Bourbaki

§ 5. Manifolds

5.1. Charts and atlases. Manifolds

5.1.1. Let X be a set. A chart of X is a triple c = (U, φ, E), where U is a subset of X, E a Banach space and φ a bijection of U onto an open subset of E. We say that U is the domain of the chart c. If E is of finite dimension n, we say that c is of dimension n. Otherwise, we set dim c = +∞.

5.1.2. We say that two charts c = (U, φ, E) and c' = (U', φ', E') of X are C^r-compatible (or simply compatible when there can be no ambiguity about r) when the following conditions are satisfied: (a) φ(U ∩ U') (resp. φ'(U ∩ U')) is open in E (resp. E'); (b) the mapping φ ∘ φ'^{-1} (resp. φ' ∘ φ^{-1}) of φ'(U ∩ U') onto φ(U ∩ U') (resp. of φ(U ∩ U') onto φ'(U ∩ U')) is of class C^r (cf. 2.3.1, 3.2.1 and 4.2.1).

5.1.3. A C^r-atlas (or simply atlas) of a set X is a set of pairwise C^r-compatible charts of X whose domains have X itself as their union. Two atlases A and B of X are said to be C^r-equivalent if A ∪ B is an atlas. The relation of C^r-equivalence between atlases is an equivalence relation.

5.1.4. Let S be a set of Banach spaces. We say that an atlas A is of type S if E ∈ S for every chart c = (U, φ, E) of A. In an analogous manner, we say that an atlas A is of Hilbert type (resp. of countable-dimensional Hilbert type) if E is a Hilbert space (resp. a Hilbert space of countable type) for every chart (U, φ, E) of A.

5.1.5. A K-manifold of class C^r (or a manifold of class C^r over K, or simply a manifold when there can be no ambiguity about K and r) is a set X endowed with a class of equivalent atlases (Ens., chap. II, § 6, No. 9) for the relation of C^r-equivalence. An atlas of this class is called an atlas of the manifold X. A chart belonging to an atlas of X is called a chart of the manifold X. A chart of X whose domain contains a point a ∈ X is called a chart of X at a. A chart centred at a is a chart (U, φ, E) of X at a such that φ(a) = 0.

If X is a set and A an atlas of X, the set X endowed with the equivalence class of A is called the manifold defined by A.

When r ≠ ω (hence K = R), a manifold of class C^r is sometimes called a differential manifold. A manifold of class C^ω is also called an analytic manifold over K (or K-analytic manifold).

When in addition K=RK = \mathbf{R} (resp. C\mathbf{C}, Qp\mathbf{Q}_p), we also say real analytic manifold (resp. complex analytic manifold, pp-adic analytic manifold).

5.1.6. Let XX be a manifold. The subsets of XX which are unions of domains of charts of XX form the set of open subsets for a topology on XX, called the topology underlying the manifold structure of XX. For every chart c=(U,φ,E)c = (U, \varphi, E) of XX, the mapping φ\varphi is a homeomorphism of the open subset UU endowed with the topology induced by that of XX onto the open subset φ(U)\varphi(U) of EE.

The topological space underlying XX is a Baire space. When KK is equal to R\mathbf{R} or C\mathbf{C}, it is locally connected.

Let XX be a manifold and let A\mathscr{A} be an atlas of XX. For the topological space XX to be separated, it is necessary and sufficient that the following condition be satisfied: whatever the charts (U,φ,E)(U, \varphi, E) and (V,ψ,F)(V, \psi, F) belonging to the atlas A\mathscr{A}, the graph of the mapping ψ∘φ−1\psi \circ \varphi^{-1} of φ(U∩V)\varphi(U \cap V) into ψ(U∩V)\psi(U \cap V) is closed in φ(U)×ψ(V)\varphi(U) \times \psi(V).

Let XX be a manifold; suppose that the topological space XX is regular. Then for every point a∈Xa \in X, there exists a chart (U,φ,E)(U, \varphi, E) of XX at aa having the following property: for a subset YY of UU to be closed in XX, it is necessary and sufficient that its image φ(Y)\varphi(Y) be closed in EE. If the space XX is paracompact, there exists on XX a distance compatible with the topology of XX and making XX a complete metric space.

5.1.7. Let S\mathfrak{S} be a set of Banach spaces. We say that a manifold is of type S\mathfrak{S} (resp. of Hilbert type, resp. of countable-dimensional Hilbert type) if it possesses an atlas of type S\mathfrak{S} (resp. of Hilbert type, resp. of countable-dimensional Hilbert type). If S\mathfrak{S} is reduced to a single element EE, a manifold of type S\mathfrak{S} is also called a purely manifold of type EE.

A pure manifold of dimension nn is a pure manifold of type KnK^n. We say that a manifold is locally of finite dimension if it is of type S\mathfrak{S} with S={Kn;n∈N}\mathfrak{S} = \{ K^n ; n \in \mathbf{N} \}.

5.1.8. Let XX be a manifold and let x∈Xx \in X. The dimension (finite or +∞+\infty) of a chart of XX at xx depends only on xx. It is called the dimension of XX at xx and is denoted by dim⁡xX\dim_x X. The dimension of XX, denoted by dim⁡X\dim X, is the supremum of the dim⁡xX\dim_x X for x∈Xx \in X.

The mapping x↦dim⁡xXx \mapsto \dim_x X is locally constant. The manifold XX is locally of finite dimension (resp. purely of dimension nn) if and only if dim⁡xX<+∞\dim_x X < +\infty (resp. dim⁡xX=n\dim_x X = n) for every x∈Xx \in X.

5.1.9. Suppose KK is locally compact. Let XX be a separated manifold. Then, for XX to be locally of finite dimension, it is necessary and sufficient that XX be locally compact.

No. 5.2

5.1.10. Let X be a manifold and let ξ1,…,ξn\xi^1, \ldots, \xi^n be mappings of a subset U of X into K. We say that ξ=(ξ1,…,ξn)\xi = (\xi^1, \ldots, \xi^n) is a system of coordinates of X in U if the triplet (U,ξ,Kn)(U, \xi, K^n) is a chart of X; this chart is also denoted by (U;ξ)(U; \xi) or (U;ξ1,…,ξn)(U; \xi^1, \ldots, \xi^n). If a∈Ua \in U, we also say that ξ\xi is a system of coordinates of X at aa; if in addition ξi(a)=0\xi^i(a) = 0 for i=1,2,…,ni = 1, 2, \ldots, n, we say that the system of coordinates ξ\xi is centred at aa.

5.2. Examples of manifolds

5.2.1. Let X be a set; there exists on X one and only one manifold structure for which the underlying topological space is discrete; this structure is a purely 0-dimensional manifold structure.

5.2.2. Let E be a Banach space. The triplet c=(E,IdE,E)c = (E, \mathrm{Id}_E, E) is a chart of E and A={c}\mathscr{A} = \{c\} is an atlas of E, and hence defines a purely manifold structure of type E on E; the underlying topology is the topology given on E. Whenever in what follows we speak of the manifold structure on E, it will always be the preceding structure to which we refer.

In particular, this applies to every finite-dimensional vector space over K, endowed with the unique separated topology compatible with its vector structure (Esp. Vect. Top., chap. I, § 2, No. 3).

5.2.3. Let X be a manifold and U an open subset of X. There exists on U a manifold structure whose charts are the charts of the manifold X with domain contained in U. This structure is said to be induced by that of X (cf. No. 5.3.4); endowed with this structure, U is called an open submanifold of X.

In particular, every open subset of a Banach space E is endowed with a canonical purely manifold structure of type E. Let X be a manifold; for a triplet (U,φ,E)(U, \varphi, E) to be a chart of X, it is necessary and sufficient that U be open in X and that φ\varphi be an isomorphism of the open submanifold U of X onto an open submanifold of E.

5.2.4. Let X be a set, and (Xi)i∈I(X_i)_{i \in I} a covering of X. Suppose that each XiX_i is endowed with a manifold structure such that the following condition is fulfilled:

For any ii and jj in I, the set Xi∩XjX_i \cap X_j is open in XiX_i and XjX_j and the manifold structures induced on Xi∩XjX_i \cap X_j by those of XiX_i and XjX_j coincide.

There then exists on X one and only one manifold structure such that XiX_i is an open submanifold of X for every ii in I. We say that the manifold X is obtained by gluing the manifolds XiX_i.

5.2.5. Let X be a manifold. The set XnX_n of points xx of X such that dim_x X = n (n integer ≥0\geq 0) is an open submanifold of X, which is purely of dimension n.

5.2.6. Let E be a Banach space. The set G(E) of vector subspaces of E admitting a topological supplement can be endowed with an analytic manifold structure in the following way: for every pair (F_0, G_0) ∈G(E)×G(E)\in G(E) \times G(E) such that E = F_0 ⊕\oplus G_0, let U_{G_0} denote the set of F ∈G(E)\in G(E) admitting G_0 as a supplement, and define a bijection φF0,G0\varphi_{F_0, G_0} of U_{G_0} onto the Banach space L(F0;G0)\mathscr{L}(F_0; G_0) by associating to each F ∈UG0\in U_{G_0} the mapping from F_0 into G_0 having as graph the subspace F of E = F_0 ×\times G_0. The charts (U_{G_0}, φF0,G0\varphi_{F_0, G_0}, L(F0;G0)\mathscr{L}(F_0; G_0)) form an atlas of G(E). Endowed with the manifold structure defined by this atlas, G(E) is called the Grassmann manifold of E.

The topological space G(E) is metrisable. If K is locally compact and E finite-dimensional, G(E) is compact.

For every integer n≥0n \geq 0, the set G_n(E) (resp. G^n(E)) of F ∈G(E)\in G(E) of dimension (resp. codimension) n is open and closed in G(E) and is a purely open submanifold of G(E). If K = R or C, G_n(E) is connected for every n.

