— or that the varieties under consideration are locally of finite dimension.
In the latter case, the expressions “Banach space, vector bundle, vector functor” mean respectively “finite-dimensional vector space, vector bundle of finite rank, vector functor in finite dimension with values of finite dimension”.
8.1. Tangent bundle
8.1.1. — Let r∈NK and let X be a K-variety of class Cr. We denote by T(X) the sum set of the tangent spaces Tx(X) (5.5.1) for x∈X, and by π the canonical mapping of T(X) onto X. Let c=(U,φ,E) be a chart of the variety X; for x∈U and t∈Tx(X), put ζc(x,t)=(x,θc−1(t)), the mapping θc being that defined in 5.5.1. The triplet c′=(U,ζc,E) is a vector chart (7.1.1) of T(X) (endowed with the mapping π:T(X)⟶X). There exists on T(X) one and only one structure of vector bundle with base X and of class Cr−1 such that, for every chart c of X, the corresponding vector chart c′ is a vector chart of T(X)[^1]. Endowed with this structure, T(X) is called the tangent bundle of X. In particular, T(X) is endowed with a structure of variety of class Cr−1 and the triplet (T(X),X,π) is a fibration (7.1.4).
Let U be an open subset of a Banach space E and let i be the canonical injection of U into E. The vector chart c′=(U,ζc,E) associated with the chart c=(U,i,E) of U defines an isomorphism of T(U) onto the trivial bundle EU (7.1.5); these two bundles are identified by means of this isomorphism.
If X is a purely variety of type E (5.1.7), T(X) is a purely variety of type E×E.
8.1.2. — Let f:X⟶Y be a morphism of varieties of class Cr. We define an f-morphism T(f):T(X)⟶T(Y) of vector bundles of class Cr−1 by
T(f)(x,t)=(f(x),Txf(t))
for x∈X and t∈Tx(X).
If f=IdX, one has T(f)=IdT(X). For f:X⟶Y and g:Y⟶Z, one has
T(g∘f)=T(g)∘T(f).
8.1.3. — Let f:X⟶Y be a morphism of manifolds of class Cr, and let f′:T(X)⟶f∗T(Y)
be the unique X-morphism such that T(f) = g ∘ f', where g is the canonical morphism f*T(Y) → T(Y) (cf. 7.2.4).
For f to be an immersion (resp. a submersion), it is necessary and sufficient that 0 → T(X) f′f∗T(Y) (resp. T(X) f′f∗T(Y)→0) be a locally direct exact sequence (7.5.6). If f is an immersion, the cokernel bundle of f' is called the normal or transverse bundle of f. If f is a submersion, the kernel bundle of f is called the tangent bundle to the fibres of f, or the relative tangent bundle of X over Y, and is denoted T(X/Y). Its fibre T_x(X/Y) at x ∈ X is the tangent space at x to the submanifold f−1(f(x)) of X. The sequence
0⟶Tx(X/Y)iTx(X)Tx(f)Tf(x)(Y)⟶0
where i is the canonical injection, is exact.
For f to be étale, it is necessary and sufficient that f' be an isomorphism.
When K has characteristic 0, for f to be a subimmersion, it is necessary and sufficient that f' be locally direct.
8.1.4. Let X_1 and X_2 be two manifolds. We have T(X_1 × X_2) = p_1^*T(X_1) ⊕ p_2^*T(X_2), where p_1 and p_2 are the canonical projections (5.6.3).
8.1.5. Let n be an integer ⩾0; the tangent bundle T(K^n) of K^n is identified with the trivial bundle with fibre K^n (8.1.1). The constant sections defined by the elements of the canonical basis of K^n form a frame of T(K^n).
If X is a manifold and (u^1, ..., u^n) a system of coordinates of X in an open set U of X (5.1.10), the preceding frame defines by transport of structure a frame of T(X) on U; it is called the tangent frame defined by (u^1, ..., u^n) and is denoted (\partial/\partial u^1, ..., \partial/\partial u^n) (cf. 5.5.8).
8.2. Vector fields
Let X be a manifold of class C^r, and let U be an open set of X.
8.2.1. A section (not necessarily continuous, cf. Notations and Conventions) of the tangent bundle T(X) above U is called a vector field on U.
8.2.2. The dual bundle T'(X) (7.7.3) of T(X) is called the cotangent bundle of X. Its sections are called covector fields. If f is a function of class C^k on X (with 1 ≤ k ≤ r), with values in K, its differential df : x ↦ d_x f (5.5.6) is a covector field of class C^{k-1}. More generally, if f is a function of class C^k on X (1 ≤ k ≤ r), with values in a Banach space E, its differential df is a section of class C^{k-1} of the bundle L(T(X);EX).
8.2.3. Let ξ be a vector field of class C^{r-1} on X, and let f∈Cr(X;F) be a function of class C^r with values in a Banach space F. We denote by ⟨ξ,df⟩, or D_\xi(f), or again ξ(f) the function x↦dxf(ξ(x)); it is an element of Cr−1(X;F). For fixed f, the mapping ξ↦Dξ(f) is Cr−1(X)-linear. If Dξf=0 for every function f of class C^r with values in a Banach space, defined in an open set of X, we have ξ=0.
Let E, F, G be Banach spaces and let (u, v) ↦ u . v be a continuous bilinear mapping of E×F into G. Let f∈Cr(X;E), g∈Cr(X;F) and let f.g∈Cr(X;G) be their product. If ξ is a vector field of class Cr−1 on X, we have
Dξ(f.g)=Dξ(f).g+f.Dξ(g).
In particular, the mapping Dξ:Cr(X)→Cr−1(X) is a derivation of the algebra Cr(X) into the Cr(X)-module Cr−1(X).
8.2.4. Suppose that X is locally of finite dimension and that one of the following two hypotheses is satisfied:
(i) r=∞ and X is separated; (ii) r=ω and X is isomorphic to an open polydisc of Kn.
