=f(g−1). f(g)^{-1} = f(g^{-1}).
(11)

The composition of two morphisms is a morphism.

If f:G1→G2f : G_1 \to G_2 and f′:G2→G1f' : G_2 \to G_1 are two inverse morphisms of one another, they are isomorphisms (using in particular formula (11)).

Let G1,G2G_1, G_2 be two Lie group germs, and ff a mapping of G1G_1 into G2G_2 satisfying conditions (ii) and (iii) above, which is analytic in an open neighbourhood of e1e_1. Using Proposition 20, it can be proved as in Proposition 4 that ff is a morphism.

Let (G,e,θ,m)(G, e, \theta, m) be a Lie group germ and Ω\Omega the set of definition of mm. Let HH be a submanifold of GG containing ee, which is stable under θ\theta. Suppose that the set Ω1\Omega_1 of (x,y)∈Ω∩(H×H)(x, y) \in \Omega \cap (H \times H) such that m(x,y)∈Hm(x, y) \in H is open in H×HH \times H. Then (H,e,θ∣H,m∣Ω1)(H, e, \theta|H, m|\Omega_1) is a Lie group germ. Such a Lie group germ is called a Lie subgroup germ of G. The canonical injection of HH into GG is a morphism. If f:L→Gf : L \to G is a morphism of Lie group germs such that f(L)⊂Hf(L) \subset H, then f:L→Hf : L \to H is a morphism of Lie group germs.

Suppose that KK is of characteristic 0. If we replace the hypothesis that HH is a submanifold of GG by the hypothesis that HH is a quasi-submanifold of GG, the results of the above paragraph remain true (cf. Differentiable and Analytic Manifolds, R, 5.8.5). HH is then called a Lie quasi-subgroup germ of GG.

If GG is a Lie group germ with identity element ee, every symmetric open neighbourhood of ee in GG is a Lie subgroup germ of GG. (This applies in particular when GG is a Lie group.) Let HH be a Lie subgroup germ of GG; if HH is a neighbourhood of ee in GG, then HH is open in GG by Proposition 20.

The product Lie group germ of a finite number of Lie group germs is defined in an obvious way.

Proposition 21

Let G,HG, H be two Lie group germs and ϕ\phi a morphism of GG into HH. The following conditions are equivalent:

(i) ϕ\phi is étale at ee; (ii) there exist open Lie subgroup germs G′,H′G', H' of G,HG, H such that ϕ∣G′\phi|G' is an isomorphism of G′G' onto H′H'.

The implication (ii) ⇒\Rightarrow (i) is obvious. Suppose that ϕ\phi is étale at ee. There exists an open Lie subgroup germ G1G_1 of GG such that ϕ(G1)\phi(G_1) is open in HH and ϕ∣G1\phi|G_1 is an isomorphism of the manifold G1G_1 onto the manifold ϕ(G1)\phi(G_1). Then there exists an open Lie subgroup germ G′G' of G1G_1 such that the product in GG of two elements of G′G' is always defined and belongs to G1G_1. If g,g′g, g' are elements of G′G' such that gg′∈G′gg' \in G', then ϕ(g)ϕ(g′)=ϕ(gg′)∈ϕ(G′)\phi(g)\phi(g') = \phi(gg') \in \phi(G'); if g,g′g, g' are elements of G′G' such that gg′∈G1−G′gg' \in G_1 - G', then

ϕ(g)ϕ(g′)=ϕ(gg′)∈ϕ(G1)−ϕ(G′). \phi(g)\phi(g') = \phi(gg') \in \phi(G_1) - \phi(G').

Hence ϕ∣G′\phi|G' is an isomorphism of the Lie group germ G′G' onto the open Lie subgroup germ ϕ(G′)\phi(G') of HH.

If the conditions of Proposition 21 hold, GG and HH are called locally isomorphic.

