Throughout this paragraph, the letter B denotes a manifold of class Cr (r⩾1) and the letter M denotes a set endowed with a mapping π from M into B. We say that B is the base of M and for every b∈B, we denote by Mb and call the fiber of M at b the subset π−1(b) of M.
7.1. Definition of vector bundles
7.1.1. A vector chart of M is a triple t=(U,φ,F), where U is an open set of B, where F is a Banach space and φ a bijection of π−1(U) onto U×F such that π(φ−1(b,h))=b for all b∈B and all h∈F. We say that U is the domain of the vector chart t and that t is a vector chart of M at b∈B if b∈U. For every b∈U, we denote by tb the bijection of F onto Mb defined by tb(h)=φ−1(b,h) for h∈F.
7.1.2. We say that two vector charts t=(U,φ,F) and t′=(U′,φ′,F′) of M are Cr-compatible (or simply compatible) if there exists a mapping λ of class Cr of the manifold U∩U′ into the Banach space L(F;F′) such that:
tb=tb′∘λ(b)for all b∈U∩U′.
7.1.3. We say that a set of vector charts of M is a Cr-vector atlas (or simply vector atlas) of M if it consists of vector charts which are pairwise Cr-compatible and whose domains have B as their union. We say that two vector atlases A and B of M are Cr-equivalent (or equivalent) if A∪B is again a vector atlas of M. This relation is an equivalence relation.
7.1.4. A structure of vector bundle of class Cr (with base B) on M is the data of an equivalence class of vector atlases (Ens., Chap. II, § 6, No. 9). A vector chart belonging to a vector atlas of this class is called a vector chart of the vector bundle M.
Let M be a vector bundle with base B. For every b∈B, there exists on the fiber Mb one and only one Banach space structure such that, for every vector chart t=(U,φ,F) of the vector bundle M at b, the mapping tb is an isomorphism of F onto Mb.
Let c=(U,ψ,E) be a chart of the manifold B and let t=(U,φ,F) be a vector chart of the vector bundle M, with the same domain U. For x∈π−1(U), put:
α(x)=(ψ(π(x)),tπ(x)−1(x)).
Then the triple (π−1(U),α,E×F) is a chart of the set M. There exists on M one and only one manifold structure of class C′ (said to be underlying M) for which all the charts thus obtained are charts of the manifold M. The triple (M,B,π) is then a fibration (6.1.1).
7.1.5. Let F be a Banach space. Put M=B×F, the mapping being the first projection. There exists on M a vector bundle structure (with basis B) and only one for which (B,IdM,F) is a vector chart. One says that B×F endowed with this structure is the trivial vector bundle with basis B and fibre F and one sometimes denotes it by FB. The manifold structure of FB is the product manifold structure and for every b∈B, the mapping h↦(b,h) is an isomorphism of Banach spaces of F onto the fibre of FB at the point b∈B. One often denotes by 0 a trivial vector bundle whose fibre is reduced to 0.
7.1.6. Let M be a vector bundle with basis B. For b∈B, one calls rank of M at b and denotes by rgb(M), the dimension (finite or +∞) of the Banach space Mb. One has dimxM=dimbB+rgbM for b=π(x). The function b↦rgbM is locally constant. One says that M is of finite rank if rgbM<+∞ for every b∈B.
7.2. Morphisms of vector bundles
7.2.1. Let B and B′ be two manifolds and let f be a morphism from B into B′. Let M be a vector bundle with basis B and M′ a vector bundle with basis B′. One says that a mapping g from M into M′ is an f-morphism of vector bundles if the following condition is satisfied:
For every point b0∈B, there exists a vector chart t=(U,φ,F) of M at b0, a vector chart t′=(U′,φ′,F′) of M′ at f(b0) and a mapping λ of class C′ from U into L(F;F′) such that f(U)⊂U′ and that gb∘tb=tf(b)′∘λ(b) for every b∈U, where gb is the restriction of g to Mb.
Under these hypotheses, g is a morphism of varieties and, for every b∈B, g induces a continuous linear mapping of Mb into Mf(b)′. The vector rank of g at b∈B is called and is denoted by rgb(g) the rank (finite or +∞) of the linear mapping gb.
Conversely, if r⩾∞ or if M is of finite rank, an f-morphism of fibrations of (M,B,π) into (M′,B′,π′) which induces on each fibre Mb a linear mapping of Mb into Mf(b)′ (for every b∈B), is an f-morphism of vector bundles.
7.2.2. Let in addition f′ be a morphism of B′ into a variety B′′. If g is an f-morphism of M into M′, and if g′ is an f′-morphism of M′ into a vector bundle M′′ with base B′′, the mapping g′∘g is an (f′∘f)-morphism of M into M′′. We have (g′∘g)b=gf(b)′∘gb for every b in B.
7.2.3. Let M and M' be two vector bundles with the same base B. A B-morphism, or simply a morphism of M into M', is called every Id_B-morphism. The composite of two morphisms is a morphism.
Let g be a morphism of vector bundles of M into M'; if g is bijective, it is an isomorphism of the variety M onto the variety M', the reciprocal mapping g−1 is a morphism of vector bundles of M' into M and we have (g−1)b=gb−1 for every b in B. The mapping g is then an isomorphism of vector bundles.