Then, for every derivation D of the algebra Cr(X), there exists one and only one vector field ξ of class Cr such that D=Dξ.
8.2.5. Let (u1,…,un) be a system of coordinates of X in U, and ξ a vector field on U. The i-th coordinate of ξ in the tangent frame (8.1.5) defined by (u1,…,un) is Dξ(ui). In other words, we have
ξ=1≤i≤n∑Dξ(ui)⋅∂/∂ui.
8.2.6. Let φ:X→Y be a morphism of manifolds of class Cr, ξ a vector field on X and η a vector field on Y. We say that ξ is φ-*related* to η if, for every x∈X, we have η(φ(x))=Tx(φ)(ξ(x)). Suppose this is so, and that ξ and η are of class Cr−1. Then, for every Banach space E, the diagram
Cr(Y;E)↓DηCr−1(Y;E)φ∗φ∗Cr(X;E)↓DξCr−1(X;E)
where φ∗(f)=f∘φ, is commutative.
When φ is étale, for every vector field η on Y, there exists one and only one vector field on X which is φ-*related* to η; it is denoted by φ∗η.
8.2.7. Let φ:X→Y be a morphism of manifolds of class Cr. If ω is a covector field(1) on Y, we define a covector field φ∗ω on X by
(φ∗ω)(x)=t(Tx(φ))(ω(φ(x)))(x∈X).
More generally, let τ be a vector functor (7.6) of class Cr−1, of type (I+,I−) with I+=∅ and I− reduced to one element (such a functor is said to be contravariant). If ω is a section of class Cr−1 of the vector bundle τ(T(Y)) based on Y, we define a section φ∗ω of τ(T(X)) by (φ∗ω)(x)=τ(Tx(φ))(ω(φ(x)))(x∈X).
1 The reader should take care not to confuse this use of the letter ω with that defined in Notations and Conventions.
8.3. Differential forms, exterior differentiation
Let us recall that, in this number and the following ones, it is assumed either that K is of characteristic 0, or that the manifolds considered are locally of finite dimension.
8.3.1. Let X be a manifold of class C' and F a vector bundle of class C^k based on X, with k + 1 \in \mathbf{N}_K and k + 1 \leq r. Let p be an integer ≥0 and let Alt^p(T(X); F) be the vector bundle of class C^k of alternating p-linear mappings of T(X) into F (7.8.1 and Errata to No. 7.8). A section of this bundle over an open set U of X is called an alternating differential form (or simply differential form) of degree p on U, with values in F. Those of these sections which are of class C^k form a Ck(U)-module, denoted ^k\Omega^p(U; F). If E is a Banach space, we write ^k\Omega^p(U; E) instead of ^k\Omega^p(U; E_X) and speak of differential forms with values in E. In this notation, k (resp. E) is sometimes omitted when it is equal to r − 1 (resp. K).
A differential form of degree 0 (resp. 1) with scalar values is a function (resp. a covector field).
Let φ:Y→X be a morphism of manifolds of class C', and let ω∈kΩp(X;F).Therethenexistsoneandonlyoneform\varphi^*\omega \in ^k\Omega^p(Y; \varphi^*F) such that
for every y \in Y and every family v_1, \ldots, v_p of elements of T_y(Y).
If ψ:Z→Y is a morphism of manifolds of class C', we have
(2)
(φ∘ψ)∗ω=ψ∗(φ∗ω).
When Y is a submanifold of X and φ is the canonical injection, we sometimes write ω∣Y instead of φ∗ω and say that ω∣Y is induced on Y by ω.
8.3.2. The definitions and results of No. 7.8 apply to differential forms. In particular, a pairing of vector bundles F' \times_X F'' \to F gives rise to an exterior product ^k\Omega^{p'}(U; F) \times ^k\Omega^{p''}(U; F'') \to ^k\Omega^p(U; F), where p = p' + p'' (7.8.2); it is denoted by (ω′,ω′′)↦ω′∧ω′′. The direct sum of the ^k\Omega^p(U; K) for p \geq 0 is thus endowed with a structure of associative and alternating graded algebra (A, III, p. 53).
Let ξ be a vector field on X and ω a differential form of degree p \geq 1 on X, with values in a vector bundle F; the interior product of ξ and ω is the differential form i(\xi, \omega) of degree p − 1 on X, with values in F, defined by
(3)
i(ξ,ω)x(v1,…,vp−1)=ωx(ξ(x),v1,…,vp−1)
for every x \in X and every family v_1, \ldots, v_{p-1} of elements of T_x(X). If ω is a differential form of degree 0, we put i(\xi, \omega) = 0. We also write i(\xi)\omega or i_\xi\omega instead of i(\xi, \omega). When F = K_X, the definition given above coincides with that of 7.8.4.
Let F' \times_X F'' \to F be a pairing of vector bundles. For ω′∈kΩp′(U;F′)and\omega'' \in ^k\Omega^{p''}(U; F''), we have
(4)
i(ξ)(ω′∧ω′′)=(i(ξ)ω′)∧ω′′+(−1)p′ω′∧i(ξ)ω′′.
8.3.3. Let (u1,…,un) be a system of coordinates in the open set U of X. Every element ω of kΩp(U;F) is written in a unique way
ω=i1<⋯<ip∑fi1,…,ip⋅dui1∧⋯∧duip
where the fi1,…,ip are sections of class Ck of F on U; in this formula, the sign ∧ is relative to the canonical pairing KX×XKX→KX and the point designating the product is relative to the canonical pairing F×XKX→F.