Proposition 22

Let HH be a Lie group, UU a Lie subgroup germ of HH and NN the set of g∈Hg \in H such that UU and gUg−1gUg^{-1} have the same germ at ee (General Topology, Chapter I, § 6, no. 10). Then NN is a subgroup of HH containing UU. There exists one and only one analytic manifold structure on NN with the following properties:

(i) NN with this structure is a Lie group; (ii) UU is an open submanifold of NN; (iii) the canonical injection of NN into HH is an immersion.

Clearly NN is a subgroup of HH. If g∈Ug \in U, then ge∈Uge \in U and geg−1∈Ugeg^{-1} \in U, hence gu∈Ugu \in U and gug−1∈Ugug^{-1} \in U for uu sufficiently close to ee in UU and hence the germ of gUg−1gUg^{-1} at ee is contained in that of UU; exchanging gg and g−1g^{-1}, we see that the germs of gUg−1gUg^{-1} and UU at ee are equal. Hence U⊂NU \subset N.

Let VV be an open neighbourhood of ee in UU such that V=V−1,V2⊂UV = V^{-1}, V^2 \subset U. Conditions (i), (ii), (iii) of Proposition 18 of no. 9 (where GG is replaced by NN) are satisfied. Hence there exists an analytic manifold structure on NN with the following properties: (α\alpha) NN with this structure is a Lie group; (β\beta) VV is open in NN; (γ\gamma) the manifold structures on NN and UU induce the same structure on VV. Since VV is a submanifold of HH, the canonical injection of NN into HH is an immersion at ee and hence at every point of NN. Let u∈Uu \in U. There exists an open neighbourhood V' of e in V such that the mapping v↦uvv \mapsto uv is an analytic isomorphism of V' onto an open neighbourhood of u in U (Proposition 20) and at the same time onto an open neighbourhood of u in N. Hence U is open in N and the identity mapping of U is an isomorphism with the given manifold structure on U and the open submanifold structure on N; in other words, U is an open submanifold of N.

Finally, we consider an analytic manifold structure on N with properties (i) and (ii) of the proposition and let N* be the Lie group thus obtained. Then the identity mapping of N into N* is étale at e and hence a Lie group isomorphism. This proves the uniqueness assertion of the proposition.

Let H be a Lie group, U a Lie quasi-subgroup germ of H and N the set of g∈Hg \in H such that U and gUg−1gUg^{-1} have the same germ at e. If K is of characteristic 0, there exists on G one and only one manifold structure with properties (i) and (ii) of Proposition 22. The proof is the same as for Proposition 22.

Corollary

Preserving the notation of Proposition 22, let G be the subgroup of H generated by U. Then G is an open subgroup of N. There exists one and only one Lie group structure on G such that U is an open submanifold of G and the canonical injection of G into H is an immersion.

Remark

Preserving the notation of Proposition 22 and its corollary, suppose that K is of characteristic 0, that H is finite-dimensional and that the topology on U admits a countable base. Even with all these hypotheses it is possible that G is not closed in H (Exercise 3). But, if G is closed, G is a Lie subgroup of H. For the mapping (g,h)↦gh(g, h) \mapsto gh is a law of analytic left operation of G on H. The orbit of e is G. Our assertion then follows from Propositions 2 (iv) and 14 (iii).

11. Law chunks of operation

Let (G,e,θ,m)(G, e, \theta, m) be a Lie group germ and X a manifold of class CrC^r.

Definition 6

A law chunk of left operation of class CrC^r of G on X is a mapping ψ\psi defined on an open subset Ω\Omega of G×XG \times X containing {e}×X\{e\} \times X, with values in X and with the following properties: (i) ψ\psi is of class CrC^r; (ii) for all x∈Xx \in X, ψ(e,x)=x\psi(e, x) = x; (iii) there exists a neighbourhood Ω1\Omega_1 of {e}×{e}×X\{e\} \times \{e\} \times X in G×G×XG \times G \times X such that, for (g,g′,x)∈Ω1(g, g', x) \in \Omega_1, the elements of m(g,g′)m(g, g'), ψ(m(g,g′),x)\psi(m(g, g'), x), ψ(g,ψ(g′,x))\psi(g, \psi(g', x)) are defined and ψ(g,ψ(g′,x))=ψ(m(g,g′),x)\psi(g, \psi(g', x)) = \psi(m(g, g'), x).