The mapping which to F ∈G(E)\in G(E) associates its orthogonal in the dual E’ of E is a morphism of G(E) into G(E’) which induces an isomorphism of G^n(E) onto G_n(E’) for every integer n.

If K = R or C, or if E is finite-dimensional, G_1(E) is nothing other than the projective space deduced from E, which is thus endowed with an analytic manifold structure.

5.3. Functions of class CrC^r and morphisms of manifolds

5.3.1. Let X be a manifold of class C', F a separated polynormed space and f a mapping from X into F. One says that f is of class C' if for every chart (V, φ\varphi, E) of X, the mapping f∘φ−1f \circ \varphi^{-1} of φ(V)\varphi(V) into F is of class C'. For this, it is enough that this condition be satisfied for the charts of an atlas of X. The set of mappings of class C' from X into F forms a vector subspace of the space of all mappings from X into F. It is denoted by C′(X;F)\mathscr{C}'(X; F).

When F = K, one sets C′(X;K)=C′(X)\mathscr{C}'(X; K) = \mathscr{C}'(X); this is a sub-K-algebra of the algebra of mappings from X into K. The elements of C′(X)\mathscr{C}'(X) are also called morphic functions.

When X is an open subset of a Banach space, this terminology agrees with that of Nos. 2.3.1, 3.2.1 and 4.2.1.

5.3.2. Let X and Y be two manifolds of class C' and let f be a mapping from X into Y. One says that f is of class C' or is a morphism of manifolds (of class C') if it is continuous and if, for every chart (V, ψ\psi, F) of Y, the mapping ψ∘f\psi \circ f of the open submanifold f−1(V)f^{-1}(V) into the Banach space F is of class C′C'. For this, it is enough that there exist an atlas A\mathscr{A} of Y such that, for every chart (V,ψ,F)∈A(V, \psi, F) \in \mathscr{A}, the set f−1(V)f^{-1}(V) be open in X and the mapping ψ∘f\psi \circ f of the open submanifold f−1(V)f^{-1}(V) of X into F be of class C′C'. The set of morphisms from X into Y is denoted by C′(X;Y)\mathscr{C}'(X; Y). When Y is a Banach space endowed with its canonical manifold structure, the definitions of 5.3.1. and 5.3.2 are coherent. A mapping of class CωC^\omega is also called a K-analytic mapping (or simply analytic). When K=CK = \mathbf{C}, one also says holomorphic mapping.

Let (U,φ,E)(U, \varphi, E) be a chart of X and (V,ψ,F)(V, \psi, F) a chart of Y such that f(U)⊂Vf(U) \subset V. The mapping ψ∘f∘φ−1\psi \circ f \circ \varphi^{-1} of the open set φ(U)\varphi(U) of E into the open set ψ(V)\psi(V) of F is called the expression of f in the given charts.

5.3.3. Suppose X and Y finite-dimensional; let a∈X, b∈Ya \in X,\ b \in Y and ff a mapping from X into Y with f(a)=bf(a) = b. Suppose first that ff is of class C′C' and consider coordinate systems (U;ξ1,…,ξm)(U; \xi^1, \ldots, \xi^m) of X at aa and (V;η1,…,ηn)(V; \eta^1, \ldots, \eta^n) of Y at bb respectively, with f(U)⊂Vf(U) \subset V. Let ξ\xi be the mapping (ξ1,…,ξm)(\xi^1, \ldots, \xi^m) of U into KmK^m. Then there exist functions uju^j of class C′C' on the open set ξ(U)\xi(U) of KmK^m, with values in K, such that the coordinates of a point y=f(x)y = f(x) of Y, (for x in U), are given by the formulae:

(1)

ηj(y)=uj(ξ1(x),…,ξm(x))for 1≤j≤n \eta^j(y) = u^j(\xi^1(x), \ldots, \xi^m(x)) \quad \text{for } 1 \leq j \leq n
which is equivalent to:

(2)

ηj∘f=uj(ξ1,…,ξm)for 1≤j≤n. \eta^j \circ f = u^j(\xi^1, \ldots, \xi^m) \quad \text{for } 1 \leq j \leq n.
One says that the preceding formulae constitute the expression of ff by means of the chosen coordinate systems.

faisceau de fonctions à valeurs dans YY if it satisfies the following condition:

Let (Ui)i∈I(U_i)_{i \in I} be a family of open sets of XX, with union UU, and let ff be a mapping of UU into YY; in order that ff belong to L(U)\mathscr{L}(U), it is necessary and sufficient that f∣Uif|U_i belong to L(Ui)\mathscr{L}(U_i) for every ii in II.

5.4.2. Let XX and YY be two manifolds; for every open set U⊂XU \subset X, let L(U)\mathscr{L}(U) be the set of morphisms of UU into YY; then L\mathscr{L} is a sheaf of mappings with values in YY.

When Y=KY = K, the sheaf thus defined is denoted by CXr′\mathscr{C}_X^{r'}.

5.4.3. Let XX be a topological space and let S\mathscr{S} be a set of Banach spaces. For every E∈SE \in \mathscr{S}, let FE\mathscr{F}_E be a sheaf of mappings on XX with values in EE. Suppose that the family of the FE\mathscr{F}_E, for E∈SE \in \mathscr{S}, satisfies the following condition:

For every x∈Xx \in X, there exist an open neighbourhood UU of xx, a space

E0∈SE_0 \in \mathscr{S}, and a homeomorphism φ\varphi of U onto an open set of E_0, such that, for every open set V⊂UV \subset U and every E∈SE \in \mathscr{S}, the set FE(V)\mathscr{F}_E(V) consists of the functions g∘φg \circ \varphi, where gg ranges over the space Cr(φ(V);E)\mathscr{C}^r(\varphi(V); E).

Then there exists on X a structure of manifold of class CrC^r and of type S\mathscr{S}, and only one, compatible with the given topology on X and for which FE\mathscr{F}_E is the sheaf of functions of class CrC^r on X with values in E.

5.4.4. Let X be a topological space and let F\mathscr{F} be a sheaf of functions with values in K, satisfying the following condition: for every point x of X, there exist an integer n, an open neighbourhood U of x and a homeomorphism φ\varphi of U onto an open set of KnK^n, such that, for every open set V of U, the set F(V)\mathscr{F}(V) consists of the functions g∘φg \circ \varphi where gg ranges over the set of functions of class CrC^r with values in K on the open set φ(V)\varphi(V) of KnK^n.

Then there exists on X a structure of locally finite-dimensional manifold of class CrC^r and only one such that F=CXr\mathscr{F} = \mathscr{C}^r_X.

5.4.5. Let XX and X′X' be two CrC^r-class varieties, locally of finite dimension, and let ff be a continuous mapping of X into X'. For ff to be a morphism, it is necessary and sufficient that for every open set U' of X' and every function g∈Cr(U′;K)g \in \mathscr{C}^r(U'; K), the function g∘fg \circ f belong to Cr(U;K)\mathscr{C}^r(U; K), where U=f−1(U′)U = f^{-1}(U').

5.5. Tangent spaces, tangent linear mappings

5.5.1. Let X be a variety and let a∈Xa \in X. Consider the pairs (c,h)(c, h) where c=(U,φ,E)c = (U, \varphi, E) is a chart of the variety X at aa and where hh is an element of E. Two such pairs (c,h)(c, h) and (c′,h′)(c', h') are said to be equivalent if the derivative at φ(a)\varphi(a) of the mapping φ′∘φ−1\varphi' \circ \varphi^{-1} (which is defined on a neighbourhood of φ(a)\varphi(a)) transforms hh into h′h'. We thus obtain an equivalence relation between pairs (c,h)(c, h) and call a tangent vector at aa to X a class of equivalent pairs (c,h)(c, h) (Ens., Chapter II, § 6, No. 9).

The tangent vectors at aa to X form a set denoted by Ta(X)T_a(X). If c=(U,φ,E)c = (U, \varphi, E) is a chart of the variety X at aa, the mapping θc\theta_c of E into Ta(X)T_a(X) which associates to an element hh of E the tangent vector represented by the pair (c,h)(c, h) is a bijection. If one transports to Ta(X)T_a(X), by means of θc\theta_c, the structure of topological K-vector space of E, the structure thus obtained does not depend on the choice of cc and makes Ta(X)T_a(X) a Banach space, called the tangent space to X at aa. The dimension (finite or +∞+\infty) of Ta(X)T_a(X) is equal to the dimension of X at aa.

5.5.2. Let X, Y be two manifolds, ff a morphism of X into Y and aa a point of X. Consider a chart c=(U,φ,E)c = (U, \varphi, E) of X at aa and a chart c′=(V,ψ,F)c' = (V, \psi, F) of Y at bb with f(U)⊂Vf(U) \subset V; the mapping Φ=ψ∘f∘φ−1\Phi = \psi \circ f \circ \varphi^{-1} of φ(U)\varphi(U) into F is of class CrC^r, and its derivative DΦ(φ(a))D\Phi(\varphi(a)) at the point φ(a)\varphi(a) is a continuous linear mapping of E into F. The continuous linear mapping θc′∘DΦ(φ(a))∘θc−1\theta_{c'} \circ D\Phi(\varphi(a)) \circ \theta_c^{-1} of Ta(X)T_a(X) into Tb(Y)T_b(Y) does not depend on the charts cc and c′c' chosen; it is denoted by Ta(f)T_a(f) and is called the tangent linear mapping to ff at aa. If gg is a morphism of YY into a manifold ZZ, we have:

Ta(g∘f)=Tf(a)(g)∘Ta(f). T_a(g \circ f) = T_{f(a)}(g) \circ T_a(f).

5.5.3. Let f:X→Yf : X \to Y be a morphism of manifolds. If ff is locally constant, we have Ta(f)=0T_a(f) = 0 for every aa in XX. Conversely, if Ta(f)=0T_a(f) = 0 for every a∈Xa \in X and if the field KK is of characteristic 0, then ff is locally constant.