8.3.4. Let p be an integer ⩾0 and r an element of NK. Let E and F be Banach spaces, U an open set of E and α∈rΩp(U;F). Let α~∈Cr(U;Altp(E;F)) be the element corresponding to α. If x∈U, the differential dxα~ of α~ at x (5.5.6) is an element of L(E;Altp(E;F)); if t∈E, we have (dxα~)(t)∈Altp(E;F), and, if t1,…,tp are in E, we have (dxα~)(t)(t1,…,tp)∈F. There exists one and only one element dα of r−1Ωp+1(U;F) such that one has
8.3.5. Let p be an integer ⩾0 and r an element of NK. Let X be a variety of class Cr+1, F a Banach space, and ω∈rΩp(X;F). There exists one and only one differential form π∈r−1Ωp+1(X;F) such that, for every chart c=(U,φ,E) of X, if ωc is the differential form on φ(U) such that ω∣U=φ∗(ωc), one has πc=dωc, where dωc is defined by formula (5) of 8.3.4.
The differential form π is called the exterior differential of ω; it is denoted dω. It has the following properties:
(6) If ψ:Y→X is a morphism of manifolds of class Cr+1, and if ω∈rΩp(X;F), one has ψ∗(dω)=d(ψ∗ω); in particular, d commutes with the operation of restriction to a submanifold. (7) For p=0, the mapping d:rΩ0(X;F)→r−1Ω1(X;F) coincides with that defined in 8.2.2. (8) If r⩾2 and if ω∈rΩp(X;F), one has d(dω)=0.\footnote{The sign ∧ indicates that the symbol above which it is placed is to be omitted (cf. 7.8.4).} (9) For every continuous linear mapping u:F→F′ of Banach spaces, one has d(u(ω))=u(d(ω)) for ω∈rΩp(U;F). (10) For every continuous bilinear mapping F′×F′′→F of Banach spaces, one has
d(ω′∧ω′′)=(dω′)∧ω′′+(−1)p′ω′∧dω′′
for ω′∈rΩp′(X;F′) and ω′′∈rΩp′′(X;F′′).
8.3.6. Let (u1,…,un) be a system of coordinates of X in U. Let
ω=i1<⋯<ip∑fi1,…,ip⋅dui1∧⋯∧duip
1 The sign ∧ indicates that the symbol above which it is placed is to be omitted (cf. 7.8.4). 2 If ω∈1Ωp(X;F) and if dω∈1Ωp+1(X;F), one still has d(dω)=0.
a differential form on U (with fi1,...,ip∈Cr(U;F)). One then has
dω=i1<...<ip∑dfi1,...,ip∧dui1∧⋯∧duip.
8.3.7. Suppose that K is of characteristic zero. Let ω∈rΩp(X;F) be a differential form of degree p≥1 such that dω=0. For every x∈X, there exists an open neighbourhood U of x and a differential form π∈rΩp−1(U;F) such that dπ=ω on U.
8.4. Infinitesimal transformations
Let r∈NK and X be a manifold of class Cr+1.
8.4. Infinitesimal transformations
8.4.1. Let τ be a vector functor for isomorphisms (7.6.6), of class Cr. If E is a Banach space, denote by GL(E) the open subset of L(E;E) consisting of the automorphisms of E. Denote by τE′ the tangent linear mapping at the identity element IdE of the morphism τ:GL(E)→GL(τ(E)); it is a continuous linear mapping of L(E;E) into L(τ(E);τ(E)).
8.4.2. Let E be a Banach space and put F=τ(E). Let U be an open subset of E, ξ a vector field on U of class Cr, ξ~:U→E the corresponding mapping (obtained by identifying T(U) and U×E (8.1.1)), and f:U→F a mapping of class Cr. For x∈U, define an element (θξ.f)(x) of F by the formula
(θξ.f)(x)=dxf(ξ~(x))−τE′(dxξ~)(f(x)).
(Let us remark that on the one hand ξ~(x)∈E and dxf∈L(E;F), whence dxf(ξ~(x))∈F, and on the other hand, dxξ~∈L(E;E), τE′(dxξ~)∈L(F;F) and f(x)∈F, whence τE′(dxξ~)(f(x))∈F. The function θξ.f:x↦(θξ.f)(x) is of class Cr−1 in U.
8.4.3. Let F denote the vector bundle τ(T(X)) (7.6.2 and 7.6.6). Let ξ be a vector field on X and f a section of F, both of class Cr. Let x be a point of X and c=(U,φ,E) a chart of X at x. Let c′=(U,ζc,E) be the corresponding vector chart of T(X) (8.1.1) and τ(c′)=(U,ψc,τ(E)) the corresponding vector chart of F (7.6.2). Put xc=φ(x); let ξc be the mapping of φ(U) into E such that ζc(ξ(u))=(u,ξc(φ(u))) and fc the mapping of φ(U) into τ(E) such that ψc(f(u))=(u,fc(φ(u))) for all u∈U. Let ac be the element (θξc.fc)(xc) of τ(E) defined in 8.4.2. There exists one and only one element a of Fx such that ψc(a)=(x,ac) for every chart c of X at x. We denote it by (θξ.f)(x). It depends only on the germs of ξ and f at x (and even only on their jet of order 1 (12.1.2)).
The mapping θξ.f:x↦(θξ.f)(x) is a section of class Cr−1 of F; it depends K-bilinearly on the pair (ξ,f).
8.4.4. Let us retain the preceding hypotheses and notations. There exists (7.6.4) one and only one morphism of vector bundles τ′ from L(T(X);T(X)) into L(F;F) such that, for every x∈X, the restriction of τ′ to the fibre L(T(X);T(X))x=L(Tx(X);Tx(X)) is equal to τT(x)′ (8.4.1). If g is a function of class Cr in X, with values in K, we have
θgξ.f=gθξ.f−τ′(dg⊗ξ)(f)
(in this formula, we identify dg⊗ξ, which is a section of T′(X)⊗T(X), with a section of L(T(X);T(X)); its image τ′(dg⊗ξ) is a section of L(F;F) and transforms f into a section of F).