Law chunks of right operation of class CrC^r are defined similarly. We often write gxgx instead of ψ(g,x)\psi(g, x). Let G' be a Lie subgroup germ of G and X' a submanifold of X. Suppose that the set Ω′\Omega' of (g,x)∈Ω∩(G′×X′)(g, x) \in \Omega \cap (G' \times X') such that ψ(g,x)∈X′\psi(g, x) \in X' is open in G′×X′G' \times X' (a condition which is always fulfilled if X′X' is open in XX). Then ψ∣Ω′\psi|_{\Omega'} is a law chunk of left operation of class CrC^r of G′G' on X′X', which is said to be derived from ψ\psi by restriction to G′G' and X′X'.

Proposition 23

Let (G,e,θ,m)(G, e, \theta, m) be a Lie group germ, XX a manifold of class CrC^r, x0x_0 a point of XX, Ω\Omega an open neighbourhood of (e,x0)(e, x_0) in G×XG \times X and ψ\psi a mapping of Ω\Omega into XX with the following properties: (i) ψ\psi is of class CrC^r; (ii) ψ(e,x)\psi(e, x) is equal to xx for xx sufficiently close to x0x_0; (iii) ψ(m(g,g′),x)=ψ(g,ψ(g′,x))\psi(m(g, g'), x) = \psi(g, \psi(g', x)) for (g,g′,x)(g, g', x) sufficiently close to (e,e,x0)(e, e, x_0). Then there exist an open neighbourhood X′X' of x0x_0 in XX and an open subset Ω′\Omega' of Ω∩(G×X′)\Omega \cap (G \times X') such that ψ∣Ω′\psi|_{\Omega'} is a law chunk of left operation of class CrC^r of GG on X′X'. There exist an open neighbourhood X′X' of x0x_0 in XX and an open neighbourhood G′G' of ee in GG such that ψ(e,x)=x\psi(e, x) = x for all x∈Xx \in X, and

ψ(g,ψ(g′,x))=ψ(m(g,g′),x) \psi(g, \psi(g', x)) = \psi(m(g, g'), x)
for (g,g′,x)∈G′×G′×X′(g, g', x) \in G' \times G' \times X'. Let Ω′\Omega' be the set of (g,x)∈Ω∩(G′×X′)(g, x) \in \Omega \cap (G' \times X') such that ψ(g,x)∈X′\psi(g, x) \in X'. Then Ω′\Omega' is open in G×X′G \times X' and X′X', Ω′\Omega' have the properties of the proposition.

Lemma 3

Let XX be a normal space and (Xi)i∈I(X_i)_{i \in I} a locally finite open covering of XX. For all (i,j)∈I×I(i, j) \in I \times I and all x∈Xi∩Xjx \in X_i \cap X_j, let Vij(x)V_{ij}(x) be a neighbourhood of xx contained in Xi∩XjX_i \cap X_j. Then we can associate with every x∈Xx \in X a neighbourhood V(x)V(x) of xx such that the following conditions are fulfilled: (a) the relation x∈Xi∩Xjx \in X_i \cap X_j implies V(x)⊂Vij(x)V(x) \subset V_{ij}(x); (b) if V(x)V(x) and V(y)V(y) meet, there exists i∈Ii \in I such that V(x)∪V(y)⊂XiV(x) \cup V(y) \subset X_i. There exists an open covering (Xi′)i∈I(X'_i)_{i \in I} of XX such that Xi′‾⊂Xi\overline{X'_i} \subset X_i for all i∈Ii \in I (General Topology, Chapter IX, § 4, Theorem 3). Let x∈Xx \in X. Let V1(x)V_1(x) be the intersection of the Vij(x)V_{ij}(x) and the Xk′X'_k which contain xx; this is an open neighbourhood of xx. Let V2(x)V_2(x) be a neighbourhood of xx contained in V1(x)V_1(x) and meeting only a finite number of XiX_i. Then V2(x)V_2(x) meets only a finite number of Xi′‾\overline{X'_i} and hence the set

