Bourbaki

§ 2. Group of tangent vectors to a Lie group

1. Tangent laws of composition

Let X and Y be manifolds of class C^r. We know (Differentiable and Analytic Manifolds, R, 8.1.4) that X \times Y is a manifold of class C^r and that the mapping (T(pr_1), T(pr_2)), the product of the tangent mappings to the canonical projections, is an isomorphism of class C^{r-1} of T(X \times Y) onto T(X) \times T(Y).\footnote{For r = 1, this means that (T(pr_1), T(pr_2)) is a homomorphism of T(X \times Y) onto T(X) \times T(Y).} This isomorphism is compatible with the vector bundle structures with base space X \times Y and allows us to identify T(X \times Y) with T(X) \times T(Y). Let a ∈ X, b ∈ Y, u ∈ T_a(X), v ∈ T_b(Y); the above identification allows us to consider (u, v) as an element of T_{(a, b)}(X × Y); then

(u,v)=(u,0)+(0,v)(u, v) = (u, 0) + (0, v)

and (u, 0) (resp. (0, v)) is the image of u (resp. v) under the tangent mapping to the immersion x ↦ (x, b) (resp. y ↦ (a, y)) of X (resp. Y) into X × Y. When it is necessary to be precise, we shall write 0_a for the zero element of T_a(x).

Now let X, Y, Z be manifolds of class C^r and f a mapping of class C^r of X × Y into Z. The tangent mapping is, using the above identification, a mapping of class C^{r-1} of T(X) × T(Y) into T(Z). For u ∈ T_a(X) and v ∈ T_b(Y),

(1)

T(f)(u,v)=T(f)(u,0b)+T(f)(0a,v),T(f)(u, v) = T(f)(u, 0_b) + T(f)(0_a, v),
(2)
T(f)(0a,0b)=0f(a,b).T(f)(0_a, 0_b) = 0_{f(a, b)}.

On the other hand, the mapping y ↦ f(a, y) is the composition of the immersion y ↦ (a, y) and f; it follows that

(3) T(f)(0,v)T(f)(0, v) is the image of v under the tangent mapping to y↦f(a,y)y \mapsto f(a, y).

Similarly

(4) T(f)(u,0)T(f)(u, 0) is the image of u under the tangent mapping to x↦f(x,b)x \mapsto f(x, b).

If the mapping f of X × Y into Z is denoted by (x, y) ↦ xy, uv is often used to denote the element T(f)(u, v) for u ∈ T(X), v ∈ T(Y).

Let X be a manifold of class C^r and m: X × X → X a law of composition of class C^r on X. Then T(m) is a law of composition of class C^{r-1} on T(X). It is called the law of composition tangent to m. The canonical projection p of T(X) onto X is compatible with the laws m and T(m); in other words,

(5)

p∘T(m)=m∘(p×p).p \circ T(m) = m \circ (p \times p).

It follows from (2) that

(6)

T(m)(0x,0y)=0m(x,y)T(m)(0_x, 0_y) = 0_{m(x, y)}

for all x, y in X; in other words, the zero section x↦0xx \mapsto 0_x of T(X) is compatible with the laws m and T(m).

Proposition 1

Let X be a manifold of class C^r and m a law of composition of class C^r on X. If m is associative (resp. commutative), then T(m) is associative (resp. commutative).

If m is associative, then m∘(m×IdX)=m∘(IdX×m)m \circ (m \times \mathrm{Id}_X) = m \circ (\mathrm{Id}_X \times m), whence

T(m)∘(T(m)×IdT(X))=T(m)∘(IdT(X)×T(m))T(m) \circ (T(m) \times \mathrm{Id}_{T(X)}) = T(m) \circ (\mathrm{Id}_{T(X)} \times T(m))

and hence T(m)T(m) is associative. Let ss be the mapping (x,y)↦(y,x)(x, y) \mapsto (y, x) of X×XX \times X into X×XX \times X. If mm is commutative, then m∘s=mm \circ s = m and hence

T(m)∘T(s)=T(m). T(m) \circ T(s) = T(m).

