13.1.1. Let E be a module over a commutative ring A. If n is an integer ≥0, denote by TSn(E) the module of symmetric tensors of rank n of E (A, III, p. 71 and A, IV, § 5, no 2); the direct sum of the TSn(E) is denoted by TS(E). If r is an integer, one of the symbols ∞,ω, put
TS(r)(E)=n≤r⨁TSn(E)andTS(r)+(E)=1≤n≤r⨁TSn(E);
consequently, if r≥0, TSr(E) is the direct sum of TS(0)(E)=A and of TS(r)+(E). One has TS(1)+(E)=E and TS(∞)(E)=TS(ω)(E)=TS(E).
If n is an integer ≥0, and if x∈E, denote by γn(x) the element x⊗⋯⊗x of TSn(E); for n=0, γ0(x)=1.
13.1.2. (« Polynomial mappings »). Suppose that A is integral infinite and that E is a free a-module. Let F be an a-module. A mapping f:E→F is said to be homogeneous polynomial of degree n if it satisfies the following equivalent conditions (cf. A, IV, § 5, no 9):
a) There exists a linear mapping f~:TSn(E)→F such that f(x)=f~(γn(x)) for all x∈E.
b) There exists a multilinear mapping u:En→F such that f(x)=u(x,…,x) for all x∈E.
Suppose that this is the case. The mapping f~ satisfying a) is then unique; if t∈TSn(E), denote by ⟨f,t⟩ the element f~(t). If u:En→F satisfies b), the corresponding linear mapping of ⊗nE into F coincides in the submodule TSn(E). When n! is invertible in A, one may choose u symmetric, and this in a unique way.
13.1.3. Let E be a Banach space over K and F a separated polynomial space over K. Let U be an open neighbourhood of 0 in E, f:U→F a mapping of class Cr (r∈NK), and t an element of TS(r)(E), cf. 13.1.1 (applied to the ring A=K). Let k be an integer ≤r such that t∈TS(k)(E). One can decompose t and f uniquely into
t=t0+⋯+tk,with ti∈TSi(E)
and
f=f0+⋯+fk+h,
where fi is a homogeneous continuous polynomial of degree i (0≤i≤k) and where h has contact of order ≥k with 0 at the point 0. Put then
⟨f,t⟩=i=0∑k⟨fi,ti⟩;
this is an element of F which does not depend on the choice of k.
13.1.4. Let E and E’ be two Banach spaces, U an open neighbourhood of 0 in E and φ:U→E′ a mapping of class Cr such that φ(0)=0. Let t∈TS(r)(E). There exists one and only one element t′∈TS(r)(E′) such that, for every separated polynormed space F, every open neighbourhood U’ of 0 in E’, and every mapping f:U′→F of class Cr, one has
(*)
⟨f,t′⟩=⟨f∘φ,t⟩.
This element t′ is denoted by φ∗(t). The mapping
φ∗:TS(r)(E)→TS(r)(E′)
thus defined is linear.
13.2. Point distributions
In this No., X denotes a manifold of class Cr, and x a point of X.
13.2.1. Let A be the set of pairs (c,t), where c=(U,φ,E) is a chart of X centred at x, and where t∈TS(r)(E). Two elements ((U,φ,E),t) and ((U′,φ′,E′),t′) of A are said to be equivalent if t′=γ∗(t), where γ=φ′∘φ−1. In this way one obtains an equivalence relation on A; an equivalence class for this relation is called a point distribution at x on X.
Let Tx(r)(X) be the set of point distributions at x on X. If c is a chart of X centred at x, the mapping
θc:TS(r)(E)→Tx(r)(X)
which associates with t the class of (c,t), is a bijection. In all that follows, Tx(r)(X) is endowed with the structure of a K-vector space obtained by transporting that of TS(r)(E) by means of θc; this structure does not depend on the choice of c. Analogously, if k≤r, Tx(k)(X) and Tx(k)+(X) denote the images of TS(k)(E) and TS(k)+(E) under θc; they do not depend on the choice of c. An element of Tx(r)(X) is said to be of order ≤k if it belongs to Tx(k)(X). One has
Tx(k)(X)=Tx(0)(X)⊕Tx(k)+(X).