7.2.4. Let f be a morphism of a variety B' into B and let M be a vector bundle with base B. Put M′=B′×BM and denote by π′ (resp. g) the restriction to M' of the projection of B′×M onto B' (resp. M). There exists on M' one and only one structure of vector bundle with base B' (relative to π′) for which g is a f-morphism. We say that M' is the vector bundle with base B' inverse image of M by f and denote it by f∗M; the f-morphism g is called the canonical f-morphism of f∗M into M.
The structure of variety of f∗M is that of the fiber product of the varieties B' and M over B (5.11.2); for every b∈B′, the mapping x↦(b,x) is an isomorphism of Banach spaces of Mf(b) onto (f∗M)b.
The formation of inverse images of vector bundles is transitive.
Let N' be a vector bundle with base B' and let h be a f-morphism of N' into M. There exists one and only one B'-morphism h~ of N' into f∗M such that h=g∘h~.
Let N be a vector bundle with base B and v a B-morphism of N into M. There exists one and only one B'-morphism, denoted by f∗v, of f∗N into f∗M such that the diagram
f∗N↓Nf∗vvf∗M↓M
is commutative.
7.2.5. Let B' be a subvariety of B and let i be the canonical injection of B' into B. If M is a vector bundle with base B, the inverse image i∗M is called the vector bundle induced on B' by M and is denoted by M∣B′. If t=(U,φ,F) is a vector chart of M, then (U∩B′,φ∣π−1(U∩B′),F) is a vector chart of M∣B′. The canonical f-morphism of M∣B′ into M is an isomorphism of varieties of M∣B′ onto the subvariety π−1(B′) of M.
7.2.6. Let f be a morphism of a variety B' into B and let M (resp. M') be a vector bundle with base B (resp. B'). A f-comorphism of M into M' is called a B'-morphism of f∗M into M'. When B = B' and f=IdB, the data of a f-comorphism of M into M′ is equivalent to the data of a B-morphism of M into M′.
Let g be a f-comorphism of M into M′. For every b∈B′, the mapping gb:x↦g(b,x) of Mf(b) into Mb′ is a continuous linear mapping.
Suppose moreover that f′ is a morphism of a variety B′′ into B′ and that h is a f′-comorphism of M′ into a vector bundle M′′ with base B′′. The mapping h∘f∗g of f′∗(f∗M)=(f∘f′)∗M into M′′ is then a (f∘f′)-comorphism of M into M′′, denoted by h∘g. For b∈B′′, one has
(h∘g)b=hb∘gf′(b).
7.3. Multilinear morphisms
7.3.1. Let M1,…,Md and N be vector bundles with base B, and let u be a mapping of the set M1×B⋯×BMd into N. We say that u is a multilinear morphism (or d-linear) if the following condition is satisfied:
For every b0∈B, there exist an open neighbourhood U of b0 in B, vector bundle charts tj=(U,φj,Fj) of Mj (for 1≤j≤d) and t=(U,φ,F) of N, and a mapping λ of class Cr from U into the Banach space L(F1,…,Fd;F) of continuous d-linear mappings from F1×⋯×Fd into F, such that:
(tb∘λ(b))(x1,…,xd)=u(tb1(x1),…,tbd(xd))
for every b∈U and all xj∈Fj.
Every multilinear morphism u is a morphism of manifolds from the fibre product M1×B⋯×BMd into N and induces for every b∈B a continuous d-linear mapping ub from (M1)b×⋯×(Md)b into Nb.
If f is a morphism from B into a manifold B′, a multilinear morphism from M1×B⋯×BMd into a vector bundle M′ with base B′ is by definition the composite of a multilinear morphism from M1×B⋯×BMd into f∗M′ with the canonical f-morphism of f∗M′ into M′.
A bilinear morphism is also called a pairing. For d=1, a linear morphism is a morphism in the sense of 7.2.1. For d=0, a 0-linear morphism is identified with a section of N (7.4).
7.3.2. A bundle of algebras of base B is called a vector bundle A of base B, endowed with a pairing of A×BA into A. Each fibre Ab is then endowed with a structure of K-algebra. If, for every b∈B, the algebra Ab has a unit element, denoted by eb, the mapping b↦eb is a section of A (cf. 7.4). A bundle of algebras A is said to be locally trivial if, for every point b0 of B, there exists a vector chart t=(U,φ,E) of A at the point b0 and a structure of K-algebra on E such that tb is an isomorphism of algebras of E onto Ab for every b∈U.
7.3.3. Let A be a bundle of associative algebras with unit element of base B. A bundle of A-modules of base B is called a vector bundle M of base B endowed with a pairing m:A×BM→M such that the mapping mb:Ab×Mb→Mb defines, for every b∈B, a structure of Ab-module on the fibre Mb.
Let M and M′ be two bundles of A-modules. An A-homomorphism of M into M′ is called any morphism g:M→M′ of vector bundles of base B which induces for every b in B an Ab-linear mapping of Mb into Mb′.
Suppose A is locally trivial. A fibré M in A-modules is said to be locally trivial if, for every point b0 of B, there exists a vector chart t=(U,φ,E) of A at b0 as in 7.3.2, a vector chart t′=(U,φ′,L) of M at b0, and a module structure of E on L such that, for every b∈U, tb′ is a tb-isomorphism of the Ab-module Mb onto the E-module L.