But T(s)T(s) is the mapping (u,v)↦(v,u)(u, v) \mapsto (v, u) of T(X)×T(X)T(X) \times T(X) into T(X)×T(X)T(X) \times T(X). Hence T(m)T(m) is commutative.

Proposition 2

Let XX be a manifold of class CrC^r, mm a law of composition of class CrC^r on XX and ee an identity element for mm.

(i) *The vector 0e0_e is an identity element for T(m)T(m). (ii) *Te(X)T_e(X) is stable under T(m)T(m) and the law of composition induced on Te(X)T_e(X) by T(m)T(m) is the vector space addition on Te(X)T_e(X). (iii) *Let UU be an open subset of XX and α\alpha a mapping of class CrC^r of UU into XX such that, for all x∈Ux \in U, α(x)\alpha(x) is the inverse of xx under mm. Then, for all u∈T(U)u \in T(U), T(α)uT(\alpha)u is the inverse of uu under T(m)T(m).

Properties (3) and (4) show that T(m)(0e,u)=T(m)(u,0e)=uT(m)(0_e, u) = T(m)(u, 0_e) = u for all u∈T(X)u \in T(X), whence (i). For u,vu, v in Te(X)T_e(X),

T(m)(u,v)=T(m)(u,0e)+T(m)(0e,v)=u+v, T(m)(u, v) = T(m)(u, 0_e) + T(m)(0_e, v) = u + v,

whence (ii). Finally the relations m(x,α(x))=m(α(x),x)=em(x, \alpha(x)) = m(\alpha(x), x) = e for all x∈Ux \in U imply

T(m)(u,T(α)(u))=T(m)(T(α)u,u)=0e T(m)(u, T(\alpha)(u)) = T(m)(T(\alpha)u, u) = 0_e

for all u∈T(U)u \in T(U), whence (iii).

Proposition 3

Let X1,X2,…,Xp,YX_1, X_2, \ldots, X_p, Y be manifolds of class CrC^r, ii an integer of {1,p}\{1, p\}, mim_i (resp. nn) a law of composition of class CrC^r on XiX_i (resp. YY) and uu a mapping of class CrC^r of X1×X2×⋯×XpX_1 \times X_2 \times \cdots \times X_p into YY. If uu is distributive relative to the variable of index ii, then T(u)T(u) is distributive relative to the variable of index ii.

The proof is analogous to that of Proposition 1.

2. Group of tangent vectors to a Lie group

Proposition 4

Let GG be a Lie group. Then T(G)T(G), with the law of composition tangent to the multiplication of GG, is a Lie group. The identity element of T(G)T(G) is the vector 0e0_e.

This follows from Propositions 1 and 2.

Proposition 5

Let GG and HH be Lie groups and ff a morphism of GG into HH. Then T(f)T(f) is a morphism of the Lie group T(G)T(G) into the Lie group T(H)T(H).

We know that T(f′)T(f') is analytic. On the other hand, let mm (resp. nn) denote the multiplication on GG (resp. HH). Then f∘m=n∘(f×f)f \circ m = n \circ (f \times f), whence

T(f)∘T(m)=T(n)∘(T(f)×T(f)), T(f) \circ T(m) = T(n) \circ (T(f) \times T(f)),

which expresses the fact that T(f)T(f) is a group homomorphism.

Corollary

Let G1,…,GnG_1, \ldots, G_n be Lie groups. The canonical isomorphism of the manifold T(G1×⋯×Gn)T(G_1 \times \cdots \times G_n) onto the manifold T(G1)×⋯×T(Gn)T(G_1) \times \cdots \times T(G_n) is a Lie group isomorphism.

pri\mathrm{pr}_i is a morphism of G1×⋯×GnG_1 \times \cdots \times G_n into GiG_i and hence T(pri)T(\mathrm{pr}_i) is a morphism of T(G1×⋯×Gn)T(G_1 \times \cdots \times G_n) into T(Gi)T(G_i).

Proposition 6

Let GG be a Lie group.* (i) The canonical projection p:T(G)→Gp : T(G) \to G is a Lie group morphism. (ii) The kernel of pp is Te(G)T_e(G). It is a Lie subgroup of T(G)T(G). The Lie group structure induced on Te(G)T_e(G) by that on T(G)T(G) is the Lie group structure of the complete normable space Te(G)T_e(G). (iii) The zero section ss is an isomorphism of the Lie group GG onto a Lie subgroup s(G)s(G) of T(G)T(G) (which subgroup we identify with GG). (iv) *The Lie group T(G)T(G) is the semi-direct product of GG by Te(G)T_e(G).