There exists one and only one element εx of Tx(0)(X) such that θc(1)=εx for every chart c of X centred at x. If t is a point distribution at x, the constant term of t is the element λ of K such that t−λεx∈Tx(r)+(X); one says that t is without constant term if its constant term is zero.
13.2.2. Let F be a separated polynormed space, and let f be a function of class Cr with values in F, defined on a neighbourhood of x. Let t be a point-supported distribution at x on X.
Consider a chart c=(U,φ,E) of X centred at x, and put fc=f∘φ−1 and tc=θc−1(t); the element ⟨fc,tc⟩ of F defined in 13.1.3 does not depend on the choice of c; we denote it by ⟨f,t⟩. One has εx(f)=f(x).¹ The constant term of t is ⟨1,t⟩.
If F′ is a separated polynormed space and if u:F→F′ is continuous linear, one has ⟨u∘f,t⟩=u(⟨f,t⟩).
Suppose that F is a Banach space, and that t is of order ≤k, with k finite. Let j∈Jxk(X,F), cf. 12.1.2, and let f∈j. Then ⟨f,t⟩ depends only on the jet j; we denote it by ⟨j,t⟩. The mapping from Tx(k)(X)×Jxk(X,F) into F thus defined is bilinear. When F=K, we have put Jxk(X,F)=Pxk(X), cf. 12.6.2, and we obtain a bilinear form on Tx(k)(X)×Pxk(X).
13.2.3. Let φ:X→Y be a morphism of manifolds of class Cr and let y=φ(x). Let c=(U,ψ,E) be a chart of X centred at x and let c′=(U′,ψ′,E′) be a chart of Y centred at y; let φ~ be the expression of φ in these charts (5.3.2), and let φ~∗ be the corresponding mapping from TS(r)(E) into TS(r)(E′), cf. 13.1.4. There exists one and only one linear mapping, denoted by Tx(r)(φ) or φ∗, from Tx(r)(X) into Ty(r)(Y), making the diagram commutative
If φ′:X→Y is a morphism of class Cr mapping x onto y, we have φ∗=φ∗′ if and only if jxr(φ)=jxr(φ′), cf. 12.1.
If φ is an immersion (resp. a submersion) at x, φ∗ is injective (resp. surjective); the converse is true if dimyY<+∞.
If φ′:Y→Z is a morphism of Cr manifolds, and if t∈Tx(r)(X), we have (φ′∘φ)∗(t)=φ∗′(φ∗(t)).
Let F be a separated polynormed space, and let f be a function of class Cr with values in F, defined in a neighbourhood of y. We then have
(1)
⟨f,φ∗(t)⟩=⟨f∘φ,t⟩for all t∈Tx(r)(X).
For given t, the relations (1) (for variable f and F) characterizeφ∗(t); when dimyY<+∞, we may restrict ourselves to F=K.
13.2.4. Suppose that X is an open submanifold of a Banach spaceE. Denote by φx the mapping y↦y−x of X into E; the chart c=(X,φx,E) is centred
¹ When K=R or C and X is locally compact, the preceding formula leads to identifying the point distribution εx with the Dirac measureεx defined in JNT, III, § 1, No. 3. More generally, every measure with finite support in X is identified with a distribution with finite support in the sense of No. 13.6.
at x. We then identify Tx(r)(X) with TS(r)(E) by means of θc−1; in particular, we have Tx(∞)(X)=TS(E).
Let F be a Banach space, let u:E→F be a continuous linear mapping, and let x∈E,y∈F be such that u(x)=y. Identify as above Tx(∞)(E) with TS(E) and Ty(∞)(F) with TS(F). The mapping
u∗:TS(E)→TS(F)(cf. 13.2.3)
coincides with the mapping TS(u) induced by the canonical extension of u to the tensor algebra.
13.2.5. If k≤r, denote by T(k)(X) (resp. T(k)+(X)) the sum set of the Ta(k)(X) (resp. of the Ta(k)+(X)) for a∈X. A mapping t:X→T(k)(X) such that t(a)∈Ta(k)(X) for all a∈X is called a field of point distributions of order ≤k.