7.3.4. Let A be a Banach algebra over K (for example a field endowed with a finite-dimensional K-algebra structure). The trivial fibré AB is then a fibré in algebras, locally trivial. A locally trivial fibré in AB-modules M is also called a vector fibré over A (with base B). The fibres Mb are then topological A-modules. An f-morphism u of M into another vector fibré over A is said to be A-linear if the mappings ub are A-linear for every b∈B.
7.4. Sections
7.4.1. Let M be a vector fibré with base B. For every open U of B, we denote by SMr(U) the set of sections of class Cr of M over U, that is to say, morphisms s of class Cr from U into M such that s(b)∈Mb for every b∈U. This set is endowed with a module structure over the ring Cr(U) of morphic functions by the rules:
(1)
(s+s′)(b)=s(b)+s′(b)
(2)
(φ⋅s)(b)=φ(b)⋅s(b)
for s,s′ in SMr(U) and φ in Cr(U). When the open set U varies, one obtains a sheaf SMr of mappings from B into M (cf. No. 5.4.1), called the sheaf of sections of M.
7.4.2. Let M1,…,Md and N be vector bundles with base M and let u be a multilinear morphism from M1×B…×BMd into N. For 1≤j≤d, let us give ourselves a section sj of Mj over an open set U of B; one defines a section u(s1,…,sd) of N over U by the formula:
u(s1,…,sd)(b)=ub(s1(b),…,sd(b))for b∈U.
The mapping (s1,…,sd)↦u(s1,…,sd) is Cr(U)-multilinear. It is sometimes denoted S(u).
7.4.3. Let f be a morphism of a manifold B′ into B and let M be a vector bundle with base B. For every open set U of B and every s∈SMr(U), the mapping x↦(x,s(f(x))) is a section of class Cr of f∗M over the open set f−1(U), denoted f∗s and called the inverse image of s by f. The mapping s↦f∗s from SMr(U) into Sf∗M(f−1(U)) is semilinear with respect to the homomorphism g↦g∘(f∣f−1(U)) of Cr(U) into Cr(f−1(U)).
If moreover N is a vector bundle with base B′ and g an f-comorphism of M into N, one sometimes denotes by S(g) the mapping s↦g∘f∗s from SMr(U) into SNr(f−1(U)).
7.4.4. Let M be a vector bundle with base B, of finite rank. A frame of M over an open subset U of B is called a finite sequence (s1,…,sn) of sections of M over U such that (s1(b),…,sn(b)) is a basis of the vector space Mb for every b∈B. The sequence (s1,…,sn) is then a basis of the Cr(U)-module SMr(U). If f is a morphism of a manifold B′ into B, the sections f∗sj form a frame of f∗M over f−1(U).
7.4.5. Let L be a field, endowed with a structure of K-algebra of finite dimension and let (M,B,π) be a fibration. Suppose that on each fibre Mb a structure of vector space over L, of finite dimension, is given. There exists then at most one structure of vector bundle over L with base B on M, compatible with the mapping π, the manifold structure of M and the L-vector space structures on the fibres (7.3.4). For such a structure to exist, it is necessary and sufficient that the following condition be satisfied:
(FV) For every b0∈B, there exists an open neighbourhood U of b0 in B and an isomorphism of manifolds φ from π−1(U) onto the product U×F of U by a vector space F over L of finite dimension, such that for every b∈U, the bijection φb of Mb onto F induced by φ is an isomorphism of vector spaces over the field L.
The triples (U,φ,F) satisfying (FV) are then vector bundle charts of the vector bundle M.
The condition (FV) is equivalent to: (FV') For every b0∈B, there exists an integer n and n sections s1,…,sn of M over an open neighbourhood U of b0 such that the mapping
(b,a1,…,an)↦a1s1(b)+⋯+ansn(b)
is an isomorphism of the manifold U×Ln onto the manifold π−1(U).
7.4.6. Let M1,…,Md and N be vector bundles with base B, the Mj being of finite rank. Suppose that for every open set U of B a mapping φU from SM1r(U)×⋯×SMdr(U) into SNr(U) is given, Cr(U)-multilinear, such that for V⊂U one has:
φU(s1,…,sd)∣V=φV(s1∣V,…,sd∣V).
There then exists one and only one multilinear morphism u from M1×B…×BMd into N such that φU(s1,…,sd)=u(s1,…,sd) whatever the sections sj of M on the open set U of B.
7.4.7. Let f be a morphism of a variety B′ into B and let M (resp. M′) be a vector bundle with base B (resp. B′) and of finite rank. Suppose that for every open set U of B a mapping φU from SMr(U) into SM′r′(f−1(U)),Cr(U)-semilinear, is given, such that
φU(s)∣f−1(V)=φV(s∣V)
for every open set V⊂U. There then exists one and only one f-comorphism g from M into M′ such that φU(s)=S(g)(s) for every s∈SMr(U).
*7.4.8. Let F be a sheaf of modules over the sheaf of rings CBr. One says that F is locally free if for every b∈B, there exists an open neighbourhood U of b and an integer n such that F∣U is isomorphic (as a sheaf of CUr-modules) to the sheaf (CUr)n.
If M is a vector bundle of base B of finite rank, the sheaf SMr is locally free. Conversely, for every locally free sheaf F on B, there exists a vector bundle M and an isomorphism of sheaves from SMr onto F. If M′ is another vector bundle of base B of finite rank, the mapping g↦S(g) is a bijection from the set of B-morphisms of M into M′ onto the set of morphisms of sheaves of Cr-modules from SMr into SM′r′*
In this number, a vector bundle means a vector bundle of base B and a morphism of vector bundles means an IdB-morphism.