8.4.5. Let I be an open set in K containing 0, U an open set in X and φ a mapping of class Cr from I×U into X. For (t,x)∈I×U, denote by εt,x the vector (1,0)∈K×Tx(X)=T(t,x)(I×U). For t∈I, denote by φt the mapping from U into X defined by φt(x)=φ(t,x) (x∈U). We suppose that the following three conditions are satisfied:
a) φ0 is the canonical injection of U into X; b) for every t∈T, the mapping φt is an isomorphism of Cr class manifolds of U onto an open set of X; c) the section (t,x)↦T(t,x)(φ)(εt,x) of the vector bundle φ∗T(X) is of class Cr. Moreover, put: d) ξ(x)=T(0,x)(φ)(ε0,x) for every x∈U.
The mapping ξ:x↦ξ(x) is a vector field of class Cr on U. We call it the initial (or starting) vector field of φ.
Let τ be a vector functor for isomorphisms, of class Cr; put F=τ(T(X)) and let f be a section of class Cr of F over X. For x∈U and t∈I, put φt∗f(x)=τ(Tx(φt)−1)(f(φt(x))); this is an element of Fx. The mapping t↦φt∗f(x) of I into Fx is of class Cr−1 for all x∈U and one has
dtd(φt∗f(x))t=0=(θξ.f)(x)pour tout x∈U.
8.4.6. Given a vector field ξ of class Cr on X and a point x0 of X, there exists an open set I of K, containing 0, an open set U of X containing x0 and a mapping φ:I×U→X of class Cr satisfying conditions a),b) and c) of 8.4.5 and such that the initial field of φ is the restriction ξ∣U of ξ to U.
8.4.7. Example. — Let F be a Banach space and let p be an integer ⩾0. If V is a Banach space, put αp(V)=Altp(V;F); if u is an isomorphism of V onto a Banach space V′, denote by αp(u) the isomorphism of αp(V) onto αp(V′) deduced from u by transport of structure. We thus obtain a vector functor for isomorphisms, denoted by αp (7.8.1 and Errata to No. 7.8). We can apply to it what precedes; we have
αp(T(X))=Altp(T(X);F).
The sections of αp(T(X)) are the differential forms of degree p on X, with values in F; if ω is such a form and if ξ is a vector field on X, both of class Cr, θξ.ω is a differential form of degree p on X, with values in F and of class Cr−1. We have
θξ.ω=d(i(ξ)ω)+i(ξ)dω
and, when r≥2,
θξ⋅dω=d(θξ⋅ω).
When p=0, the functor αp is the constant vector functor defined by F; the bundle α0(T(X)) is identified with the trivial bundle FX, and the sections of this bundle are identified with the functions on X with values in F. If ξ is a vector field of class Cr on X and if f∈Cr(X;F), we have
θξ⋅f=⟨ξ,df⟩=Dξ(f)(8.2.3);
it is an element of Cr−1(X;F).
8.4.8. Let τ and τ1 be two vector functors for isomorphisms. A morphism of vector functors h:τ1→τ defines a morphism of vector bundles, also denoted by h, from τ1(T(X)) into τ(T(X)) (7.6.4), and makes correspond to a section f of τ1(T(X)) a section h(f) of τ(T(X)). If ξ is a vector field on X, and if f and ξ are of class Cr, we have θξ⋅h(f)=h(θξ⋅f).
More generally, let τ,τ1,…,τd be vector functors for isomorphisms and let h:(τ1,…,τd)→τ be a d-linear morphism (7.6.4). Let fi∈Sτi(T(X))r(X) (for i=1,…,d), and let ξ be a vector field of class Cr on X. We have
For example, if f is a section of τ(T(X)) and g a function on X, with values in K, both of class Cr, we have
θξ⋅(gf)=(θξ⋅g)f+g(θξ⋅f).
If ω1 and ω2 are two differential forms of class Cr, with values in Banach spaces F1 and F2 respectively, we have
θξ⋅(ω1∧ω2)=(θξ⋅ω1)∧ω2+ω1∧(θξ⋅ω2),
the exterior product being taken with respect to a continuous bilinear mapping of F1×F2 into a Banach space F.
8.4.9. Let τ be a vector functor for isomorphisms, contravariant (8.2.7), of class Cr, φ:X→Y a morphism of manifolds of class Cr+1, ξ (resp. η) a vector field of class Cr on X (resp. Y) such that ξ is φ-related to η (8.2.6). For every section f of τ(T(Y)), of class Cr, we have
φ∗(θη⋅f)=θξ⋅φ∗f.
8.5. The bracket
Let r∈Nk and let X be a manifold of class Cr+1.