Suppose k finite and X locally of finite dimension. For every a∈X, the bilinear form (j,t)↦⟨j,t⟩ (cf. 13.2.2) defines an isomorphism ia of Ta(k)(X) onto the dual Pak(X)∗ of Pak(X). The ia define a bijection i:T(k)(X)→Pk(X)∗, where Pk(X)∗ denotes the dual of the vector bundle Pk(X); by transport of structure by means of i−1, one endows T(k)(X) with a structure of vector bundle with base X and of class Cr−k. A field of point distributions of order ≤k is said to be of class Cs, with s≤r−k, if it is a section of class Cs of the vector bundle T(k)(X).
Let φ:X→Y be a morphism of manifolds of class Cr and suppose that Y is, like X, locally of finite dimension. Then φ∗:T(k)(X)→T(k)(Y) is a φ-morphism of vector bundles of class Cr−k.
13.2.6. Let k be an integer such that 0≤k≤r. If U is an open set of a Banach space E of finite dimension, the vector bundle T(k)(U) is identified, thanks to 13.2.4, with the trivial bundle of fibre TS(k)(E).
Take in particular E=Kn, with n≥0, and let (e1,…,en) be the canonical basis of E. If α=(α1,…,αn) is an element of Nn, denote by Δα the element γα1(e1)…γαn(en) of TS(E) (the product used being the symmetric product of symmetric tensors, cf. A, IV, § 5, No. 3).(1) The Δα, for ∣α∣≤k, form a basis of TS(k)(E); if a∈Kn, denote by Δaα the corresponding elements of Ta(k)(E). Let F be a separated polynormed space, and f a function of class Cr, with values in F, defined in a neighbourhood of a; if α∈Nn, the element ⟨f,Δaα⟩ coincides with the element (Δαf)(a) defined in 2.5.3, 3.2.1 and 4.2.1.
Suppose X is locally of finite dimension, and let ξ=(ξ1,…,ξn) be a system of coordinates of X in an open set U. If a∈U and if α is a multi-index such that ∣α∣≤k, denote by (Δξα)a the point distribution at a whose image by ξ is the point distribution Δξ(a)α on Kn; the fields of distributions
Δξα:a↦(Δξα)a(∣α∣≤k)
form a frame on U of the vector bundle T(k)(X). If f:U→F is of class Cr, put ⟨f,(Δξα)a⟩=Δξαf(a) and denote by Δξαf the function a↦Δξαf(a); this is a function of class Cr−k in U.
(1) If m=∣α∣ and if Σ is the set of mappings σ of {1,…,m} into {1,…,n} such that Cardσ−1(i)=αi, we have
Δα=γα1(e1)…γαn(en)=σ∈Σ∑eσ(1)⊗⋯⊗eσ(n),cf. A, IV, § 5, n° 4.
13.3. Point distributions and tangent spaces
In this No., X denotes a variety of class Cr.
13.3.1. Let x∈X and let k be an integer ≤r. Let c=(U,φ,E) be a chart of X centred at x; the isomorphism θc:TS(X)(E)→Tx(X) defined in 13.2.1 maps TS(k)(E) onto Tx(k)(X) and TS(k−1)(E) onto Tx(k−1)(X); by restriction and passing to the quotient, it induces an isomorphism
The isomorphism ik is independent of the choice of c; for k=1, it is used to identify Tx(1)(X)=Tx(1)(X)/Tx(0)(X) with the tangent space Tx(X). If t∈Tx(k)(X), we allow ourselves to denote by ik(t) the image by ik of the class of t modulo Tx(k−1)(X).
Put
grkTx(r)(X)=Tx(k)(X)/Tx(k−1)(X)
and
grTx(r)(X)=0≤k≤r⨁grkTx(r)(X).
We say that grTx(r)(X) is the graded associated with the increasing filtration (Tx(k)(X))k≤r of Tx(r)(X). The ik define an isomorphism of graded vector spaces
i:grTx(r)(X)⟶TS(r)(Tx(X)).