7.5.1. Let M be a vector bundle. A subset M′ of M is called a vector subbundle of M if, for every point b∈B, there exists a vector chart t=(U,φ,E) of M at b and a closed vector subspace F of E admitting a topological supplement, such that
φ(π−1(U)∩M′)=U×F.
Under these conditions, there exists on M′ one and only one structure of vector bundle for which the canonical injection of M′ into M is a morphism. For each b∈B, the fibre Mb′ of M′ is the closed vector subspace M′∩Mb of Mb; M′ is a closed submanifold of M and the structure of underlying manifold of the structure of vector bundle of M′ coincides with that induced by the manifold structure of M.
7.5.2. Let M′ be a vector subbundle of M. Let R{x,y} denote the following relation between points x,y of M: « there exists an element b of B such that x∈Mb,y∈Mb and x−y∈Mb′ ». Then R is a regular equivalence relation on M (cf. n° 5.9.7). On the set M/R, there exists one and only one structure of vector bundle such that the canonical mapping of M onto M/R is a morphism. We denote it by M/M′ and call the vector bundle thus defined the quotient of M by M′; for each point b of B, the bundle (M/M′)b is the quotient topological vector space Mb/Mb′ and the manifold structure on M/M′ is the quotient of that of M.
7.5.3. Retain the hypotheses of 7.5.2. For every point b0 of B, one can find an open neighbourhood U of b0, a Banach space F, direct sum of two closed subspaces F′ and F′′, and an isomorphism of vector bundles ι from FU onto M∣U with the following properties: (i) The restriction ι′ of ι to U×F′ is an isomorphism from FU′ onto M′∣U. (ii) If ρ is the canonical mapping of M onto M′′=M/M′, and if ι′′ is the restriction of ι to U×F′′, the map ρ∘ι′′ is an isomorphism from FU′′ onto M′′∣U.
7.5.4. If a morphism of vector bundles g:L→M has its image contained in M′, it is a morphism of vector bundles from L into M′. Consider now a morphism of vector bundles h:M→N and suppose that, for every b in B, the restriction of hb to Mb′ is zero. If ρ is the canonical morphism from M onto M/M′, there exists one and only one morphism hˉ from M/M′ into N such that h=hˉ∘ρ.
7.5.5. Let P and Q be two vector bundles, and let g be a morphism from P into Q; for every point b of B, denote by Nb and Ib respectively the kernel and the image of the linear mapping gb:Pb→Qb. Put N=⋃b∈BNb and I=⋃b∈BIb. The morphism g is said to be locally direct if N is a vector subbundle of P and I a vector subbundle of Q. The morphism g then defines by passing to the quotient an isomorphism of P/N onto I. It is said that N is the kernel of g and it is denoted by Kerg. Analogously, the subbundle I is called the image of g and is denoted by Img. If r⩾∞, the morphism g is locally direct if and only if g is a subimmersion. If P is of finite rank, the morphism g is locally direct if and only if the vector rank of g is locally constant or equivalently if and only if g is a subimmersion.
7.5.6. Let MfM′gM′′ be two morphisms of vector bundles. We say that the sequence (f,g) is locally direct exact if the two morphisms f and g are locally direct and if Imf=Kerg. If g∘f=0, the set D of points b∈B such that the sequence MbfbMb′gbMb′′ is direct exact (that is to say, such that Kerfb and Imgb admit topological complements and that Imfb=Kergb) is open and the sequence M∣DfM′∣DgM′′∣D is locally direct exact.
One defines analogously the locally direct exact sequences of arbitrary length. By abuse of language, one sometimes says direct exact sequence instead of locally direct exact sequence.
7.5.7. Let 0→MfM′gM′′→0 be a sequence of morphisms of vector bundles. For this sequence to be locally direct exact, it is necessary and sufficient that f be an isomorphism of M onto a vector subbundle f(M) of M′ and that g define by passing to the quotient an isomorphism of the quotient vector bundle M′/f(M) onto M′′.
7.6. Vector functors
In this no. and in the following three no.s 7.7 to 7.9, the letter I denotes a finite set, union of two disjoint subsets I+ and I−. We denote by V=(Vi)i∈I (and analogously by V′,V′′,…) a family of Banach spaces indexed by I. We denote by Hom(V,V′) the Banach space ∏i∈I+L(Vi;Vi′)×∏i∈I−L(Vi′;Vi) and by f=(fi) an element of Hom(V,V′). We denote by IdV the element (IdVi)i∈I of Hom(V,V). For f∈Hom(V,V′) and f′∈Hom(V′,V′′), we denote by f′∘f the element of Hom(V,V′′) whose components are given by:
(f′∘f)i=fi′∘fiif i∈I+
(f′∘f)i=fi∘fi′if i∈I−
7.6.1. A vector functor (resp. vector functor of finite dimension) of type I and of class Cr is called the data, for every family V=(Vi)i∈I of Banach spaces (resp. of finite-dimensional vector spaces over K), of a Banach space τ(V) and, for every f∈Hom(V,V′), of an element τ(f)∈L(τ(V);τ(V′)), these data being subject to the following two conditions: (a) One has τ(IdV)=Idτ(V) and τ(f′∘f)=τ(f′)∘τ(f). (b) The mapping τ:Hom(V,V′)→L(τ(V);τ(V′)) is of class Cr.