8.5.1. Apply the definitions of No. 8.4 taking for τ the identity functor: τ(V)=V for every Banach space V and τ(u)=u for every isomorphism u of Banach spaces. We then have τ(T(X))=T(X). If ξ and η are two vector fields of class Cr on X, we put
(1)
[ξ,η]=θξ⋅η;
this is a vector field of class Cr−1 on X, called the bracket of ξ and η.8.5.2.Themapping(\xi, \eta) \mapsto [\xi, \eta]isK−bilinearandalternating.IffandgarefunctionsofclassC^ronX,withvaluesinK,onehas:(2)[fξ,gη]=fg[ξ,η]+(fDξg)η−(gDηf)ξ.8.5.3.LetFbeaBanachspace.Onehas(3)D[ξ,η]=Dξ∘Dη−Dη∘Dξinthespaceofmappingsfrom\mathscr{C}^{r+1}(X; F)into\mathscr{C}^{r-1}(X; F).Ifr \geqslant 2andif\zetaisathirdvectorfieldofclassC^ronX,onehas(4)$
[[\xi, \eta], \zeta] = [\xi, [\eta, \zeta]] - [\eta, [\xi, \zeta]] \tag{1}
$orequivalently[ξ,[η,ζ]]+[η,[ζ,ξ]]+[ζ,[ξ,η]]=0.Moregenerally,ifr \geqslant 2andif\tauisavectorfunctorforisomorphisms,onehas(5)θ[ξ,η]=θξ∘θη−θη∘θξinthespaceofmappingsfrom\mathscr{S}_{F}^{r}(X)into\mathscr{S}_{F}^{r-2}(X)(withF = \tau(T(X))).Ifr \geqslant \infty,thevectorfieldsofclassC^ronXformaK−Liealgebraforthebracket(LIE,I,§1,No.2).8.5.4.LetEbeaBanachspace,UanopensubsetofE,\xiand\etatwovectorfieldsofclassC^ronU,identifiedwithelementsof\mathscr{C}^r(U; E).Theirbracketisgivenbytheformula:(6)[ξ,η]=Dξη−Dηξ.8.5.5.Let(u^1, \ldots, u^n)beacoordinatesystemofXinanopensubsetU.Let\xi = \sum a^i \frac{\partial}{\partial u^i}and\eta = \sum b^i \frac{\partial}{\partial u^i}betwovectorfieldsofclassC^ronU.Put[ξ,η]=∑ci∂ui∂.Onehas(7)ci=∑1≤j≤n(aj∂uj∂bi−bj∂uj∂ai).8.5.6.Let\varphi : X \to YbeamorphismofvarietiesofclassC^{r+1},\xi_1and\xi_2twovectorfields1Theformula(4)isstillexactifoneassumesonlythatthevectorfields\xi, \eta, \zeta, [\xi, \eta], [\eta, \zeta], [\zeta, \xi]areallofclassC^1.ofvectorfieldsonX,\eta_1and\eta_2twovectorfieldsonY,allofclassC^r.If\xi_1and\xi_2are\varphi−relatedto\eta_1and\eta_2respectively,[ \xi_1, \xi_2 ]is\varphi−relatedto[ \eta_1, \eta_2 ].8.5.7.Let\omegabeadifferentialformofdegreeponX,withvaluesinaBanachspaceFandofclassC^r,andlet\xi_0, \ldots, \xi_pbevectorfieldsofclassC^ronX.Wehave(8)(θξ0⋅ω)(ξ1,…,ξp)=Dξ0ω(ξ1,…,ξp)−∑i=1pω(ξ1,…,[ξ0,ξi],…,ξp)and(9)dω(ξ0,…,ξp)=∑i=0p(−1)iDξiω(ξ0,…,ξ^i,…,ξp)+∑i<j(−1)i+jω([ξi,ξj],ξ0,…,ξ^i,…,ξ^j,…,ξp).If\xiand\etaarevectorfieldsofclassC^r,wehave(10)θξ∘i(η)−i(η)∘θξ=i([ξ,η])inthespaceofmappingsfromr\Omega^p(X; F)intor-1\Omega^{p-1}(X; F).Forp = 1,formula(9)reads:(11)dω(ξ,η)=Dξ⟨η,ω⟩−Dη⟨ξ,ω⟩−⟨[ξ,η],ω⟩.$
### 8.6. Liftings
Let $r \in \mathbf{N}_kandletg : X \to YbeamorphismofclassC^{r+1}manifolds.8.6.1.AliftingofgintoT(Y)isamapping\psi : X \to T(Y)suchthatthediagramX↗ψgT(Y)↓πYiscommutative,\pidenotingthecanonicalprojectionofT(Y)ontoY.Thedataof\psiisequivalenttothatofasectionofthevectorbundleg^*T(Y)inverseimagebygofT(Y).Thenotionofliftinggeneralizesthatofvectorfield(towhichitreduceswheng = \mathrm{Id}_X);animportantpartofthedefinitionsandresultsconcerning\theta_{\xi}andi(\xi)extendstoliftings;weshallrestrictourselvestobrieflyindicatingafewofthem.8.6.2.Let\psi : X \to T(Y)bealiftingofg : X \to Y,andlet\omegabeadifferentialformofdegreeponY,withvaluesinavectorbundleFwithbaseY.Whenp \geqslant 1,wedenotebyi(\psi, \omega)ori(\psi)\omegaori_\psi(\omega)thedifferentialformonX,ofdegreep - 1,withvaluesing^* F,suchthat(1)i(ψ,ω)x(v1,…,vp−1)=ωg(x)(ψ(x),Tx(g).v1,…,Tx(g).vp−1)foreveryx \in Xandeveryfamilyv_1, \ldots, v_{p-1}ofelementsofT_x(X).Whenp = 0,weseti(\psi, \omega) = 0.Wesaythati(\psi, \omega)istheinteriorproductof\psiand\omega;if\psiand\omegaareofclassC^r,thesameistrueofi(\psi, \omega).LetF' \times_Y F'' \to FbeapairingofvectorbundlesoverY.For\omega' \in {}^r\Omega^{p'}(Y; F')and\omega'' \in {}^r\Omega^{p''}(Y; F''),wehave(2)i(ψ)(ω′∧ω′′)=(i(ψ)ω′)∧g∗ω′′+(−1)p′g∗ω′∧i(ψ)ω′′.8.6.3.∗Examples∗a)Avectorfield\xionXdefinesaliftingT(g) \circ \xiofgintoT(Y),andwehave:(3)i(T(g)∘ξ)=i(ξ)∘g∗.b)Avectorfield\etaonYdefinesalifting\eta \circ gofgintoT(Y),andwehave:(4)i(η∘g)=g∗∘i(η).c)Moregenerally,leth : Y \to ZbeamorphismofvarietiesofclassC^{r+1}.If\psiisaliftingofgintoT(Y),thenT(h) \circ \psiisaliftingofh \circ gintoT(Z)andwehave:(5)i(T(h)∘ψ)=i(ψ)∘h∗.If\varphiisaliftingofhintoT(Z),then\varphi \circ gisaliftingofh \circ gintoT(Z)andwehave:(6)i(φ∘g)=g∗∘i(φ).8.6.4(∗Infinitesimaltransformations∗).Let\psibealiftingofclassC^rofgandlet\omegabeadifferentialformofdegreeponY,withvaluesinaBanachspaceF,andofclassC^r.ThedifferentialformonX(7)d(i(ψ)ω)+i(ψ)dωisofdegreepandofclassC^{r-1};wedenoteitby\theta_{\psi}.