5.5.4. Let UU be an open set of a manifold XX and ff the canonical injection of UU into XX; for every point aa of UU, the mapping Ta(f)T_a(f) of Ta(U)T_a(U) into Ta(X)T_a(X) is an isomorphism of topological vector spaces, by means of which these two spaces will henceforth be identified.

5.5.5. Let UU be an open set of a Banach space EE; if ι\iota is the canonical injection of UU into EE, the triple c=(U,ι,E)c = (U, \iota, E) is a chart of UU, and the atlas {c}\{c\} defines the manifold structure of UU. Given a point aa of UU, the mapping θc\theta_c is an isomorphism of topological vector spaces of EE onto Ta(U)T_a(U); the inverse isomorphism will be denoted by ζE(a)\zeta_E(a).

Let ff be a mapping of class CrC^r of UU into an open set VV of a Banach space FF; for every point aa of UU, we have:

Df(a)∘ζE(a)=ζF(f(a))∘Ta(f) Df(a) \circ \zeta_E(a) = \zeta_F(f(a)) \circ T_a(f)

where Df(a)Df(a) is the derivative of ff at aa (1.2.1). In other words, the diagram:

Ta(U)→Ta(f)Tf(a)(V)↓ζE(a)↓ζF(f(a))E→Df(a)F \begin{array}{ccc} T_a(U) & \xrightarrow{T_a(f)} & T_{f(a)}(V) \\ \downarrow \zeta_E(a) & & \downarrow \zeta_F(f(a)) \\ E & \xrightarrow{Df(a)} & F \end{array}

is commutative.

5.5.6. A cotangent vector to XX at aa, or also a tangent covector or simply a covector at aa, is any continuous linear form on the topological vector space Ta(X)T_a(X); these covectors are therefore the elements of the topological dual Ta(X)′T_a(X)' of the tangent space to XX at aa. This space will be endowed with the topology of uniform convergence on the bounded subsets of Ta(X)T_a(X): endowed with this topology, Ta(X)′T_a(X)' is a Banach space which is called the cotangent space to XX at aa. It is also denoted by Ta′(X)T'_a(X).

Let ff be a morphism of XX into a Banach space EE. The differential of ff at aa, denoted by dafd_a f or dfadf_a, is the continuous linear mapping ζE(f(a))∘Ta(f)\zeta_E(f(a)) \circ T_a(f) of Ta(X)T_a(X) into EE. For t∈Ta(X)t \in T_a(X), one sometimes denotes by t(f) or t . f the value daf(t)d_a f(t) of the linear mapping dafd_a f at the point tt. The mapping f↦daff \mapsto d_a f is linear.

If E=KE = K, the differential dafd_a f of ff at aa is a covector of XX at aa.

Let E,F,GE, F, G be three Banach spaces and let (u,v)↦u.v(u, v) \mapsto u . v be a continuous bilinear mapping of F×GF \times G into EE. Let ff (resp. gg) be a morphism of XX into FF (resp. GG). Then the mapping f.g:x↦f(x).g(x)f . g : x \mapsto f(x) . g(x) is a morphism of XX into EE and, for every t∈Ta(X)t \in T_a(X), we have:

da(f.g)(t)=f(a).dag(t)+daf(t).g(a). d_a(f . g)(t) = f(a) . d_a g(t) + d_a f(t) . g(a).

Take in particular G=EG = E and F=KF = K, the bilinear mapping considered being multiplication. We then have:

da(fg)=f(a)dag+g(a) daf. d_a(fg) = f(a)d_ag + g(a)\ d_af.

5.5.7. Let XX be a locally finite-dimensional variety and let ξ1,…,ξm\xi^1, \ldots, \xi^m be morphic functions defined in an open neighbourhood UU of a point aa of XX. The following conditions are equivalent:

(i) there exists an open neighbourhood VV of aa contained in UU, such that the functions ξi∣V\xi^i|V (for 1≤i≤m1 \leq i \leq m) form a system of coordinates of XX in VV;

(ii) the differentials daξid_a \xi^i for 1≤i≤m1 \leq i \leq m form a basis of Ta(X)′T_a(X)';

(iii) the mapping ξ=(ξ1,…,ξm)\xi = (\xi^1, \ldots, \xi^m) of UU into KmK^m is étale at aa (see No. 5.7.6).

For the differentials daξ1,…,daξmd_a \xi^1, \ldots, d_a \xi^m to be linearly independent, it is necessary and sufficient that there exist a neighbourhood VV of aa, contained in UU, such that the functions ξi∣V\xi^i|V form part of a system of coordinates of XX in VV.

5.5.8. Let XX be a locally finite-dimensional variety and let ξ=(ξ1,…,ξm)\xi = (\xi^1, \ldots, \xi^m) be a system of coordinates of XX in an open set UU. Let a∈Ua \in U: we denote by (∂1,a,…,∂m,a)(\partial_{1,a}, \ldots, \partial_{m,a}) the basis of the tangent space Ta(X)T_a(X) dual to the basis (daξ1,…,daξm)(d_a \xi^1, \ldots, d_a \xi^m) of Ta(X)′T_a(X)'. We therefore have:

∂i,a(ξj)=δi,j(Kronecker index). \partial_{i,a}(\xi^j) = \delta_{i,j} \quad \text{(Kronecker index)}.

The tangent vector ∂i,a\partial_{i,a} is also denoted by (∂/∂ξi)a(\partial/\partial \xi^i)_a.

Let ff be a function of class CrC^r on UU, with values in a Banach space EE. We denote by ∂if\partial_i f or ∂f/∂ξi\partial f/\partial \xi^i the function a↦∂i,a(f)a \mapsto \partial_{i,a}(f): this is a function of class Cr−1C^{r-1} on UU with values in EE (a continuous function if r=1r = 1). Its value at a point aa of UU is sometimes denoted by (∂f/∂ξi)a(\partial f/\partial \xi^i)_a. For all t∈Ta(X)t \in T_a(X), we have:

daf(t)=∑i=1mdaξi(t)∂f∂ξi(a), d_a f(t) = \sum_{i=1}^m d_a \xi^i(t) \frac{\partial f}{\partial \xi^i}(a),

which may also be written:

daf=∑i=1m∂f∂ξi(a)daξi. d_a f = \sum_{i=1}^m \frac{\partial f}{\partial \xi^i}(a) d_a \xi^i.

These notations are consistent with those of 1.6.3.

5.5.9. Let XX and YY be two varieties and ff a morphism of XX into YY. The rank of ff at a point aa of XX, denoted rgaf\mathrm{rg}_a f, is the rank (finite or equal to +∞+\infty) of the linear mapping Ta(f)T_a(f). The mapping a↦rgafa \mapsto \mathrm{rg}_a f is lower semicontinuous.

5.6. Products of varieties

5.6.1. Let XX and X′X' be two sets and let c=(U,φ,E)c = (U, \varphi, E) (resp. c′=(U′,φ,E′)c' = (U', \varphi, E')) be a chart of XX (resp. X′X'). The triple

(U×U′,φ×φ′,E×E′) (U \times U', \varphi \times \varphi', E \times E')

is a chart of the product set X×X′X \times X', denoted c×c′c \times c'.

5.6.2. Let XX and X′X' be two varieties of class CrC^r. There exists on the product set X×X′X \times X' one and only one structure of a variety of class CrC^r such that c×c′c \times c' is a chart of X×X′X \times X' for every chart cc of XX and every chart c′c' of X′X'. Endowed with this structure, X×X′X \times X' is called the product variety of the varieties XX and X′X'. One defines analogously the product of any finite number of varieties. The topology of the variety X×X′X \times X' is the product topology of the topologies of XX and X′X'. For a∈Xa \in X and b∈X′b \in X', one has:

dim⁡(a,b)(X×X′)=dim⁡aX+dim⁡bX′. \dim_{(a,b)}(X \times X') = \dim_a X + \dim_b X'.

5.6.3. Let XX and X′X' be two varieties, and let X×X′X \times X' be their product. Let a∈Xa \in X and let a′∈X′a' \in X'. The canonical projections

pr1:X×X′→Xandpr2:X×X′→X′ \mathrm{pr}_1 : X \times X' \to X \quad \text{and} \quad \mathrm{pr}_2 : X \times X' \to X'

are morphisms. Let πi (i=1,2)\pi_i \ (i = 1, 2) be their tangent linear mappings at the point (a,a′)(a, a'). The mapping

(π1,π2):T(a,a′)(X×X′)→Ta(X)×Ta′(X′) (\pi_1, \pi_2) : T_{(a,a')}(X \times X') \to T_a(X) \times T_{a'}(X')

is an isomorphism, which makes it possible to identify the tangent space to X×X′X \times X' at (a,a′)(a, a') with the product Ta(X)×Ta′(X′)T_a(X) \times T_{a'}(X').

The injection of Ta(X)T_a(X) into T(a,a′)(X×X′)T_{(a,a')}(X \times X') resulting from this identification is the tangent linear mapping at aa of the morphism x↦(x,a′)x \mapsto (x, a') of XX into X×X′X \times X'; there is an analogous result for the injection of Ta′(X′)T_{a'}(X') into T(a,a′)(X×X′)T_{(a,a')}(X \times X').

5.6.4. Let W,XW, X and X′X' be three varieties, and let f:W→X,f′:W→X′f : W \to X, f' : W \to X' be two mappings. For the mapping

(f,f′):W→X×X′ (f, f') : W \to X \times X'

to be a morphism, it is necessary and sufficient that ff and f′f' be morphisms (this justifies the use of the term « product », cf. Ens., Chap. IV, § 2, No. 4). In this case, if aa is a point of WW, we have:

Ta(f,f′)=(Ta(f),Ta(f′)) T_a(f, f') = (T_a(f), T_a(f'))

taking into account the identification made above.