7.6.2. Let M=(Mi)i∈I be a family of vector bundles with base B. For b∈B, put Mb=(Mbi)i∈I. Let τ be a vector functor and let τ(M) be the sum set of the τ(Mb) for b∈B; there exists on τ(M) one and only one structure of vector bundle (with base B relative to the mapping π of τ(M) into B such that, for every b∈B, one has τ(Mb)={b}) possessing the following property:
Let U be an open set of B and, for every i, let ti=(U,φi,Fi) be a vector chart of Mi, with domain U; put F=(Fi)i∈I and let ψb be the element of Hom(Mb,F) defined by (ψb)i=(tbi)−1 for i∈I+ and (ψb)i=tbi for i∈I−; for x∈π−1(U), put ψ(x)=(π(x),τ(ψπ(x))(x)). Then the triplet (U,ψ,τ(F)) is a vector chart of the vector bundle τ(M).
Endowed with this structure, τ(M) is called the vector bundle deduced from the family M by the vector functor τ.
7.6.3. Let f be a morphism from B into a manifold B′. Let M=(Mi) (resp. M′=(M′i)) be a family indexed by I of vector bundles with base B (resp. B′). For every i∈I+, let gi be an f-morphism from Mi into M′i and for every i∈I−, let gi be an f-comorphism from M′i into Mi. Put g=(gi)i∈I and for b∈B, put gb=((gi)b)i∈I (cf. 7.2.1 and 7.2.6). There exists one and only one f-morphism, denoted τ(g), from τ(M) into τ(M′) such that τ(g)b=τ(gb) for every b∈B.
If in particular Mi=f∗M′i, the gi being the canonical morphisms or comorphisms, then the B-morphism from τ(M) into f∗τ(M′) defined by τ(g) (7.2.4) is an isomorphism: this fact is expressed by saying that τcommutes with inverse images.
In particular let B′ be a submanifold of B and put M∣B′=(Mi∣B′)i∈I. The vector bundles τ(M)∣B′ and τ(M∣B′) are then canonically B′-isomorphic.
7.6.4. Let τ,τ1,…,τd be vector functors (of type I and of class C′). A d-linear morphismθ from (τ1,…,τd) into τ is the data, for every family V of Banach spaces indexed by I, of a continuous d-linear mapping θV from τ1(V)×⋯×τd(V) into τ(V), this data satisfying the following condition: for every f∈Hom(V′,V′′) one has
τ(f)∘θV=θV′′∘(τ1(f)×⋯×τd(f)).
For d=1, one simply says a morphism from τ1 into τ.
Let M be then a family indexed by I of vector bundles over B. There exists one and only one B-d-linear morphism θM from τ1(M)×B⋯×Bτd(M) into τ(M) such that (θM)b=θMb for every b∈B.
With the notations of 7.6.3., one has
τ(g)∘θM=θM′∘(τ1(g)×⋯×τd(g)).
If d=1, and if θ is an isomorphism (which means that θV is an isomorphism for every family V), then θM is an isomorphism.
7.6.5. The definitions and results of Nos. 7.6.2 to 7.6.4 extend to the case of finite-dimensional vector functors, provided that the vector bundles given are assumed everywhere to have finite rank.
They also extend to the following case: let L be a field endowed with a structure of finite-dimensional K-algebra; one takes for τ a vector functor over L (i.e. satisfying the hypotheses of 7.6.1 in which K has been replaced by L), and one considers only vector bundles over L in the sense of 7.3.4.
7.6.6. One calls a vector functor (resp. finite-dimensional vector functor) for isomorphisms the data, for every Banach space V (resp. every finite-dimensional vector space over K), of a Banach space τ(V) and, for every isomorphism f of V onto a Banach space V’, of an isomorphism τ(f) of τ(V) onto τ(V’), these data being subject to condition (a) of 7.6.1 and to the following condition:
(b’) The mapping f↦τ(f) from the open subset of L(V;V′) constituted by the isomorphisms of V onto V’, into L(τ(V);τ(V′)), is of class Cr.
The definitions and results of the preceding Nos. extend to the case of vector functors for isomorphisms (by taking I+={1} and I−=∅), with the exception of those of the first paragraph of No. 7.6.3.
7.7. Direct sums, bundles of multilinear mappings, dual
7.7.1. Assume that I−=∅. One defines a vector functor σ called the direct sum functor by putting σ(V)=⨁i∈IVi and σ(f)=⨁i∈Ifi. If M=(Mi)i∈I is a family of vector bundles over B, the vector bundle σ(M) is called the direct sum of the Mi and is denoted by ⨁i∈IMi. For every b∈B, the fibre at b of ⨁i∈IMi is the direct sum of the fibres of the Mi at b.
Let U be an open subset of B and let si∈FMi(U) (for i∈I). The mapping b↦∑i∈Isi(b) is then a section, denoted ∑isi, of class Cr of M=⨁i∈IMi and the mapping (si)i∈I↦∑isi is an isomorphism of Cr(U)-modules from ⨁i∈IFMi(U) onto FM(U).
The underlying manifold of ⨁i∈IMi is identified with the fibre product ∏BMi.