Assertion (i) follows from (5). Assertion (ii) is obvious taking account of Proposition 2 (ii). Assertions (iii) and (iv) follow from (6) and § 1, Proposition 8.

Let u∈T(G)u \in T(G) and g∈Gg \in G. By (3) and (4), the products ug,guug, gu calculated in the group T(G)T(G) are the images of uu under T(δ(g−1))T(\delta(g^{-1})) and T(γ(g))T(\gamma(g)). It follows from § 1, Corollary 2 to Proposition 17 that the mapping (g,u)↦gu(g, u) \mapsto gu of G×Te(G)G \times T_e(G) into T(G)T(G) is an isomorphism of the trivial vector bundle G×Te(G)G \times T_e(G) with base space GG onto the vector bundle T(G)T(G). The inverse isomorphism is called the left trivialization of T(G)T(G). By considering the mapping (g,u)↦ug(g, u) \mapsto ug, the right trivialization of T(G)T(G) is defined similarly.

Proposition 7

Let GG be a Lie group, MM a manifold of class CrC^r and ff and gg mappings of class CrC^r of MM into GG, so that fgfg is a mapping of class CrC^r of MM into GG. Let m∈M,x=f(m),y=g(m),u∈Tm(M)m \in M, x = f(m), y = g(m), u \in T_m(M). Then

(Tfg)u=T(f)u.y+x.T(g)u. (T fg)u = T(f)u.y + x.T(g)u.

Let mm be the multiplication of GG. Then fg=m∘(f,g)fg = m \circ (f, g). Now

T(f,g)(u)=(T(f)u,T(g)u), T(f, g)(u) = (T(f)u, T(g)u),

hence T(fg)u=T(f)u.T(g)uT(fg)u = T(f)u.T(g)u. It then suffices to apply (1) with ff replaced by mm.

Corollary

Let n∈Zn \in \mathbf{Z}. The tangent mapping at ee to the mapping g↦gng \mapsto g^n of GG into GG is the mapping x↦nxx \mapsto nx of Te(G)T_e(G) into Te(G)T_e(G).

For n⩾0n \geqslant 0, this follows by induction on nn from Proposition 7. On the other hand, the tangent mapping at ee to the mapping g↦g−1g \mapsto g^{-1} is the mapping x↦−xx \mapsto -x (no. 1, Proposition 2).

Let G be a Lie group, X a manifold of class C^r and (g, x) \mapsto gx a law of left operation of class C^r of G on X. Arguing as for Proposition 1, we derive a law of left operation of class C^{r-1} of T(G) on T(X), which we shall also denote by (u, v) \mapsto uv. Identifying G (resp. X) with the image of the zero section of T(G) (resp. T(X)), we see by (6) that the law of left operation of T(G) on T(X) extends the law of left operation of G on X. For all u \in T_g(G) and v \in T_x(X), by (1),

uv=gv+ux. uv = gv + ux.

If g \in G and v \in T_x(X), gv is, by (3), the image of v under the tangent mapping at x to the mapping y \mapsto gy of X into X. This tangent mapping is an isomorphism of T_x(X) onto T_{gx}(X). In particular,

g(v+v′)=gv+gv′,g(λv)=λ(gv)for v,v′ in Tx(X),λ∈K. g(v + v') = gv + gv', \quad g(\lambda v) = \lambda(gv) \quad \text{for } v, v' \text{ in } T_x(X), \lambda \in K.

If x \in X and u \in T_g(G), ux is by (4) the image of u under the tangent mapping at g to the mapping h \mapsto hx of G into X. Hence

(u+u′)x=ux+u′x,(λu)x=λ(ux)for u,u′ in Tg(G),λ∈K. (u + u')x = ux + u'x, \quad (\lambda u)x = \lambda(ux) \quad \text{for } u, u' \text{ in } T_g(G), \lambda \in K.