\omega.IffisafunctionofclassC^ronX,wehave(8)θfψ.ω=df∧i(ψ)ω+fθψ.ω.8.6.5(∗Characterizationof\theta_{\psi}.\omegaasaderivative∗).Retainthehypothesesof8.6.4.Letx \in X.TherethenexistsanopenneighbourhoodUofx,anopenneighbourhoodIof0inK,andamorphismG : I \times U \to Y,ofclassC^r,havingthefollowingtwoproperties:a)Foreveryx \in U,wehaveG(0, x) = g(x).b)Foreveryx \in U,theimagebyT_{(0, x)}(G)ofthetangentvector(1, 0)isequalto\psi(x).Supposethat(U,I,G)satisfiestheseconditions;foreveryt \in I,denotebyG_tthemappingx \mapsto G(t, x)fromXintoY;put\omega_t = G_t^*(\omega).Foreveryx \in X,themappingt \mapsto \omega_t(x)fromIinto\mathrm{Alt}^p(T_x(X), F)isofclassC^r,anditsderivativeattheoriginisgivenbytheformuladtd(ωt(x))t=0=(θψ⋅ω)(x).$
### 8.7. Weakening of structure
In this no., we suppose $K = \mathbf{R}.8.7.1.LetXbeamanifoldofclassC^s,andletFbeavectorbundleoverX,ofclassC^k,with0 \leq k \leq s.If0 \leq k' \leq k,thereexistsonFoneandonlyonestructureF_{k'}ofvectorbundlewithbaseXandofclassC^{k'}suchthateveryvectorbundlechartofFisavectorbundlechartofF_{k'}.WesaythatF_{k'}isobtainedfromFbyweakeningthebundlestructureofF.8.7.2.Letr \in \mathbf{N}_\mathbf{R}withr \leq s,andletX_rbethemanifoldofclassC^runderlyingX(5.13.1).ThetangentbundleT(X_r)isofclassC^{r-1};itisthebundleobtainedbyweakeningthestructurefromthetangentbundleT(X),whichisofclassC^{s-1}.LetFbeavectorbundleoverX,ofclassC^k,for0 \leq k \leq r - 1.ThedifferentialformsofclassC^konXwithvaluesinFarecanonicallyidentifiedwiththecorrespondingdifferentialformsonX_r$. The various operations on these forms described in the remainder of this paragraph are compatible with this identification.
In what follows in this fascicle, we shall often leave it to the reader to make explicit the other results of this kind.
### 8.8. Almost complex manifolds and complex manifolds
In this no., we suppose $K = \mathbf{R}anddenotebyXa(real)manifoldofclassC^r,withr \in \mathbf{N}_\mathbf{R}.8.8.1(«Complexvectorbundles»).AvectorbundleoverCwithbaseX(7.3.4)isalsocalledacomplexvectorbundlewithbaseX.IfMissuchavectorbundle,wecallacomplexvectorbundlechartofMatriple(U, \varphi, E),whereUisanopensetinX,EacomplexBanachspaceand\varphianisomorphismofcomplexvectorbundlesofM|UontothetrivialbundleE_U = U \times E.Afamily(U_i, \varphi_i, E_i)ofcomplexvectorbundlechartsofMwhosedomainsofdefinitionU_icoverXiscalledacomplexvectorbundleatlasofM;everycomplexvectorbundlepossessesacomplexvectorbundleatlas(7.3.3).WesometimessayC−vectorialinsteadofcomplexvectorial.Analogously,wesometimessay,whenwewishtospeakofavectorbundleover\mathbf{R},ofanordinaryvectorbundlechartofthisbundle,etc.,«realvectorial»or«R−vectorial»insteadof«vectorial».LetMbeacomplexvectorbundlewithbaseX.Thereexistsoneandonlyoneautomorphismjoftheunderlying(real)vectorbundleofM(7.3.3)whoserestrictiontoeachfibreismultiplicationbytheelementiofC;thesquareofjisequalto−1.Conversely,ifMisa(real)vectorbundlewithbaseXandjanautomorphismofMsuchthatj^2 = -1,thereexistsonMauniquestructureofcomplexvectorbundleadmittingMasitsunderlying(real)vectorbundleandforwhichjismultiplicationbyi.8.8.2(«Complexificationofavectorbundle»).Wedefineavectorialfunctorτbysetting\tau(E) = E \otimes \mathbf{C}forevery(real)BanachspaceE,and\tau(u) = u \otimes \mathrm{Id}_\mathbf{C}foreverymorphismu : E \to E'of(real)Banachspaces.IfFisavectorbundlewithbaseXandofclassC^k(0 \leq k \leq r),itstransformbythevectorbundlefunctor\tauisdenotedF \otimes \mathbf{C}.ThereexistsonF \otimes \mathbf{C}auniquestructureofcomplexvectorbundlewithbaseXandofclassC^k,compatiblewithitsstructureasa(real)vectorbundleandinducingoneachfibre(F \otimes \mathbf{C})_x = F_x \otimes \mathbf{C}itscanonicalstructureasacomplexBanachspace(EVT,II,§8,No.1,Example).Endowedwiththisstructure,F \otimes \mathbf{C}iscalledthecomplexificationofF.LetUbeanopensubsetofX;thecanonicalmappingSFk(U)⊗C→SF⊗Ck(U)isanisomorphism,bywhichthesetwospacesareidentified.Ifs \in \mathscr{S}_{F \otimes \mathbf{C}}^k(U),theelements\sigma, \tauof\mathscr{S}_F^k(U)suchthats = \sigma + i \tauarecalledtherealpartandtheimaginarypartofs;thesection\bar{s} = \sigma - i \tauiscalledtheconjugateofs.