We denote by pri the morphism of vector bundles from ⨁i∈IMi into Mi which on each fibre ⨁i∈IMbi is the i-th projection. We define analogously the canonical injection ji of Mi into ⨁i∈IMi.
Let f be a morphism from B into a manifold B′; let H be a second finite set and let N=(Nh)h∈H be a family of vector bundles with basis B′. The mapping u→⨁i∈I(prh∘u∘ji)(h,i)∈H×I is a bijection from the set of f-morphisms of ⨁i∈IMi into ⨁h∈HNh onto the set of matrices (uh,i)(h,i)∈H×I, where uh,i is an f-morphism from Mi into Nh.
If I={1,2}, the sequence
0→M1j1M1⊕M2pr2M2→0
is exact direct.
Conversely, let M be a vector bundle with basis B and let M′ be a vector subbundle of M. Suppose that the manifold B is paracompact and that one of the following two conditions is satisfied : (i) K is different from R or C; (ii) K=R,r=ω and the manifold B admits partitions of unity of class Cr (5.3.6).
There then exists a vector subbundle M′′ of M such that M is identified with the direct sum M′⊕M′′.
7.7.2. Suppose that I+={0} and that I−={1,2,…,d}. We define a vector functor ηd of type I and of class Cr by setting ηd(V)=L(V1,…,Vd;V0) and ηd(f)(u)=f0∘u∘(f1×⋯×fd) for u∈ηd(V). If M=(Mi)i∈I is a family of vector bundles with basis B, the vector bundle ηd(M) is denoted by L(M1,…,Md;M0).
Let u be a multilinear morphism from M1×B⋯×BMd into M0. The mapping u^:b↦ub is then a section of L(M1,…,Md;M0) and the mapping u↦u^ is bijective.
7.7.3. Retain the notations of 7.7.2 and suppose moreover that d=1. The vector bundle L(M1;M0) is then called the bundle of homomorphisms from M1 into M0. Its sections correspond to the B-morphisms from M1 into M0.
If moreover M0 is the trivial bundle KB, the vector bundle L(M1;KB) is called the dual of M=M1 and is denoted by M′: the fibre (M′)b is the space of continuous linear forms on the fibre Mb of M at the point b∈B.
If s (resp. t) is a section of M (resp. M′) over an open set U of B, the mapping b↦(b,⟨s(b),t(b)⟩) is a section, denoted ⟨s,t⟩ of the trivial bundle KB.1
7.8. Bundles of alternating multilinear mappings
In nos. 7.8.1 to 7.8.5, it is assumed that K is of characteristic 0 or that the vector bundles considered are of finite rank. It is not known whether the definition of the vector functor αd given in 7.8.1 can be made without any restriction.
7.8.1. Let I+={0},I−={1} and let d be an integer ⩾1. We define a
1 When M is of finite rank, M∗ is written instead of M′.
vector functor αd by denoting by αd(V) the Banach space of continuous alternating d-linear mappings from V1d into V0 and by setting αd(f)(u)=f0∘u∘f1d for u∈αd(V). The vector bundle αd((M1,M0)) is denoted Altd(M1;M0) and is called the vector bundle of alternating d-linear mappings from M1 into M0.
The canonical injection of Altd(M1;M0) into L(M1,…,M1;M0) is a morphism of vector bundles; Altd(M1;M0) is a vector subbundle of L(M1,…,M1;M0).
One has Alt1(M1;M0)=L(M1;M0). We put Alt0(M1;M0)=M0.
If ω is a section 1 of Altd(M1;M0) and if s1,…,sd are sections of M1, there exists one and only one section of M0, denoted ω(s1,…,sd), such that
ω(s1,…,sd)(b)=ω(b)(s1(b),…,sd(b))for all b∈B.
7.8.2 Let φ be a pairing of N×BN′ into N′′ (cf. No. 7.3.1); for each b, we have a bilinear mapping φb of Nb×Nb′ into Nb′′ which defines (cf. A, III, p. 142) a bilinear mapping of
Altd(M;N)b×Alte(M;N′)b
into Altd+e(M;N′′)b. The collection of these bilinear mappings defines a pairing u of
Altd(M;N)×BAlte(M;N′)
into Altd+e(M;N′′); if ω and ω′ are respectively sections of Altd(M;N) and Alte(M;N′) on an open set U, the section u(ω,ω′) of Altd+e(M;N′′) on U will be denoted by ω∧φω′, and called the exterior product of ω and ω′. We have the formula: (1)
where the si are sections of M on U, and where the summation is extended to the permutations σ of {1,2,…,d+e} such that
σ(1)<⋯<σ(d)andσ(d+1)<⋯<σ(d+e).
7.8.3. Let M be a vector bundle and A an algebra bundle, with base B. Suppose that the fibres Ab of A are associative and commutative algebras, possessing a unit element, denoted by eb. For every open set U of B, we shall denote by Ωd(U) the Cr(U)-module formed by the sections of the bundle Altd(M;A) and by Ω∗(U) the direct sum of the Ωd(U) for d≥0. The multiplications on each fibre define a pairing of A×BA into A, whence (7.8.2) a graded algebra structure on Ω∗(U), which is associative and anticommutative. The subalgebra Ω0(U) is the algebra of
¹ The reader should take care not to confuse this use of the letter ω with that defined p. 10.
sections of A. An element ω of Ω1(U) identifies with a U-morphism of M∣U into A∣U (7.7.3): if s∈SM′(U), we shall denote by ⟨ω,s⟩ the section ω(s) of A (7.4.2). Let sj∈SM′(U) and ωj∈Ω1(U) (for 1≤j≤d); we have:
(2)
ω(s1,…,sd)=det(⟨ωi,sj⟩)for ω=ω1∧…∧ωd.