V(x)=V2(x)∩⋂i∈I,x∉Xi′‾(X−Xi′‾) V(x) = V_2(x) \cap \bigcap_{i \in I, x \notin \overline{X'_i}} (X - \overline{X'_i})
is a neighbourhood of xx. If x∈Xi∩Xjx \in X_i \cap X_j, then V1(x)⊂Xi∩XjV_1(x) \subset X_i \cap X_j and hence V(x)⊂Xi∩XjV(x) \subset X_i \cap X_j. Let x,yx, y be in XX and suppose that V(x)V(x) and V(y)V(y) meet. There exists i∈Ii \in I such that x∈Xi′x \in X'_i. Then V1(x)⊂Xi′V_1(x) \subset X'_i, hence V(x)⊂Xi′V(x) \subset X'_i and hence V(y)∩Xi′‾≠∅V(y) \cap \overline{X'_i} \neq \emptyset. Then y∈Xi′‾y \in \overline{X'_i} by definition of V(y)V(y), whence y∈Xiy \in X_i and V(y)⊂XiV(y) \subset X_i. Thus XiX_i contains V(x)V(x) and V(y)V(y).

§ 2.1

Proposition 24

Let G be a Lie group germ, X a manifold of class C^r and (X_i)_{i \in I} a locally finite open covering of X. For all i \in I, let ψi\psi_i be a law chunk of left operation of class C^r of G on X_i. Suppose that the underlying topological space of X is normal and that, for all (i, j) \in I \times I and all x \in X_i \cap X_j, ψi\psi_i and ψj\psi_j coincide on a neighbourhood of (e, x). There exists a law chunk ψ\psi of left operation of class C^r of G on X such that, for all i \in I and all x \in X_i, ψi\psi_i and ψ\psi coincide on a neighbourhood of (e, x).

For all (i, j) \in I \times I and all x \in X_i \cap X_j, choose an open neighbourhood V_{ij}(x) of x in X_i \cap X_j such that ψi\psi_i and ψj\psi_j are defined and equal on a neighbourhood of {e} \times V_{ij}(x) in G \times X. For all x \in X choose an open neighbourhood V(x) of x in X such that conditions (a) and (b) of Lemma 3 are fulfilled. Let I_x be the set of i \in I such that x \in X_i. This is a finite set. Let U_x be the set of (g, y) \in G \times V(x) such that the ψi\psi_i for i \in I_x are defined and coincide on a neighbourhood of (g, y). Then U_x is open and (e, x) \in U_x. The ψi\psi_i for i \in I_x all have the same restriction to U_x. Let x, y be in X. If U_x and U_y meet, V(x) and V(y) meet and hence there exists i \in I such that

V(x)∪V(y)⊂Xi. V(x) \cup V(y) \subset X_i.

Then i \in I_x, i \in I_y, ψi∣Ux=ψx,ψi∣Uy=ψy\psi_i|U_x = \psi_x, \psi_i|U_y = \psi_y and hence

ψx∣(Ux∩Uy)=ψy∣(Ux∩Uy). \psi_x|(U_x \cap U_y) = \psi_y|(U_x \cap U_y).

The ψx\psi_x therefore define a mapping ψ\psi of U=⋃x∈XUxU = \bigcup_{x \in X} U_x into X and U is an open neighbourhood of {e} \times X in G \times X. Clearly ψ\psi is of class C^r and ψ(e,x)=x\psi(e, x) = x for all x \in X. For all i \in I and all x \in X_i, ψ\psi coincides with ψx\psi_x and hence with ψi\psi_i in a neighbourhood of (e, x) and hence ψ\psi satisfies condition (iii) of Definition 6.

Exercises

See the exercises for § 1.

Transcribed from the printed volume. Not translated.