In particular, a section of the bundle T(X)⊗C is called a complex vector field on X.
Let H be a complex vector bundle with base X. The canonical isomorphisms AltRp(Tx(X);H)→AltCp(Tx(X)⊗C;H) define an isomorphism of complex vector bundles, called canonical, of AltRp(T(X);H) onto AltCp(T(X)⊗C;H) (7.8.5), by means of which we shall identify these two bundles. Let ω be a section of this bundle, in other words a differential form of degree p on X, with values in H, and let ζ be a complex vector field, with real part ξ and imaginary part η. We then put:
i(ζ,ω)=i(ξ,ω)+i.i(η,ω).
If H is a trivial bundle, we put
θζ.ω=θξ.ω+iθη.ω.
Analogously, we extend the bracket by linearity to complex vector fields. The formulas of the preceding Nos. remain valid.
8.8.3. A structure of class Ck (k≤r−1) on X is called an almost complex structure if it is the datum of a complex vector bundle structure on T(X), of class Ck, whose underlying real structure is the natural structure of T(X). Such a structure is equivalent to the datum of an automorphism j of class Ck of T(X) such that j2=−1.
Let us give ourselves such a structure. Extend j to a C-automorphism of T(X)⊗C. There exists one and only one decomposition of T(X)⊗C into a direct sum of two complex subbundles:
(1)
T(X)⊗C=T′(X)⊕T′′(X)
such that j coincides with multiplication by i on T′(X) and by −i on T′′(X). The projector p′ of T(X)⊗C onto T′(X) is equal to 21(1−ij); the projector p′′=1−p′ of T(X)⊗C onto T′′(X) is equal to 21(1+ij). The composites
π′:T(X)→T(X)⊗Cp′T′(X)
π′′:T(X)→T(X)⊗Cp′′T′′(X)
are R-isomorphisms; the first transforms j into multiplication by i, the second into multiplication by −i.
8.8.4 (“Forms of type (p,q)”). Retain the hypotheses and notations of the preceding no. Let F be a complex vector bundle with base X, let p and q be two integers ⩾0, and let n=p+q. Let ω be a differential form of degree n on an open set U of X, with values in F; identify (8.8.2) ω with a section of Altcn(T(X)⊗C;F). We say that ω is of type(p,q) if, for every x∈U, one has
ωx(v1,…,vn)=0
as soon as p+1 of the vectors vi belong to Tx′(X) or q+1 of the vectors vi belong to Tx′′(X). If one identifies ⋀nTx(X)⊗C with the direct sum of the spaces ⋀m′Tx′(X)⊗⋀m′′Tx′′(X) for m′+m′′=n (A, III, p. 85) and ωx with a C-linear form on ⋀nTx(X), it amounts to saying that ωx is zero on ⋀m′Tx′(X)⊗⋀m′′Tx′′(X) for (m′,m′′)=(p,q).
For every pair (p,q) of integers ⩾0 with p+q=n, there exists one and only one complex vector subbundle Altp,q(T(X);F) of Altn(T(X);F) such that its sections on an open set U of X are the differential forms of type (p,q) on X, with values in F. The vector bundle Altn(T(X);F) is the direct sum of the vector bundles Altp,q(T(X);F) for p⩾0,q⩾0 and p+q=n. Every differential form ω of degree n with values in F decomposes uniquely into
ω=p+q=n∑ωp,q
where ωp,q is a form of type (p,q), called the component of type(p,q) of ω. If ω is of class Ch, with 0≤h≤k, the same is true of its components.
Examples
(a) A differential form of degree n is of type (n,0) if and only if it is C-multilinear.
(b) Let F′,F′′ and F be three complex vector bundles and let F′×XF′′→F be a C-bilinear pairing. If ω′ (resp. ω′′) is a differential form of type (p′,q′) (resp. (p′′,q′′)) with values in F′ (resp. F′′), the differential form ω′∧ω′′ is of type (p′+p′′,q′+q′′).
(c) Suppose that F is the complexification of a (real) vector bundle (8.8.2). If ω is a differential form of type (p,q) with values in F, its conjugate ω (8.8.2) is of type (q,p).
8.8.5. In addition to the preceding hypotheses, suppose that k⩾1. The following three conditions are then equivalent:
(i) For every open set U of X and for every pair (ξ,η) of complex vector fields on U, of class Ck, such that ξ(x) and η(x) belong to Tx′(X) for every x∈U, one has [ξ,η](x)∈Tx′(X) for every x∈U.
(ii) For every open set U of X and every pair (ξ,η) of (real) vector fields on U, of class Ck, one has
[ξ,η]+j[jξ,η]+j[ξ,jη]−[jξ,jη]=0.
(iii) For every open set U of X and for every complex differential form ω of type (1,0) on U, of class Ck, the component of type (0,2) of dω is zero.
Suppose that these conditions are satisfied. Let ω be a differential form of type (p,q) on an open set U of X, of class Ch with 1≤h≤k, with values in a complex Banach space F. We denote by d′ω (resp. d′′ω) the component of type (p+1,q) (resp. (p,q+1)) of dω; the other components of dω are then zero and we have
dω=d′ω+d′′ω.
If k≥2, we have
d′(d′ω)=0,d′′(d′′ω)=0,d′(d′′ω)+d′′(d′ω)=0.
With the notation of 8.8.4, example b), we have
d′(ω′∧ω′′)=d′(ω′)∧ω′′+(−1)p′+q′ω′∧d′(ω′′)
d′′(ω′∧ω′′)=d′′(ω′)∧ω′′+(−1)p′+q′ω′∧d′′(ω′′).