7.8.4. Let d≥1. There exists a pairing i of M×BAltd(M;A) into Altd−1(M;A) whose restriction to each fibre is given by the right interior product (cf. A, III, p. 156). If s is a section of M on the open set U and if ω∈Ωd(U), we denote by i(s)ω the section i(s,ω) of Altd−1(M;A) on U; we set i(s)ω=0 for ω in Ω0(U).
One thus associates with every section s of M on U an endomorphism of the C′(U)-module Ω∗(U). We have the following formulas:
In the last formula, the ωi are in Ω1(U) and the sign ^ indicates that the symbol which it surmounts is to be omitted.
All the operations described above on the sections are multilinear over the ring C′(U;K).
7.8.5. Let L be a Banach algebra over K. The definitions and results of Nos. 7.7 and 7.8 extend to the case of vector bundles over L: one defines in an analogous manner the bundles of L-multilinear or L-multilinear alternating mappings.
We retain the notations of 7.6. Moreover, we denote by L a commutative field endowed with a structure of finite-dimensional K-algebra and we call vector bundle a vector bundle over L, with base B and locally finite rank.
7.9.1. Suppose that I−=∅. If V and V′ are two families indexed by I of finite-dimensional vector spaces over L, we denote by τ(V) the tensor product of the Vi for i∈I (A, II, p. 71) and if f∈Hom(V,V′), we set τ(f)=⊗fi. We thus define a vector functor over L in finite dimension and if M=(Mi)i∈I is a family of vector bundles, we denote by ⨂i∈IMi and call the tensor product (over L) of the Mi the vector bundle τ(M).
If si is a section of Mi on the open set U of B (for i∈I), the mapping b↦⨂i∈Isi(b) is a section of ⨂i∈IMi, denoted by ⨂i∈Isi. The mapping (si)i∈I↦⨂i∈Isi is multilinear over the ring Cr(U;L).
7.9.2. The canonical isomorphisms defined in Alg., Chap. II provide isomorphisms of vector functors. It follows from 7.6.4 that there are isomorphisms of vector bundles. For example, there are canonical isomorphisms:
(M1⊕M2)⊗M3⟶(M1⊗M3)⊕(M2⊗M3)
M1∗⊗M2⟶L(M1;M2)
etc.
7.9.3. Let M be a vector bundle and let I and J be two finite disjoint sets. The tensor bundle TJI(M) is defined as the tensor product ⨂α∈I∪JMα, where Mα=M if α∈I, and Mα=M∗ if α∈J, M∗ denoting the dual of M (A, III, p. 63). The fibre TJI(M)b of this bundle at a point b is equal to the tensor space TJI(Mb) defined in Alg., loc. cit. When I={1,…,p} and J={p+1,…,p+q}, one writes Tqp(M) in place of TJI(M); one has
Tqp(M)=(⨂pM)⊗(⨂qM∗).
The giving of a total order on I and on J defines a canonical isomorphism of TJI(M) onto Tqp(M).
7.9.4. If I (resp. J) is the union of two disjoint subsets I′ and I′′ (resp. J′ and J′′), TJI(M) is canonically identified with the tensor product TJ′I′(M)⊗TJ′′I′′(M). In particular, if s′ (resp. s′′) is a section of TJ′I′(M) (resp. of TJ′′I′′(M)), the tensor product s′⊗s′′ is identified with a section of TJI(M).
7.9.5. The dual of TJI(M) is identified with TIJ(M).
7.9.6. Let i∈I and j∈J. For every b∈B, the homomorphism of contraction of the indices i and j is defined (cf. Alg., loc. cit.); it is a homomorphism (cji)b:TJI(Mb)→TJ−{j}I−{i}(Mb). The collection of the (cji)b defines a morphism of vector bundle spaces
cji:TJI(M)→TJ−{j}I−{i}(M),
also called contraction of the indices i and j. One defines analogously the contraction of the indices i1,…,ik of I with the indices j1,…,jk of J.
7.9.7. Let d be an integer ⩾0; let V and V′ be two vector spaces of finite dimension over L and f∈HomL(V,V′); put λd(V)=⋀d(V) and λd(f)=⋀d(f). One thus defines a vector functor over L in finite dimension and, if M is a vector bundle, one denotes by ⋀d(M) and calls d-th exterior power of M (over L) the vector bundle λd(M).
The canonical mapping of ⨂dV onto ⋀d(V) defines a morphism of vector functors, whence a canonical morphism of ⨂dM into ⋀d(M). This morphism is surjective.
The canonical isomorphisms of the space of alternating d-multilinear mappings on the space ⋀d(V∗) or on (⋀d(V))∗ provide isomorphisms, called canonical, of the vector bundle Altd(M;L) of alternating d-multilinear mappings on L, onto the vector bundle ⋀d(M∗) or onto (⋀d(M))∗.
7.9.8. Put now λ(V)=⋀(V) and λ(f)=⋀(f); one thus again defines a vector functor on L in finite dimension. The vector bundle λ(M) is denoted by ⋀(M). Its fibre ⋀(M)b at b∈B is the exterior algebra (over L) of the fibre Mb. The vector bundle ⋀(M) is a locally trivial vector bundle in algebras.