If r=∞ or ω, i is an isomorphism of grTx(r)(X) onto TS(Tx(X)). When one wishes to specify x (or X, or both), one writes ix (or iX, or ix,X) in place of i.
13.3.2. In addition to the above hypotheses, suppose that X is locally finite-dimensional. The isomorphisms ik corresponding to the different points of X define an isomorphism of vector bundles
ik:T(k)(X)/T(k−1)(X)⟶TS(k)(T(X))
where TS(k)(T(X)) denotes the vector bundle of class Cr−1 obtained from T(X) by the finite-dimensional vector functor TS(k) (cf. 7.6.5). For k≥1, ik is an isomorphism of class Cr−k; for k=0, it is the identity mapping of the trivial bundle KX.
13.3.3. Example. — The hypotheses being as above, let ξ=(ξ1,…,ξn) be a system of coordinates of X at x. Let (∂i)1≤i≤n be the basis of Tx(X) defined by ξ (cf. 5.5.8) and let (Δxα)∣α∣≤k be that of Tx(k)(X) defined in 13.2.6. We have:
ik(Δxα)=0if ∣α∣<k
ik(Δxα)=γα1(∂1,x)⋯γαn(∂n,x)if ∣α∣=k.
[^1]
13.3.4. Suppose that K is of characteristic zero or that X is locally of finite dimension. Let k be an integer such that 0≤k≤r and let x∈X; let F be a Banach space. By 12.6.8, we have an exact sequence
(i)
0→Pk(Tx(X);F)iPxk(FX)σPxk−1(FX)→0,
with σ=rk,k−1. On the other hand, by 13.3.1, we have an exact sequence
(ii)
0←TSk(Tx(X))iTx(k)(X)sTx(k−1)(X)←0,
where s is the inclusion of Tx(k−1)(X) in Tx(k)(X) and i=ik. These exact sequences are “coupled in F”. More precisely, write them in the form:
(i)
0→AiBσC→0
(ii)
0←A′iB′sC′←0.
If a∈A and a′∈A′, the element ⟨a,a′⟩ of F is defined by 13.1.2; if b∈B and b′∈B′, (resp. if c∈C and c′∈C′), the element ⟨b,b′⟩ (resp. ⟨c,c′⟩) of F is defined by 13.2.2; we then have:
13.3.5. Let Y be a variety of class Cr, let φ:X→Y be a morphism of class Cr and let x∈X,y∈Y be such that y=φ(x). The mapping φ∗:Tx(r)(X)→Ty(r)(Y) defined in 13.2.3 is compatible with the filtrations of these spaces; it defines by passing to the associated graded spaces a linear mapping
gr(φ∗):grTx(r)(X)→grTy(r)(Y).
On the other hand, let TS(r)(Tx(φ)) be the mapping of TS(r)(Tx(X)) into TS(r)(Ty(Y)) induced by the canonical extension TS(Tx(φ)) of Tx(φ). The diagram
13.3.6. Suppose that K is of characteristic zero. Using the isomorphisms φM:Sk(M)→TSk(M) defined in A, IV, § 5, No. 8, we can replace, in all that precedes, the TSk(Tx(X)) by the k-th symmetric powers Sk(Tx(X)) of the tangent space Tx(X).
13.4. Tensor product of point distributions
13.4.1. Let X1 and X2 be two manifolds of class Cr, let x1∈X1,x2∈X2 and t1∈Tx1(k1)(X1),t2∈Tx2(k2)(X2) with k1+k2≤r. Put X=X1×X2,x=(x1,x2) and k = k_1 + k_2. For i = 1, 2, let c_i = (U_i, \varphi_i, E_i) be a chart of X_i centred at x_i and denote by t~i the element of TS(E_i) such that θc(t~i)=ti, cf. 13.2.1. Let σ be the canonical isomorphism of TS(E_1) ⊗ TS(E_2) onto TS(E_1 × E_2), cf. A, IV, § 5, no. 5. The element σ(t~1⊗t~2) is the symmetric product of t~1 and t~2 (identified with elements of TS(E_1 × E_2) by means of the canonical injections TS(E_i) → TS(E_1 × E_2)); it belongs to TS^{(k)}(E_1 × E_2). Put c=c1×c2; this is a chart of X centred at x. The image by θc of σ(t~1⊗t~2) is an element of T_x^{(k)}(X), which does not depend on the choice of the charts ci. It is called the direct product, or the symmetric tensor product (or simply the tensor product) of t1 and t2 and is denoted by t1×t2 or t1⊗t2.