5.6.5. Let f:X→Yf : X \to Y and f′:X′→Y′f' : X' \to Y' be morphisms of varieties. Then f×f′:X×X′→Y×Y′f \times f' : X \times X' \to Y \times Y' is a morphism. If a∈Xa \in X and a′∈X′a' \in X', we have:

T(a,a′)(f×f′)=Ta(f)×Ta′(f′) T_{(a, a')}(f \times f') = T_a(f) \times T_{a'}(f')
rg(a,a′)(f×f′)=rgaf+rga′f′. \mathrm{rg}_{(a, a')}(f \times f') = \mathrm{rg}_a f + \mathrm{rg}_{a'} f'.

5.6.6. Let X1,X2X_1, X_2 and ZZ be three varieties and ff a morphism of X1×X2X_1 \times X_2 into ZZ. Let a∈X1a \in X_1 and b∈X2b \in X_2. The tangent linear mapping to the partial mapping x↦f(x,b)x \mapsto f(x, b) (resp. y↦f(a,y)y \mapsto f(a, y)) of X1X_1 (resp. X2X_2) into ZZ is denoted by T(a,b)1(f)T^1_{(a, b)}(f) (resp. T(a,b)2(f)T^2_{(a, b)}(f)). If we identify T(a,b)(X1×X2)T_{(a, b)}(X_1 \times X_2) with Ta(X1)×Tb(X2)T_a(X_1) \times T_b(X_2), we have:

T(a,b)(f)⋅(u,v)=T(a,b)1(f)⋅u+T(a,b)2(f)⋅v T_{(a, b)}(f) \cdot (u, v) = T^1_{(a, b)}(f) \cdot u + T^2_{(a, b)}(f) \cdot v

for every u∈Ta(X1)u \in T_a(X_1) and every v∈Tb(X2)v \in T_b(X_2).

5.6.7. (« Theorem on implicit functions »). With the hypotheses and notation of the preceding number, suppose in addition that T(a,b)2(f)T^2_{(a, b)}(f) is bijective. There then exists an open neighbourhood UU of aa in X1X_1 and an open neighbourhood VV of bb in X2X_2 having the following property: for every x∈Ux \in U, there exists one and only one point g(x)g(x) of VV such that f(x,g(x))=f(a,b)f(x, g(x)) = f(a, b), and the mapping gg is a morphism of UU into VV. For every x∈Ux \in U, we have:

Tx(g)=−(T(x,g(x))2(f))−1∘T(x,g(x))1(f). T_x(g) = - (T^2_{(x, g(x))}(f))^{-1} \circ T^1_{(x, g(x))}(f).

5.7. Immersions, étale morphisms

5.7.1. Let XX and YY be two manifolds, ff a morphism of XX into YY and aa a point of XX. Put b=f(a)b = f(a). The following conditions are equivalent: (i) The linear mapping Ta(f)T_a(f) is injective and its image is a closed vector subspace of Tb(Y)T_b(Y) admitting a topological supplement 1^1;

(ii) There exist a Banach space FF, a closed vector subspace EE of FF admitting a topological supplement, and charts (U,φ,E)(U, \varphi, E) of XX at aa and (V,ψ,F)(V, \psi, F) of YY at bb such that f(U)⊂Vf(U) \subset V and φ=ψ∘(f∣U)\varphi = \psi \circ (f|U).

(iii) There exist an open neighbourhood UU of aa, an open neighbourhood VV of bb containing f(U)f(U), a manifold ZZ, a point cc of ZZ, and an isomorphism of manifolds gg of U×ZU \times Z onto VV such that f(x)=g(x,c)f(x) = g(x, c) for all x∈Ux \in U.

(iv) There exist an open neighbourhood UU of aa, an open neighbourhood VV of bb and a morphism qq of VV into XX with f(U)⊂Vf(U) \subset V, and q(f(x))=xq(f(x)) = x for all x∈Ux \in U.

When XX and YY are finite-dimensional, the preceding conditions are also equivalent to the following:

(v) There exist an open neighbourhood UU of aa, a coordinate system (η1,…,ηn)(\eta^1, \ldots, \eta^n) of YY in an open neighbourhood VV of bb containing f(U)f(U) and an integer m≤nm \leq n such that ηj∘f=0\eta^j \circ f = 0 for m<j≤nm < j \leq n and the functions η1∘f,…,ηm∘f\eta^1 \circ f, \ldots, \eta^m \circ f form a coordinate system of XX in UU.

If properties (i) to (iv) are satisfied, we say that ff is an immersion at aa.

The set of points where ff is an immersion is open in XX; if this open set is all of XX, we say that ff is an immersion.

For ff to be an immersion, it is necessary and sufficient that XX can be covered by open sets UiU_i such that, for every i,f∣Uii, f|U_i is an isomorphism of UiU_i onto a subvariety of YY (cf. No. 5.8.3).

5.7.2. Examples: (a) If XX is an open set in a variety YY, the canonical injection of XX into YY is an immersion. (b) Let EE and FF be two Banach spaces and let uu be a continuous linear mapping of EE into FF. Then uu is an immersion if and only if uu is an isomorphism of EE onto a closed vector subspace of FF admitting a topological supplement.

5.7.3. Let f:X→Yf : X \to Y and g:Y→Zg : Y \to Z be two morphisms. If ff and gg are immersions, g∘fg \circ f is an immersion. Conversely, if g∘fg \circ f is an immersion, then ff is an immersion. If f:X→Yf : X \to Y and f′:X′→Y′f' : X' \to Y' are immersions, f×f′f \times f' is an immersion.

5.7.4 Suppose that XX and YY are varieties of finite dimension over a field KK of characteristic 0. If f:X→Yf : X \to Y is an injective morphism, the set of points of XX where ff is an immersion is an open dense set

1^1 Every closed vector subspace of finite codimension of a Banach space admits a topological supplement; if K=RK = \mathbf{R} or C\mathbf{C}, the same is true of every finite-dimensional vector subspace (necessarily closed).

in XX. If K = R or C, this result remains true if one supposes only that X is of finite dimension.

5.7.5. Let f:X→Yf : X \to Y be an immersion and let gg be a continuous mapping of a variety Z into X. For gg to be a morphism, it is necessary and sufficient that f∘gf \circ g be a morphism.

5.7.6. Let f:X→Yf : X \to Y be a morphism, and let aa be in X. The following two properties are equivalent: (i) Ta(f)T_a(f) is bijective. (ii) There exists an open neighbourhood U of aa and an open neighbourhood V of f(a)f(a) such that ff induces an isomorphism of U onto V. When these properties are satisfied, one says that ff is a local isomorphism at aa, or that ff is étale at aa. If this holds for every a∈Xa \in X, one says that ff is étale, or is an étalement, or that X is étalé in Y (by means of ff). For a morphism to be étale, it is necessary and sufficient that it be both an immersion (No. 5.7.1) and a submersion (No. 5.9.1.).

5.7.7. Let f:X→Yf : X \to Y be an immersion; suppose the variety X is purely finite-dimensional. Then ff is étale in each of the following two cases: (i) dim⁡Y=dim⁡X\dim Y = \dim X; (ii) ff is surjective and the topology of X has a countable basis of open sets.

5.7.8. For a morphism ff to be an isomorphism onto an open subvariety (resp. an isomorphism), it is necessary and sufficient that ff be étale and injective (resp. étale and bijective).

5.8. Inverse images of structures of variety, subvarieties

5.8.1. Let X be a topological space, Y a variety and ff a mapping of X into Y. Consider the following conditions: (QR) (resp. (R)) For every a∈Xa \in X, there exists an open neighbourhood U of a in X and a chart (V,φ,E)(V, \varphi, E) of the variety Y at f(a)f(a) such that f(U)⊂Vf(U) \subset V and φ∘f\varphi \circ f induces a homeomorphism of U onto the intersection of φ(V)\varphi(V) with a closed vector subspace (resp. closed and admitting a topological complement) F of E. If condition (QR) is satisfied, there exists on X one and only one structure of variety for which the triples (U,φ∘f,F)(U, \varphi \circ f, F) (with the notation of (QR)) are charts of X. It is called the inverse image structure of the structure of variety of Y by ff. Its topology is that of X. For there to exist on X a structure of variety compatible with the given topology and for which ff be an immersion, it is necessary and sufficient that condition (R) be satisfied. This structure is then unique: it is the inverse image by ff of the structure of variety of YY.

5.8.2. The condition (R) above is in particular satisfied when, for every a∈Xa \in X, there exists an open neighbourhood UU of aa such that f∣Uf|U is a homeomorphism of UU onto an open set of YY. In this case, the variety XX obtained is étalé over YY (5.7.6).

5.8.3. Suppose now that XX is a topological subspace of YY, ff being the canonical injection. If ff satisfies condition (R) (resp. (QR)) of No. 5.8.1, one says that XX, endowed with the inverse image structure of the structure of variety of YY by ff, is a subvariety (resp. quasi-subvariety) of YY. A subvariety is a quasi-subvariety.

Every quasi-subvariety is locally closed; every open set is a subvariety and its structure is that defined in No. 5.2.3.

celle de YY (Ens., ch. IV, § 2, No. 4).

5.8.6. If XX is a submanifold (resp. quasi-submanifold) of YY and if YY is a submanifold of a manifold ZZ, then XX is a submanifold (resp. quasi-submanifold) of ZZ.

5.8.7. Let XiX_i (for i=1,2i = 1, 2) be a manifold and let YiY_i be a submanifold (resp. quasi-submanifold) of XiX_i (for i=1,2i = 1, 2). Then Y1×Y2Y_1 \times Y_2 is a submanifold (resp. quasi-submanifold) of X1×X2X_1 \times X_2.