The definitions and properties of the interior products given in Alg., Chap. III, 3rd ed., § 10 extend immediately to the sections of the vector bundles ⋀(M) and ⋀(M∗) (cf. also the formulas (1) to (7) of Nos. 7.8.2 to 7.8.4).
7.9.9. Let M be a vector bundle. For every integer n, let Bn be the set (open) of the points b∈B such that the dimension (over L) of Mb is equal to n. For b∈Bn, put Nb=⋀n(Mb) and let N be the sum set of the Nb for b∈B. There exists on N one and only one structure of vector bundle such that the obvious mapping of N∣Bn into ⋀n(M)∣Bn is an isomorphism for every n. Endowed with this structure, the vector bundle N of rank one at each point is denoted by det(M).
7.10. Vector bundles and principal bundles
7.10.1. Let F be a Banach space. A vector bundle M with base B is said to be pure of type F if all the fibres Mb of M (for b∈B) are isomorphic (as Banach spaces) to F.
Let M be a vector bundle of base B pure of type F and let P be the open subvariety of the vector bundle L(FB;M) composed of the pairs (b,u) where b∈B and where u is an isomorphism of Fb=F onto Mb. The group GL(F) of automorphisms of F operates on the right on P by setting (b,u)⋅g=(b,u∘g) for (b,u)∈P and g∈GL(F). Let πP denote the mapping (b,u)↦b from P into B. The quadruplet λ=(P,GL(F),B,πP) (where GL(F) is endowed with its canonical structure of group variety (5.12.2)) is a principal fibration (6.2.1): it is called the fibration of frames of M. The mapping ((b,u),h)↦u(h) from P×F into M endows M with a structure of associated bundle to λ, of fibre type F (6.5.1).
When F=Kn, one can identify an isomorphism u of F onto Mb with the basis of Mb image by u of the canonical basis of Kn. The fibred space of frames of M is then identified with the open subvariety of M×B…×BM formed by the bases (e1,…,en) of the different fibres Mb.
Let U be an open subset of B and let t=(U,φ,E) be a vector chart of M of domain U, with E=F. The mapping b↦(b,tb) is then a section of P, denoted t~, and the mapping t↦t~ is a bijection from the set of vector charts of M of the form (U,φ,F) onto the set of sections of P over U.
7.10.2. Conversely, let λ=(Q,G,B,πQ) be a principal fibration and let φ be a homomorphism of group varieties of G into the group GL(F) of automorphisms of a Banach space F. Let G operate on the left on F by setting g.h=φ(g)(h) (h∈F,g∈G). Let M be a fibred space associated with λ, of fibre type F, and let ρ:Q×F→M be its frame mapping (6.5.1). Let π be the mapping from M into B defined by
π(ρ(q,h))=πQ(q)(q∈Q and h∈F).
Let s be a section of Q over an open subset U of B; the mapping
s~:(b,h)↦(s(b),h)
is then a bijection from U×F onto π−1(U). There exists on the pair (M,π) one and only one structure of vector bundle of base B for which the triples ts=(U,s~−1,F) are vector charts (for every section s of Q). The structure of underlying variety of this structure is that of the fibred space associated with λ.
Let q∈Q; put b=πQ(q) and let u be the isomorphism of F onto Mb defined by u(h)=ρ(q,h) (for h∈F). The mapping f:q↦(b,u) is a B-morphism of principal fibrations, compatible with φ, from (Q,G,B,πQ) into the frame bundle (P,GL(F),B,πP) of the vector bundle M.
7.10.3. Let us take up again the notations of No. 7.6. For every i∈I, let
λi=(Pi,Gi,B,πi)
be a principal fibration with base B and suppose that Gi operates on the left on a Banach space Vi by means of a homomorphism
φi:Gi→GL(Vi).
Let Mi be a fibre space associated with λi of fibre type Vi. Put M=(Mi)i∈I and V=(Vi)i∈I and let λ be the product principal fibration of the λi over B (6.2.5). Put λ^=(P,G,B,πp), with G=∏i∈IGi.
Let now τ be a vector functor. For g=(gi)∈G, let φ(g) be the element of Hom(V,V) defined by:
φ(g)i=φi(gi)si i∈I+
φ(g)i=φi(gi)−1si i∈I−
The group G then operates on τ(V) by means of the morphism g↦τ(φ(g)) from G into GL(τ(V)).
Let on the other hand x=(xi) be a point of P and let b=πp(x). For each i, the mapping θxi defined in No. 6.5.2 is an isomorphism of Vi onto (Mi)b. Let θx be the element of Hom(V,((Mi)b)i∈I) defined by:
(θx)i=θxisi i∈I+
(θx)i=θxi−1si i∈I−
Let ρ be the mapping (x,h)↦(b,τ(θx)(h)) from P×τ(V) into the vector bundle τ(M); the mapping ρ endows τ(M) with a structure of fibre space associated with λ of fibre type τ(V).
These considerations generalize to the case of finite-dimensional vector functors, or of vector functors over a field L endowed with a structure of finite-dimensional K-algebra.
7.11. Change of structure
The structures and operations described in this paragraph are compatible with the changes of structure described in Nos. 5.13 and 5.14.