With that of 8.8.4, example c), we have
d′(ω)=d′′(ω).
In formulas (4) and (5) (resp. (6)), the differential forms ω′ and ω′′ (resp. ω) are assumed to be of class C1.
8.8.6 (« Almost complex structure of a complex manifold »). Let Xc be a complex analytic manifold, having X as underlying real analytic manifold (5.14.2.a)). The bundle T(X) is identified with the underlying (real) vector bundle of the complex vector bundle T(Xc), which endows X with an almost complex structure of class Cω, said to be defined by the complex analytic manifold structure given on X. This almost complex structure satisfies conditions (i) to (iii) of 8.8.5; in particular, the operators d′ and d′′ of 8.8.5 are defined. If f:Xc→Yc is a morphism of complex analytic manifolds, f∗ commutes with the operators d′ and d′′.
8.8.7. Retain the hypotheses and notations of 8.8.6. Let Xc be the conjugate of Xc (5.14.2.b)). Embed X by the diagonal mapping x↦(x,x) in the complex manifold Xc×Xc, and let g be the anti-automorphism (x,y)↦(y,x) of Xc×Xc. The pair (Xc×Xc,g) is a complexification of X (5.14.8). In particular, for every x∈X, the tangent space Tx(Xc×Xc) is identified with Tx(X)⊗C; under this identification, Tx(Xc×{x}) corresponds to Tx′(X) and Tx({x}×Xc) corresponds to Tx′′(X).
8.8.8. Suppose r=ω; every almost complex structure of class Cω on X satisfying the equivalent conditions (i) to (iii) of 8.8.5 is the almost complex structure defined by one and only one complex analytic manifold structure on X.¹
8.8.9 (« Holomorphic forms »). Let us take up the hypotheses and notations of 8.8.6, and let F be a complex Banach space. Morphic (in other words complex analytic) differential forms on Xc, with values in F, are also called holomorphic differential forms; the identification of T(Xc) with T(X) identifies them with (real analytic) differential forms on X. For a differential form ω of degree p on X, with values in F and of class Ck with k⩾1, to be holomorphic, it is necessary and sufficient that it be of type (p,0) and that d′′ω=0.
Let ξ be a vector field on X. We say that ξ is holomorphic if it is a complex analytic mapping of Xc into T(Xc)=T(X), in other words a complex analytic vector field on Xc. If this is the case, and if ω is a holomorphic differential form on X with values in F, the form θξ.ω (resp. i(ξ)ω) is the same, whether we consider ξ as a complex analytic vector field on Xc and ω as an element of Ωp(Xc;F), or ξ as a vector field on X and ω as an element of Ωp(X;F). Moreover, let ξ′=π′(ξ) and ξ′′=π′′(ξ) be the components of ξ in T′(X) and T′′(X) respectively (8.8.3); we have
θξ.ω=θξ′.ω,θξ′′.ω=0(7)
i(ξ)ω=i(ξ′)ω,i(ξ′′)ω=0.(8)
8.8.10. Example. — Suppose that Xc is an open set in Cn and denote by z1,…,zn the coordinate functions on Xc. For 1⩽k⩽n, put zk=xk+iyk, where xk and yk are real-valued. Consider the tangent frame (∂/∂zk) of T(Xc) defined by the coordinate system (zk) and the tangent frame (∂/∂xk,∂/∂yk) of T(X) defined by the coordinate system (xk,yk) (8.1.5). We have j(∂/∂xk)=∂/∂yk and the image of ∂/∂zk by the mapping π′:T(X)=T(Xc)→T′(X) (8.8.3) is equal to
21(∂/∂xk−i∂/∂yk);
we shall still denote it in general by ∂/∂zk.² The complex conjugate vector field of this field ∂/∂zk is denoted by ∂/∂zˉk; we have
∂/∂zˉk=21(∂/∂xk+i∂/∂yk).(9)
If f is a function of class C1 on X, with values in a complex Banach space F, the differential forms d′f and d′′f (8.8.5) are given by
d′f=1⩽k⩽n∑∂zk∂fdzk,d′′f=1⩽k⩽n∑∂zˉk∂fdzˉk.(10)
¹ If X is finite-dimensional, this remains true for an almost complex structure of class Ck with k⩾dimX (cf. A. NEWLANDER and L. NIRENBERG, Complex analytic coordinates in almost complex manifolds, Ann. of Math. LXV (1957), p. 391–404). ² Care must be taken that the canonical identification of T(Xc) with T(X) does not transform the vector field ∂/∂zk on Xc into the vector field ∂/∂zk on X, but into the vector field ∂/∂zk+∂/∂zˉk. However, the two vector fields denoted by ∂/∂zk take the same value on holomorphic differential forms (8.8.9), which are moreover the only forms to which a vector field of Xc can be applied.
Let us recall that we have
df=d′f+d′′f
dzk=dxk+idykdzˉk=dxk−idyk.
If A=(i1,…,ip) is a strictly increasing sequence of integers belonging to the interval [1,n], put
dzA=dzi1∧⋯∧dzip
and denote by dzˉA the conjugate of dzA. Every differential form ω of type (p,q) with values in F can be written uniquely
(11)
ω=A,B∑fA,BdzA∧dzˉB
where A (resp. B) runs through the set of strictly increasing sequences of p (resp. q) elements of [1,n], the fA,B being functions with values in F. If ω is of class Ck, with k⩾1, the same is true of the functions fA,B and we have
(12)
d′ω=A,B∑d′fA,B∧dzA∧dzˉB
(13)
d′′ω=A,B∑d′′fA,B∧dzA∧dzˉB.
[^1]: When K=R and r=1, T(X) is a topological vector bundle (cf. Notations and conventions).