5.8.8. Let X be a quasi-submanifold of a manifold Y; let x be a point of X, and denote by f the canonical injection of X into Y. The map T_x(f) : T_x(X) → T_x(Y) is injective and enables us to identify the space T_x(X) with a closed vector subspace of T_x(Y). The quotient topological vector space T_x(Y)/T_x(X) is a Banach space, called the transverse space to X (in Y) at x. Its dimension (finite or +∞) is called the codimension of X in Y at the point x.

If moreover X is a submanifold of Y, then the space T_x(X) admits a topological complement in T_x(Y).

5.8.9. Let f be a mapping of a manifold X into a manifold Y, and let Γ be its graph. In order that f be a morphism, it is necessary and sufficient that the following two conditions be satisfied: (i) Γ is a submanifold of X × Y. (ii) For every (x, y) ∈ Γ, one has

T(x,y)(X×Y)=T(x,y)(Γ)⊕Ty(Y). T_{(x,y)}(X \times Y) = T_{(x,y)}(\Gamma) \oplus T_y(Y).
If this is so, the mapping pr1\mathrm{pr}_1 induces an isomorphism of Γ\Gamma onto X, and T(x,y)(Γ)T_{(x,y)}(\Gamma) is identified with the graph of Tx(f)T_x(f).

In particular, the diagonal of X × X is a submanifold of X × X.

5.8.10. Let Y be a manifold, and let (f_i)_{i∈I} be a finite family of morphic functions on Y. Let X be the set of x ∈ Y such that f_i(x) = 0 for every i. Make the following assumption: (J) For every x ∈ X, the differentials d_x f_i are linearly independent in T'_x(Y).

Then X is a closed submanifold of Y, the tangent space T'_x(X) is the subspace of T_x(Y) formed by the α such that α . f_i = 0 for every i in I. Moreover, the codimension of X in Y is equal to Card (I) at each of its points.

5.8.11. (Simple zeros of an ideal). Let a be an ideal of the algebra of polynomials K[X_1, ..., X_n]. A point x = (x_1, ..., x_n) of K^n is called a zero of a if f(x) = 0 for every f ∈ a. If x ∈ K^n, let S_x denote the subalgebra of K(X_1, ..., X_n) formed by the fractions f/g, with f, g ∈ K[X_1, ..., X_n], and g(x) ≠ 0; let a_x denote the ideal of S_x generated by a in S_x. A point x is called a simple zero of a if it is a zero of a and if the following condition is satisfied: (S) There exists a finite sequence (f_1, ..., f_m) of elements of a which generate the ideal a_x and whose differentials at x are linearly independent. (This condition is equivalent to saying that the local ring S_x/a_x is regular (Alg. Comm., to appear).) Let Z (resp. Z_s) be the set of zeros (resp. of simple zeros) of a. The set Z is closed in K^n, Z_s is open in Z, and Z_s is a submanifold of KnK^n. If x∈Zsx \in Z_s, the ideal K[X1,…,Xn]∩axK[X_1, \ldots, X_n] \cap a_x consists of the polynomials ff vanishing in a neighbourhood of xx in ZsZ_s.

Let aˉ\bar{a} be the ideal of polynomials vanishing on ZZ. If KK is algebraically closed, the set of simple zeros of aa is dense in ZZ.

5.8.12. Let XX be a manifold and LL the set of pairs (x,Z)(x, Z) where xx is a point of XX and ZZ a submanifold of XX containing xx. Given two pairs π=(x,Z)\pi = (x, Z) and π′=(x′,Z′)\pi' = (x', Z'), we shall denote by R{π,π′}R\{\pi, \pi'\} the relation:

« If x=x′x = x' and there exists a neighbourhood UU of xx such that U∩Z=U∩Z′U \cap Z = U \cap Z' ». Then RR is an equivalence relation in LL; we denote by γx(Z)\gamma_x(Z) the equivalence class of the pair (x,Z)∈L(x, Z) \in L. On the set J=L/RJ = L/R, there exists one and only one manifold structure satisfying the following condition: For every subvariety ZZ of XX, the mapping x↦γx(Z)x \mapsto \gamma_x(Z) from ZZ into JJ is an isomorphism of ZZ onto an open subvariety of JJ. We say that JJ is the manifold of germs of subvarieties of XX (cf. Top. Gén., Chapter I, 4th ed., § 6, No. 10). The mapping ρ:J→X\rho : J \to X defined by ρ(γx(Z))=x\rho(\gamma_x(Z)) = x is an immersion; we call it the canonical immersion of JJ into XX. If XX is a separated analytic manifold of finite dimension, the same is true of JJ.

5.9. Submersions and quotient manifolds

5.9.1. Let f:X→Yf : X \to Y be a morphism of varieties, let aa be a point of XX, and put b=f(a)b = f(a). The following conditions are equivalent: (i) The linear mapping Ta(f)T_a(f) is surjective, and its kernel admits a topological complement in Ta(X)T_a(X). (ii) There exist a chart (U,φ,E)(U, \varphi, E) of XX at aa, a chart (V,ψ,F)(V, \psi, F) of YY at bb, and a surjective continuous linear mapping uu of EE into FF such that

f(U)⊂V,ψ∘f=u∘φ f(U) \subset V, \quad \psi \circ f = u \circ \varphi
and such that the kernel of uu admits a topological complement in EE. (iii) There exist an open neighbourhood UU of aa, an open neighbourhood VV of bb containing f(U)f(U), and a morphism gg of UU into a variety ZZ such that the mapping (f,g)(f, g) of UU into V×ZV \times Z is an isomorphism. (iv) There exist an open neighbourhood VV of bb and a morphism ss of VV into XX such that s(b)=as(b) = a and f(s(y))=yf(s(y)) = y for every yy in VV ("local section"). When XX and YY are finite-dimensional, the preceding conditions are equivalent to the following condition: (v) There exist an open neighbourhood UU of aa, an open neighbourhood VV of bb containing f(U)f(U), and a coordinate system (η1,…,ηn)(\eta^1, \ldots, \eta^n) on VV such that the functions ηi∘f\eta^i \circ f on UU form part of a coordinate system on UU. If properties (i) to (iv) are satisfied, one says that ff is a submersion at aa. The set of points where ff is a submersion is open in XX; if this open set is the whole of XX, one says that ff is a submersion.

5.9.2. Let f:X→Yf : X \to Y and g:Y→Zg : Y \to Z be two morphisms. If ff and gg are submersions, then so is g∘fg \circ f; conversely, if g∘fg \circ f is a submersion and ff is surjective, then gg is a submersion.

5.9.3. If f:X→Yf : X \to Y and f′:X′→Y′f' : X' \to Y' are submersions, f×f′f \times f' is a submersion.

5.9.4. A submersion f:X→Yf : X \to Y is an open mapping (cf. Top. gén., chap. I, 4e éd., § 5, n° 1); in particular, the equivalence relation R defined by ff is open, ff defines by passing to the quotient a homeomorphism of X/RX/R onto f(X)f(X), and f(X)f(X) is open in YY.

5.9.5. Let RR be an equivalence relation on a variety XX. We say that RR is regular if there exists on the quotient space X/RX/R a structure of a variety such that the canonical projection p:X→X/Rp : X \to X/R is a submersion; this structure of a variety is then unique; it is the quotient of that of XX (Ens., ch. IV, § 2, n° 6): in other words, in order that a mapping gg of X/RX/R into a variety YY be a morphism, it is necessary and sufficient that g∘pg \circ p be a morphism of XX into YY.

Let C⊂X×XC \subset X \times X be the graph of RR. In order that RR be regular, it is necessary and sufficient that the following two conditions be satisfied: (i) CC is a subvariety of X×XX \times X. (ii) The mapping pr1\mathrm{pr}_1 of CC into XX is a submersion.

Condition (ii) also means that if aa and bb are congruent modulo RR, there exists an open neighbourhood UU of aa and a morphism ss of UU into XX such that s(a)=bs(a) = b and such that s(x)s(x) is congruent to xx modulo RR for every x∈Ux \in U.

Suppose that RR is regular. In order that the quotient variety X/RX/R be separated, it is necessary and sufficient that the graph of RR be closed in X×XX \times X.

5.9.6. Let XX and X′X' be two manifolds, RR and R′R' regular equivalence relations on XX and X′X', and let f:X→X′f : X \to X' be a morphism compatible with the relations RR and R′R'. The mapping f~:X/R→X′/R′\tilde{f} : X/R \to X'/R' deduced from ff by passing to the quotients is then a morphism.

5.9.7. (« Transitivity of quotients ») Let RR and SS be two equivalence relations on a manifold XX such that RR entails SS, and let S/RS/R be the quotient equivalence relation on X/RX/R. Suppose that RR is regular. Then, for SS to be regular, it is necessary and sufficient that S/RS/R be so; if this is the case, the canonical bijection

(X/R)/(S/R)→X/S (X/R)/(S/R) \to X/S

is an isomorphism of manifolds.

5.9.8. (« Products of quotients ») Let (Xi)i∈I(X_i)_{i \in I} be a finite family of manifolds, each endowed with a regular equivalence relation RiR_i. Let X=∏i∈IXiX = \prod_{i \in I} X_i, and let RR be the equivalence relation on XX which is the product of the RiR_i (cf. Top. Gén., chap. I, 4e éd., § 5, n° 3, cor. de la prop. 8). Then RR is regular, and the canonical bijection of X/RX/R onto ∏i∈I(Xi/Ri)\prod_{i \in I} (X_i/R_i) is an isomorphism of manifolds.

5.10. Subimmersions

5.10.1. Let f:X→Yf : X \to Y be a morphism of manifolds and let Γ\Gamma be the graph of ff. The mapping j:x↦(x,f(x))j : x \mapsto (x, f(x)) is an immersion of XX into X×YX \times Y, whose image is the submanifold Γ\Gamma, and f=pr2∘jf = \mathrm{pr}_2 \circ j is the composite of the immersion jj followed by the submersion pr2\mathrm{pr}_2.