We have εx1⊗εx2=εx. The constant term of t1⊗t2 is the product of the constant terms of t1 and t2.
The isomorphism (y1,y2)↦(y2,y1) of X1×X2 onto X2×X1 transforms t1⊗t2 into t2⊗t1.
13.4.2 (« Associativity and functoriality »). Let Xi (i=1,2,3) be manifolds of class Cr, and let xi∈Xi, ti∈Txi(ki)(Xi) with k1+k2+k3≤r. We have
(t1⊗t2)⊗t3=t1⊗(t2⊗t3);
we denote this point distribution by t1⊗t2⊗t3. Finite tensor products of any number of factors are defined analogously.
Let φ1:X1→Y1 and φ2:X2→Y2 be morphisms of manifolds of class Cr, and let t1∈T(k1)(X1), t2∈T(k2)(X2) with k1+k2≤r. We have
(φ1×φ2)∗(t1⊗t2)=φ1∗(t1)⊗φ2∗(t2).
13.4.3. The notation being that of 13.4.1, let F1,F2 and F be separated polynormed spaces, and let (u1,u2)↦u1.u2 be a continuous bilinear mapping of F1×F2 into F. Let
fi:Xi→Fi(i=1,2)
be a mapping of class Cr, and define f1⊗f2:X→F by
(f1⊗f2)(y1,y2)=f1(y1).f2(y2).
The mapping f1⊗f2 is of class Cr, and we have
⟨f1⊗f2,t1⊗t2⟩=⟨f1,t1⟩.⟨f2,t2⟩.
13.4.4. The notation being that of 13.4.1, let f be a mapping of class Cr of X into a separated polynormed space F. If y1∈X1, denote by fy1 the mapping y2↦f(y1,y2) of X2 into F and put g(y1)=⟨fy1,t2⟩. The function g:X1→F thus defined is of class Cr−k1, and we have
be the linear mapping defined by the bilinear mapping (t1,t2)↦t1⊗t2; this mapping is injective; it is bijective if k1 and k2 are infinite.
If n≤r, denote by Tx(n)(X1,X2) the vector subspace of Tx1(n)(X1)⊗Tx2(n)(X2) generated by the subspaces Tx1(k1)(X1)⊗Tx2(k2)(X2) for k1+k2≤n. There exists one and only one linear mapping
αn:Tx(n)(X1,X2)→Tx(n)(X)
extending the mappings αk1,k2 defined above; it is an isomorphism. In what follows, we identify Tx(n)(X), by means of αn−1, with the vector subspace Tx(n)(X1,X2) of Tx1(n)(X1)⊗Tx2(n)(X2).
13.4.6. Retain the notation of 13.4.5 and 13.3.1. By passing to the quotient, αk1,k2 defines a linear mapping
is commutative (in this diagram, i12 denotes the homomorphism induced by ix1⊗ix2 and σ is the isomorphism defined in A, IV, § 5, No. 5). If r is infinite, this diagram is simply written:
In this number, X denotes a variety of class Cr and x a point of X.
13.5.1. Let k≤r. If Δ is the diagonal mapping y↦(y,y) of X into X×X, then Δ∗ maps Tx(k)(X) into T(x,x)(k)(X×X), which is a vector subspace of
Tx(k)(X)⊗Tx(k)(X), cf. 13.4.5. One thus obtains a linear mapping, again denoted by Δ∗ (or c):
Tx(k)(X)⟶Tx(k)(X)⊗Tx(k)(X);
it is called the coproduct attached to Tx(k)(X). Endowed with this coproduct, Tx(k)(X) is a coassociative and cocommutative cogebra (A, III, p. 144-145); its counit is the linear form which associates to a point distribution its constant term (13.2.1). If k′≤k, the inclusion of Tx(k′)(X) in Tx(k)(X) is a morphism of cogebras. If c=(U,φ,E) is a chart of X centred at x, the isomorphism
θc:T(k)(E)⟶Tx(k)(X)
is an isomorphism of cogebras, T(k)(E) being endowed with the coproduct induced by that of T(E), cf. A, IV, § 5, no 7. If φ:X⟶Y is a morphism of manifolds of class Cr, the mapping of T(k)(X) into Tφ(x)(k)(Y) induced by φ∗ is a morphism of cogebras.