The above can be applied to the case of a Lie group operating on itself by left (resp. right) translation. The corresponding law of operation of T(G) on T(G) is defined by left (resp. right) translation of the Lie group T(G). Formulae (7), (8) and (9) are therefore valid in T(G).

Proposition 8

Let G_1 and G_2 be Lie groups, X_1 and X_2 manifolds of class C^r and f_i a law of left operation of class C^r of G_i on X_i (i = 1, 2). Let \phi be a morphism of G_1 into G_2 and \psi a \phi-morphism of X_1 into X_2. Then T(\psi) is a T(\phi)-morphism of T(X_1) into T(X_2).

f_2 \circ (\phi \times \psi) = \psi \circ f_1, whence

T(f2)∘(T(ϕ)×T(ψ))=T(ψ)∘T(f1). T(f_2) \circ (T(\phi) \times T(\psi)) = T(\psi) \circ T(f_1).

Let G be a Lie group, X a manifold of class C^r and (g, x) \mapsto gx a law of left operation of class C^r of G on X. Let I be an open subset of K containing 0 and \gamma: I \to G a mapping of class C^r such that \gamma(0) = e. Let

a=T0(γ)1∈Te(G). a = T_0(\gamma)1 \in T_e(G).

Let x \in X. Using (4), ax is the image under the tangent mapping to \lambda \mapsto \gamma(\lambda)x of the tangent vector 1 to I at 0. Hence the vector field x \mapsto ax on X is the vector field defined by the mapping (\lambda, x) \mapsto \gamma(\lambda)x in the sense of Differentiable and Analytic Manifolds, R, 8.4.5.

3. Case of group germs

Let (G, e, \theta, m) be a Lie group germ and \Omega the set of definition of m. Then T(\Omega) is identified with an open subset of T(G) \times T(G) and T(m) is an analytic mapping of T(Ω)T(\Omega) into T(G)T(G). It can be verified as in no. 2 that (T(G),0e,T(θ),T(m))(T(G), 0_e, T(\theta), T(m)) is a Lie group germ. The products of GG and T(G)T(G) are often written multiplicatively. The canonical projection of T(G)T(G) onto GG is a morphism of Lie group germs. The restriction of Te(m)T_e(m) to Te(G)T_e(G) is the vector space addition of Te(G)T_e(G). The zero section of T(G)T(G) is an isomorphism of the Lie group germ GG onto a Lie subgroup germ of T(G)T(G) which we identify with GG. If ff is a morphism of GG into a Lie group germ HH, T(f):T(G)→T(H)T(f): T(G) \to T(H) is a morphism of Lie group germs.

The mapping ϕ:(g,u)↦gu\phi: (g, u) \mapsto gu of G×Te(G)G \times T_e(G) into T(G)T(G) is an isomorphism of the trivial vector bundle G×Te(G)G \times T_e(G) with base space GG onto the vector bundle T(G)T(G); for ϕ\phi and ϕ−1\phi^{-1} are analytic and are vector bundle morphisms, so that it suffices to apply Differentiable and Analytic Manifolds, R, 7.2.1. (The proof of no. 2 could also be adapted.) The isomorphism ϕ−1\phi^{-1} is called the left trivialization of T(G)T(G). The inverse isomorphism of the mapping (g,u)↦ug(g, u) \mapsto ug is called the right trivialization.

Let XX be a manifold of class CrC^r and ψ\psi a law chunk of left operation of class CrC^r of GG on XX. Then T(ψ)T(\psi) is a law chunk of left operation of class Cr−1C^{r-1} of T(G)T(G) on T(X)T(X) extending ψ\psi. Formulae (7), (8) and (9) remain valid if gxgx is defined. If II is an open subset of KK containing 0, if γ:I→G\gamma: I \to G is a mapping of class CrC^r such that γ(0)=e\gamma(0) = e and if a=T0(γ)1a = T_0(\gamma)1, the vector field x↦axx \mapsto ax defined on XX is the vector field defined by the mapping (λ,x)↦γ(λ)x(\lambda, x) \mapsto \gamma(\lambda)x in the sense of Differentiable and Analytic Manifolds, R, 8.4.5.

Transcribed from the printed volume. Not translated.