Let a∈Xa \in X. We say that ff is a submersion at aa if there exists an open neighbourhood UU of aa, a manifold ZZ, a submersion ss of UU into ZZ and an immersion ii of ZZ into YY such that f∣U=i∘sf|U = i \circ s. The set of points of XX at which ff is a submersion is an open set of XX; if this open set is all of XX, we say that ff is a submersion.

5.10.2. Immersions and submersions are submersions. If f:X→Yf : X \to Y is a submersion, g:Y→Zg : Y \to Z an immersion and h:W→Xh : W \to X a submersion, then g∘f∘hg \circ f \circ h is a submersion.

If ff and f′f' are submersions, f×f′f \times f' is a submersion.

5.10.3. In order that f:X→Yf : X \to Y be a submersion at a point aa of XX, it is necessary and sufficient that there exist a chart (U,φ,E)(U, \varphi, E) of XX at aa, a chart (V,ψ,F)(V, \psi, F) of YY at the point f(a)f(a) and a continuous linear mapping gg of EE into FF such that: (i) f(U)⊂V, g(φ(U))⊂ψ(V)f(U) \subset V,\ g(\varphi(U)) \subset \psi(V) and f∣U=ψ−1∘g∘φf|U = \psi^{-1} \circ g \circ \varphi; (ii) the kernel (resp. the image) of gg is a closed subspace of EE (resp. FF) admitting a topological complement.

5.10.4. Let XX and YY be two finite-dimensional manifolds. In order that a morphism ff of XX into YY be a submersion at a point aa of XX, it is necessary and sufficient that there exist a coordinate system (ξ1,…,ξm)(\xi^1, \ldots, \xi^m) of XX at aa, a coordinate system (η1,…,ηn)(\eta^1, \ldots, \eta^n) of YY at f(a)f(a) and an integer k≤inf⁡(m,n)k \leq \inf(m, n) such that

ηi∘f=ξifor 1≤i≤kηi∘f=0for k<i≤n. \eta^i \circ f = \xi^i \quad \text{for } 1 \leq i \leq k \\ \eta^i \circ f = 0 \quad \text{for } k < i \leq n.

5.10.5. Let f:X→Yf : X \to Y be a submersion. For every b∈Yb \in Y, f−1(b)f^{-1}(b) is a submanifold of XX; the subspace of Ta(X)T_a(X) tangent to the submanifold f−1(b)f^{-1}(b) at the point a∈f−1(b)a \in f^{-1}(b) is the kernel of Ta(f)T_a(f).

5.10.6. (“Constant rank theorem”) Let f:X→Yf : X \to Y be a morphism of manifolds and let a∈Xa \in X. If ff is a subimmersion at aa, then rgxf=rgaf\mathrm{rg}_x f = \mathrm{rg}_a f for xx in a neighbourhood of aa.

Conversely, suppose that the field KK is of characteristic zero. Let (U,φ,E)(U, \varphi, E) be a chart of XX at aa and let (V,ψ,F)(V, \psi, F) be a chart of YY at f(a)f(a), with f(U)⊂Vf(U) \subset V. Put g=ψ∘f∘φ−1g = \psi \circ f \circ \varphi^{-1}. If there exist a closed vector subspace E1E_1 of EE and a closed vector subspace F1F_1 of FF such that for every x∈Ux \in U, the subspace E1E_1 (resp. F1F_1) is a topological complement of the kernel (resp. of the image) of the derivative Dg(φ(x))Dg(\varphi(x)) of gg at the point φ(x)\varphi(x), then ff is a subimmersion at aa.

If KK is of characteristic zero (resp. K=RK = \mathbf{R} or C\mathbf{C}) and if YY is finite-dimensional (resp. rgaf<+∞\mathrm{rg}_a f < +\infty), then ff is a subimmersion at aa if and only if rgxf=rgaf\mathrm{rg}_x f = \mathrm{rg}_a f for every xx sufficiently near aa.

If KK is of characteristic zero (resp. K=RK = \mathbf{R} or C\mathbf{C}) and YY (resp. XX) is finite-dimensional, the set of points x∈Xx \in X where ff is a subimmersion is an open dense set in XX.

5.10.7. ("Canonical factorization of a subimmersion") Every subimmersion is the composite of a submersion and an immersion. More precisely, let f:X→Yf : X \to Y be a subimmersion, and let JJ be the manifold of germs of submanifolds of YY (5.8.12). Let xx be in XX and y=f(x)y = f(x); there exist open neighbourhoods UU of xx such that f∣Uf|U is a submersion of UU onto a submanifold ZZ of YY; the element γy(Z)\gamma_y(Z) of JJ depends only on xx and not on the neighbourhood UU chosen, and, if it is denoted by λ(x)\lambda(x), the mapping λ\lambda is a submersion of XX into JJ; if ρ\rho denotes the canonical immersion of JJ into YY, one has f=ρ∘λf = \rho \circ \lambda. If ff is an immersion, the morphism λ\lambda of XX into JJ is étale.

If gg is a surjective submersion of XX onto a manifold ZZ and hh a morphism of ZZ into YY such that f=h∘gf = h \circ g, there exists a unique submersion μ\mu of ZZ into JJ such that λ=μ∘g\lambda = \mu \circ g.

5.11. Fibered products and inverse images

5.11.1. Let FF be a Banach space, (Ei)i∈I(E_i)_{i \in I} a finite family of Banach spaces, and f=(fi)i∈If = (f_i)_{i \in I} a family of continuous linear mappings fif_i of EiE_i into FF. Let EE be the product of the EiE_i and ff the continuous linear mapping of EE into FIF^I product of the fif_i. Finally, let DD be the closed subspace of FIF^I consisting of the (yi)i∈I(y_i)_{i \in I} such that yiy_i is independent of ii (the “diagonal” of FIF^I). One says that the family ff is transversal if the continuous linear mapping obtained by composing ff with the canonical projection of FIF^I onto the quotient FI/DF^I/D is surjective and if its kernel f−1(D)f^{-1}(D) admits a topological complement.

If the spaces EiE_i and FF are all finite-dimensional, the family ff is transversal if and only if one has:

codim⁡(⋂iIm⁡fi)=∑icodim⁡(Im⁡fi) \operatorname{codim} \left( \bigcap_i \operatorname{Im} f_i \right) = \sum_i \operatorname{codim} (\operatorname{Im} f_i)

If I={1,2}I = \{1, 2\}, the pair (f1,f2)(f_1, f_2) is transversal if and only if the mapping

f1+f2:E1⊕E2→F f_1 + f_2 : E_1 \oplus E_2 \to F

is surjective and if its kernel admits a topological complement (if E1E_1 and E2E_2 are finite-dimensional, it comes to the same thing to say that

Im⁡f1+Im⁡f2=F). \operatorname{Im} f_1 + \operatorname{Im} f_2 = F).

5.11.2. Let SS be a manifold, (Xi)i∈I(X_i)_{i \in I} a finite family of manifolds, and f=(fi)i∈If = (f_i)_{i \in I} a family of morphisms fif_i of XiX_i into SS. Let PP be the subset of the product XX of the XiX_i consisting of the points (xi)i∈I(x_i)_{i \in I} such that fi(xi)f_i(x_i) is independent of ii. Let x∈Px \in P and let y=fi(xi)∈Fy = f_i(x_i) \in F. One says that the family ff is transversal at the point xx of PP if the mappings Tx(fi)T_x(f_i) form a transversal family of continuous linear mappings with values in the Banach space Ty(S)T_y(S). One says that ff is transversal if it is transversal at every point of PP.

If ff is transversal, then PP is a submanifold of XX, called the fibred product of the family of the XiX_i over SS (relative to the family ff), and denoted by ∏i∈IXi\prod_{i \in I} X_i (or more simply X1×SX2X_1 \times_S X_2 when I={1,2}I = \{1, 2\}, for example). For every point x=(xi)x = (x_i) of PP, the tangent space Tx(∏i∈IXi)T_x(\prod_{i \in I} X_i) is the subspace of ∏Txi(Xi)\prod T_{x_i}(X_i) formed by the tangent vectors t=(ti)t = (t_i) such that Txi(fi)⋅tiT_{x_i}(f_i) \cdot t_i is independent of the index ii.

If f1:X1→Yf_1 : X_1 \to Y is a submersion and f2:X2→Yf_2 : X_2 \to Y a morphism, the pair (f1,f2)(f_1, f_2) is transversal.

5.11.3. (Universal property of fibred products) Let f=(fi)i∈If = (f_i)_{i \in I} be a transversal family of morphisms fi:Xi→Sf_i : X_i \to S, and let PP be the fibred product of the XiX_i over SS; for every i∈Ii \in I, let πi\pi_i denote the morphism from PP into XiX_i obtained by restriction to PP of the projection of XX onto XiX_i; then fi∘πif_i \circ \pi_i is a morphism from PP into SS independent of ii. Let TT be a manifold, and let gi:T→Xig_i : T \to X_i be morphisms of manifolds such that fi∘gif_i \circ g_i is a morphism from TT into SS independent of i∈Ii \in I; then there exists one and only one morphism hh from TT into PP such that gi=πi∘hg_i = \pi_i \circ h for every i∈Ii \in I.