13.5.2. Let F1, F2 and F be separated polynomial spaces, and (u1,u2)↦u1.u2 a continuous bilinear mapping of F1×F2 into F. Let t∈Tx(r)(X) and let
c(t)=j∑uj⊗vj(uj,vj in Tx(r)(X))
be its image by the coproduct. Let fi:X⟶Fi(i=1,2) be mappings of class Cr. We have
⟨f1.f2,t⟩=j∑⟨f1,uj⟩⟨f2,vj⟩,
which we write, by abuse of notation:
⟨f1.f2,t⟩=⟨f1⊗f2,c(t)⟩.
13.5.3. Let t∈Tx(k)(X), where k≤r.
a) For c(t)=t⊗1, it is necessary and sufficient that t be equal to 0 or to εx.
b) For c(t)=1⊗t+εx⊗t, it is necessary and sufficient that t be a tangent vector, i.e. that t∈Tx(1)(X).
13.6. Distributions with finite support
13.6.1. Let X be a manifold of class Cr, and let k≤r. We denote by T(k)(X) the direct sum of the spaces Tx(k)(X), for x∈X. An element of T(k)(X) is called a distribution with finite support on X, of order ≤k. If f:X⟶F is a function of class Cr with values in a separated polynomial space F, and if
t=x∈X∑tx,
where tx∈Tx(k)(X) for all x∈X is a distribution with finite support, we put
⟨f,t⟩=x∈X∑⟨f,tx⟩.
Analogously, we put
c(t)=x∈X∑c(tx),
which equips T(k)(X) with a coassociative, cocommutative cogebra structure, possessing a counit t↦⟨1,t⟩. The inclusions
T(k′)(X)⟶T(k)(X),
where k′≤k, are cogebra morphisms.
If φ:X→Y is a morphism of varieties of class Cr, the mappings Tx(k)(φ):Tx(k)(X)→Tφ(x)(k)(Y) relative to the various points x of X define a linear mapping φ∗ of T(k)(X) into T(k)(Y), which is a cogebra morphism.
If X1 and X2 are two varieties of class Cr, the mappings αk−1 (cf. 13.4.5) define a linear mapping
T(k)(X1×X2)→T(k)(X1)⊗T(k)(X2)
which is injective; it is a cogebra morphism. If k=∞ or ω, it is an isomorphism of cogebras.
13.6.2. Let X be a variety of class Cr, and let V be a K-vector space of finite dimension. An element of T(r)(X)⊗V is called a distribution with finite support on X with values in V. When K=R, V=C, such a distribution is also called a complex distribution with finite support on X. The definitions and results of the preceding No.s extend immediately, by linearity, to distributions with values in V.
13.7. Weakening of structure
Suppose K=R. Let r′∈NK such that r′≤r. Let X be a variety of class Cr, and let X′ be the variety of class Cr′ obtained by weakening of structure (5.13.1). Let x∈X, and let t∈Tx(k)(X), with k≤r′. Choose a chart c=(U,φ,E) of X centred at x, and let t~ be the element of TS(k)(E) such that θc(t~)=t. Since c is a chart of X′ centred at x, the element t′=θc(t~) of Tx(k)(X′) is defined; it is independent of the choice of c. The mapping t↦t′ is a bijection of Tx(k)(X) onto Tx(k)(X′) by which these two spaces are identified. The identifications thus obtained are compatible with the operations ⟨f,t⟩,φ∗(t),t1⊗t2,c(t),… of the preceding No.s.
[^1]: Here again, the product of the γαi(∂i,x) is a symmetric product in TS(Tx(X)).