5.11.4. (Associativity of the fibred product) Let f=(fi)i∈If = (f_i)_{i \in I} be a finite family of morphisms of manifolds fi:Xi→Sf_i : X_i \to S, and let (Jλ)λ∈Λ(J_\lambda)_{\lambda \in \Lambda} be a partition of II. Suppose that, for every λ\lambda in Λ\Lambda, the family fλ=(fi)i∈Jλf_\lambda = (f_i)_{i \in J_\lambda} is transversal, and let PλP_\lambda denote the fibred product of this family; for every point x=(xi)i∈Jλx = (x_i)_{i \in J_\lambda} of PλP_\lambda, the element fi(xi)f_i(x_i) of SS is independent of i∈Jλi \in J_\lambda and will be denoted by uλ(x)u_\lambda(x);

then uλu_\lambda is a morphism of PλP_\lambda into SS. For the family u=(uλ)u = (u_\lambda) to be transverse, it is necessary and sufficient that the family ff be so. The canonical bijection of ∏λ∈Λ∏i∈JλXi\prod_{\lambda \in \Lambda} \prod_{i \in J_\lambda} X_i onto ∏i∈IXi\prod_{i \in I} X_i then yields by restriction an isomorphism of the fibred product ∏λ∈Λ∏sPλ\prod_{\lambda \in \Lambda} \prod_{s} P_\lambda onto the fibred product ∏i∈I∏sXi\prod_{i \in I} \prod_{s} X_i.

5.11.5. Let SS be a variety. A variety over SS is a variety XX endowed with a morphism λ:X→S\lambda : X \to S. Let (X,λ)(X, \lambda) be a variety over SS and let f:S′→Sf : S' \to S be a morphism of varieties such that the pair (f,λ)(f, \lambda) is transverse. We then denote by f∗(X)f^*(X) the variety S′×SXS' \times_S X, endowed with the morphism f∗(λ):S′×SXf^*(\lambda) : S' \times_S X defined by f∗(λ)(s′,x)=s′f^*(\lambda)(s', x) = s'. One says that f∗(X)f^*(X) is deduced from XX by base change from SS to S′S' along ff. If λ\lambda is a submersion (resp. an immersion, a subimmersion, an étale morphism), the same is true of f∗(λ)f^*(\lambda).

5.11.6. Let f:X→Yf : X \to Y be a morphism of manifolds and let WW be a submanifold of YY; let ii be the canonical injection of WW into YY. One says that ff is transversal to WW at a point x∈f−1(W)x \in f^{-1}(W) if the pair (f,i)(f, i) is transversal at the point (x,f(x))(x, f(x)) of X×WX \times W. For this, it is necessary and sufficient that the following conditions be satisfied: (i) the tangent space Tf(x)(Y)T_{f(x)}(Y) is the sum of Tf(x)(W)T_{f(x)}(W) and of the image of Tx(f)T_x(f); (ii) the inverse image Tx(f)−1(Tf(x)(W))T_x(f)^{-1}(T_{f(x)}(W)) of the tangent space to WW at f(x)f(x) admits a topological complement. One says that ff is transversal to WW if it is transversal to WW at every point of f−1(W)f^{-1}(W). For a morphism ff of XX into YY to be a submersion, it is sufficient that it be transversal to every point of YY, and it is necessary that it be transversal to every submanifold of YY. For a finite family of morphisms fi:Xi→Sf_i : X_i \to S to be transversal, it is necessary and sufficient that the morphism gg of ∏i∈IXi\prod_{i \in I} X_i into SIS^I defined by g((xi)i∈I)=(fi(xi))i∈Ig((x_i)_{i \in I}) = (f_i(x_i))_{i \in I} be transversal to the diagonal of SIS^I.

5.11.7. Suppose that the morphism f:X→Yf : X \to Y is transversal to the submanifold WW of YY. Then f−1(W)f^{-1}(W) is a submanifold of XX, and for xx in f−1(W)f^{-1}(W) the subspace of Tx(X)T_x(X) tangent to f−1(W)f^{-1}(W) is the inverse image under Tx(f)T_x(f) of the subspace Tf(x)(W)T_{f(x)}(W). By passing to the quotient, the linear mapping Tx(f)T_x(f) defines an isomorphism of topological vector spaces from the transversal space to f−1(W)f^{-1}(W) at xx onto the transversal space to WW at y=f(x)y = f(x). If WW is of codimension dd at yy in YY, the submanifold f−1(W)f^{-1}(W) of XX is of codimension dd at every point of f−1(W)f^{-1}(W). Finally, the mapping x↦(x,f(x))x \mapsto (x, f(x)) is an isomorphism of manifolds of f−1(W)f^{-1}(W) onto the fibre product X×YWX \times_Y W.

5.11.8. Let Y1Y_1 and Y2Y_2 be two subvarieties of a variety XX, and let ιj\iota_j be the injection of YjY_j into XX. We say that Y1Y_1 and Y2Y_2 are transverse if the pair (ι1,ι2)(\iota_1, \iota_2) is transverse; equivalently, it is enough to suppose that, for every point xx of Y1∩Y2Y_1 \cap Y_2, the subspaces Tx(Y1)T_x(Y_1) and Tx(Y2)T_x(Y_2) of Tx(X)T_x(X) have Tx(X)T_x(X) as their sum, and that their intersection admits a topological complement in Tx(X)T_x(X). Under these conditions, Y1∩Y2Y_1 \cap Y_2 is a subvariety of XX and for every xx in Y1∩Y2Y_1 \cap Y_2, we have:

Tx(Y1∩Y2)=Tx(Y1)∩Tx(Y2); T_x(Y_1 \cap Y_2) = T_x(Y_1) \cap T_x(Y_2);

if, moreover, YiY_i is of codimension did_i at xx, then Y1∩Y2Y_1 \cap Y_2 is of codimension d1+d2d_1 + d_2 at xx.

5.11.9. Let ff and gg be two morphisms of a variety XX into a variety YY. If the morphism (f,g):X→Y×Y(f, g) : X \to Y \times Y is transverse to the diagonal of Y×YY \times Y, the subset NN of XX formed by the points xx such that f(x)=g(x)f(x) = g(x) is a subvariety of XX; we call it the kernel of the double arrow

f,g:X↠Y. f, g : X \twoheadrightarrow Y.

5.12. Group varieties

5.12.1. Let GG be a group. A variety structure on GG is said to be compatible with the group structure of GG if the mapping (x,y)↦xy(x, y) \mapsto xy of G×GG \times G into GG is a morphism. The mapping x↦x−1x \mapsto x^{-1} is then a morphism of GG into itself. The set GG, endowed with its group structure and its variety structure, is called a group variety (« of class CrC^r » if one wishes to be precise), or also a Lie group. If r=ωr = \omega, it is also called an analytic group. If K=RK = \mathbf{R} (resp. C,Qp\mathbf{C}, \mathbf{Q}_p), it is also called a real Lie group (resp. complex, p-adic). Every group variety is purely. We call homomorphism of group varieties (or simply homomorphism) any mapping of one group variety into another which is both a group homomorphism and a morphism of varieties.

If GG is a group variety, the topological structure underlying the variety structure of GG makes it a metrizable and complete topological group; it is locally compact if KK is locally compact and GG finite-dimensional.

5.12.2. Examples : (i) If VV is a Banach space, the canonical manifold structure of VV is compatible with its structure of a commutative group. (ii) Let AA be a complete normed KK-algebra, possessing a unit element, and let A∗A^* be the group of invertible elements of AA. This is an open subspace of AA and the manifold structure induced on this open set by the canonical manifold structure of the KK-vector space AA is compatible with the group structure in A∗A^*. In particular, let us take for AA the algebra

L(E) of continuous endomorphisms of a Banach space E; the group A∗A^* consists of the automorphisms of the topological vector space E; we denote by GL(E)\mathrm{GL}(E) the group manifold thus resulting. When A=Mn(K)A = M_n(K), we obtain a group manifold structure on GL(n,K)\mathrm{GL}(n, K).

(iii) If G1,…,GnG_1, \ldots, G_n are group manifolds, the product group G=G1×⋯×GnG = G_1 \times \cdots \times G_n is a group manifold, when it is endowed with the product manifold structure of those of the GiG_i.

5.12.3. Let GG be a group manifold, let HH be a topological group and let f:H→Gf : H \to G be a continuous homomorphism satisfying condition (QR) of No. 5.8.1. The inverse image manifold structure of that of GG by ff then makes HH a group manifold. This applies in particular when HH is a subgroup of GG which is a submanifold (resp. quasi-submanifold); such a subgroup is called a submanifold (resp. quasi-submanifold) of the group GG; it is necessarily closed in GG.

If Hi(i=1,…,n)H_i (i = 1, \ldots, n) is a submanifold of the group GiG_i, then H1×⋯×HnH_1 \times \cdots \times H_n is a submanifold of the group G1×⋯×GnG_1 \times \cdots \times G_n.

5.12.4. Let GG be a group manifold and let HH be a submanifold of the group GG. The equivalence relation x−1y∈Hx^{-1} y \in H is regular, which makes it possible to endow the space G/HG/H of the left cosets xHxH with a manifold structure called the quotient of that of GG. The canonical mapping (g,x)↦g⋅x(g, x) \mapsto g \cdot x of G×(G/H)G \times (G/H) into G/HG/H is a morphism. There are analogous results for the homogeneous space H\GH\backslash G of the right cosets HxHx. If HH is normal, the manifold structure of G/HG/H is compatible with its group structure.

5.12.5. Let GG be a group manifold and let XX be a manifold. A left operation law of the group GG on the manifold XX is any morphism (s,x)↦sx(s, x) \mapsto sx of G×XG \times X into XX such that

s(tx)=(st)xifs,t∈G,x∈Xex=xifx∈X (e being the identity element of G). s(tx) = (st)x \quad \text{if} \quad s, t \in G, x \in X \\ ex = x \quad \text{if} \quad x \in X \ (\text{$ e $ being the identity element of } G).

The group manifold GG is also said to operate on the left on XX; right operation laws are defined analogously.

Let x∈Xx \in X and let GxG_x be its stabilizer in GG. Suppose that the mapping g↦g⋅xg \mapsto g \cdot x is a subimmersion (an assumption automatically satisfied if the characteristic of KK is 0 and XX is finite-dimensional). Then GxG_x is a subgroup submanifold of GG and the mapping of G/GxG/G_x into XX obtained by passing to the quotient from g↦g⋅xg \mapsto g \cdot x is an immersion. If, moreover, the orbit G⋅xG \cdot x of xx is locally closed and if the topology of GG has a countable basis, G⋅xG \cdot x is a submanifold of XX and the mapping G/Gx→G⋅xG/G_x \to G \cdot x is an isomorphism of manifolds.

5.13. Weakening of structure

Throughout this No., the letters r,s,r′,s′r, s, r', s' denote either integers ⩾1\geqslant 1, or one of the symbols ∞\infty and ω\omega. We suppose that K=RK = \mathbf{R}.

5.13.1. Let r⩽sr \leqslant s, and let XX be a manifold of class CsC^s. There exists on the topological space XX a structure of a manifold of class CrC^r and only one such that every chart of XX for the given structure is a chart of XX for this new structure. Let XrX_r be the manifold of class CrC^r thus defined. We say that it is obtained from XX by weakening the manifold structure of XX, or also that its manifold structure underlies that of XX. The notion of weakening is transitive: if r′⩽rr' \leqslant r, we have Xr′=(Xr)r′X_{r'} = (X_r)_{r'}; it commutes with products: if YY is of class CsC^s, we have (X×Y)r=Xr×Yr(X \times Y)_r = X_r \times Y_r; there is an analogous result for X×sYX \times_s Y under the hypotheses of No. 5.11.2.

Let a∈Xa \in X and let cc be a chart of XX at aa; it is also a chart of XrX_r at aa. From (5.5.1) we deduce isomorphisms

θc:E→Ta(X),θc′:E→Ta(Xr), \theta_c : E \to T_a(X), \quad \theta'_c : E \to T_a(X_r),

whence an isomorphism θc′∘θc−1:Ta(X)→Ta(Xr)\theta'_c \circ \theta_c^{-1} : T_a(X) \to T_a(X_r). This isomorphism is independent of the choice of cc; it permits us to identify the tangent spaces to XX and to XrX_r at aa.

5.13.2. Let XX (resp. X′X') be a manifold of class CsC^s (resp. Cs′C^{s'}), and let rr be such that r⩽inf⁡(s,s′)r \leqslant \inf(s, s'). A mapping f:X→X′f : X \to X' is said to be of class CrC^r if it is a morphism from XrX_r into Xr′X'_r; such a mapping is of class Cr′C^{r'} for every r′⩽rr' \leqslant r. Moreover, the tangent linear mapping to f:Xr→Xr′f : X_r \to X'_r at a point a∈Xa \in X coincides with the tangent linear mapping to ff considered as a morphism from XrX_r into Xr′X'_r.

Most often, the same symbol will be used to denote the manifolds XX and XrX_r; thus, if XX is of class CsC^s, the expression "a submanifold of XX of class CrC^r" (r⩽sr \leqslant s) means "a submanifold of XrX_r".

5.14. Restriction of the base field

In this number, two complete non-discrete valued commutative fields KK and LL are given, together with an isomorphism σ\sigma of the valued field KK onto a subfield of LL. If EE is a Banach space over LL, σ∗(E)\sigma_*(E) denotes the vector space over KK obtained by restriction of scalars (cf. Alg., chap. II, 3rd ed., § 8); the topology given on EE is compatible with the structure of KK-vector space, and σ∗(E)\sigma_*(E) is a Banach space over KK.

5.14.1. Let XX be an analytic manifold over LL and let c=(U,φ,E)c = (U, \varphi, E) be a chart of XX. The mapping φ\varphi is a bijection of UU onto an open subset of σ∗(E)\sigma_*(E) and the triple cσ=(U,φ,σ∗(E))c_\sigma = (U, \varphi, \sigma_*(E)) is a chart of XX. There exists on XX one and only one structure of analytic manifold over K for which cσc_\sigma is a chart for every chart cc of the L-analytic manifold X. The analytic manifold over K thus obtained is denoted by XσX_\sigma, and it is said that XσX_\sigma is obtained from X by restriction of scalars (from L to K by means of σ\sigma). The underlying topological space of XσX_\sigma is the same as that underlying X.

5.14.2. Examples: (a) Take K=R, L=C, σK = \mathbf{R},\ L = \mathbf{C},\ \sigma to be the canonical injection of R into C. Thus every complex analytic manifold is canonically endowed with a structure of a real analytic manifold; this real analytic structure itself defines differential structures of class CrC^r for every r. (b) Take K=C, L=C, σK = \mathbf{C},\ L = \mathbf{C},\ \sigma to be the conjugation x↦xˉx \mapsto \bar{x}. Thus to every complex analytic manifold X one associates a complex analytic manifold X‾\overline{X}, called the conjugate of X. If ff is a complex-valued function, defined on an open set U of X, ff is analytic for the structure of X‾\overline{X} if and only if the conjugate function fˉ\bar{f} is analytic for the structure of X. The manifolds X and X‾\overline{X} define by restriction of scalars the same real analytic manifold.

5.14.3. Let X be an analytic manifold over K, and Y an analytic manifold over L. A mapping f:X→Yf : X \to Y is said to be σ\sigma-analytic if it is a morphism of X into YσY_\sigma. If K⊂L, σK \subset L,\ \sigma is the injection of K into L, and X is an analytic manifold over L, a K-analytic mapping is called a σ\sigma-analytic mapping of XσX_\sigma into Y.

5.14.4. Let V be an L-Banach space. We have Vσ=σ∗(V)V_\sigma = \sigma_*(V): the canonical structure of an analytic manifold over K on σ∗(V)\sigma_*(V) is obtained by restriction of scalars from the canonical structure of an analytic manifold over L on V.

5.14.5. Let X be an analytic manifold over L, and let a∈Xa \in X. Let c be a chart of X at a; then cσc_\sigma is a chart of XσX_\sigma at a and we deduce from it the isomorphisms

θc:E→Ta(X),θcσ:σ∗(E)→Ta(Xσ) \theta_c : E \to T_a(X), \quad \theta_{c_\sigma} : \sigma_*(E) \to T_a(X_\sigma)
whence an isomorphism θcσ∘σ∗(θc)−1\theta_{c_\sigma} \circ \sigma_*(\theta_c)^{-1} of σ∗(Ta(X))\sigma_*(T_a(X)) onto Ta(Xσ)T_a(X_\sigma); this isomorphism does not depend on the choice of c; it allows us to identify Ta(Xσ)T_a(X_\sigma) with σ∗(Ta(X))\sigma_*(T_a(X)) and even with Ta(X)T_a(X) by abuse of notation. If ff is an analytic mapping over L of X into a manifold Y, the tangent linear mapping to ff at a (ff being regarded as a morphism of XσX_\sigma into YσY_\sigma) is equal to σ∗(Ta(f))\sigma_*(T_a(f)).

5.14.6. Let X and Y be two analytic manifolds over L, and let f:Xσ→Yf : X_\sigma \to Y be a σ\sigma-analytic mapping. Suppose that the characteristic of K is 0. Then, for ff to be analytic over L, it is necessary and sufficient that, for every a∈Xa \in X, the mapping Ta(f)T_a(f) be L-linear.

When K=R, L=CK = \mathbf{R},\ L = \mathbf{C} (case (a) of No. 5.14.2), we have a more precise result: if XX and YY are finite-dimensional and if f:X→Yf : X \to Y is a mapping of class C1C^1 whose tangent mapping at every point of xx is C\mathbf{C}-linear, then ff is complex analytic.

5.14.7. Let XX be a complex analytic manifold and gg a mapping of XX into itself satisfying the conditions: (i) We have g∘g=IdXg \circ g = \mathrm{Id}_X. (ii) The mapping gg is an isomorphism of the analytic manifold XX onto the conjugate manifold X‾\overline{X} (5.14.2).

The set X0X_0 of points xx of XX such that g(x)=xg(x) = x is a closed analytic submanifold of the underlying real analytic manifold of XX. For x∈X0x \in X_0, one has Tx(X)=Tx(X0)⊕iTx(X0)T_x(X) = T_x(X_0) \oplus i T_x(X_0).

Let UU be a connected open set of XX and let ff and gg be two complex analytic mappings of UU into a separated locally convex complex space or into a separated complex analytic manifold. If ff and gg coincide on a nonempty subset of U∩X0U \cap X_0, open in X0X_0, then f=gf = g.

Assume X0X_0 paracompact. If ff is a real analytic mapping of X0X_0 into a separated locally convex space or into a separated complex analytic manifold, there exists an open neighbourhood UU of X0X_0 in XX and a complex analytic mapping of UU into the value space of ff, extending ff. Two such extensions coincide on a neighbourhood of X0X_0 in XX.

Assume XX finite-dimensional. For every point a∈X0a \in X_0, there exists a system of (complex) coordinates ζ1,…,ζn\zeta^1, \ldots, \zeta^n in an open neighbourhood UU of aa, such that ζi∘g=ζˉi\zeta^i \circ g = \bar{\zeta}^i for 1≤i≤n1 \leq i \leq n; the restriction ζi\zeta^i of ζi\zeta^i to U∩X0U \cap X_0 is then real-valued and (ξ1,…,ξn)(\xi^1, \ldots, \xi^n) is a system of coordinates at aa of the real analytic manifold X0X_0.

5.14.8. For every real analytic manifold YY, paracompact, there exists a pair (X,g)(X, g) consisting of a complex analytic manifold XX and a mapping gg of XX into itself satisfying conditions (i) and (ii) of 5.14.7, and an isomorphism ff of YY onto X0X_0. One says that XX (equipped with ff and gg) is a complexification of YY.

Machine translation of the English, by gpt-5-6-mini, gpt-5.4, gpt-5-6, against glossary version 34. Not checked by a person. That model is a cut down version of the one the rest of this language was written by.