Bourbaki

§ 9. Differential equations and foliations

In nos. os^{\mathrm{os}} 9.1, 9.2 and 9.3, K is assumed to be of characteristic 0. In no. ∘^\circ 9.4, K is assumed to be of characteristic p≠0p \ne 0.

9.1. Integral curves

In this no. ∘^\circ, X denotes a variety of class CrC^r and ξ\xi a vector field of class CsC^s on X, with rr, ss in NK\mathbf{N}_K, s≤r−1s \leq r - 1.

9.1.1. Let I be an open subset of K, and let f:I→Xf : I \to X be a mapping of class CkC^k (k∈NKk \in \mathbf{N}_K, k≤rk \leq r) from I into X. For t∈It \in I, we denote by f′(t)f'(t) the vector Tt(f)(1)∈Tf(t)(X)T_t(f)(1) \in T_{f(t)}(X); this vector is sometimes called the velocity of ff at the instant tt. When X is a Banach space, f′(t)f'(t) is the derivative of ff at tt.

One says that ff is an integral curve of ξ\xi if one has

(1)f′(t)=ξ(f(t))(1)\qquad f'(t)=\xi(f(t))

for all t∈It \in I. Such a curve is of class Cs+1C^{s+1}.

If xx is a point of X, one calls integral curve of ξ\xi with origin xx an integral curve f:I→Xf : I \to X of ξ\xi such that 0∈I0 \in I and f(0)=xf(0)=x; for every x∈Xx \in X, there exists an integral curve of ξ\xi with origin xx, defined in a suitable open neighborhood of 0 in K.

If f:I→Xf : I \to X is an integral curve of ξ\xi and if a∈Ka \in K, the mapping t↦f(t−a)t \mapsto f(t-a) from I+aI+a into X is an integral curve of ξ\xi.

Let f1:I1→Xf_1 : I_1 \to X and f2:I2→Xf_2 : I_2 \to X be two integral curves of ξ\xi, and t∈I1∩I2t \in I_1 \cap I_2. If f1(t)=f2(t)f_1(t)=f_2(t), the mappings f1f_1 and f2f_2 coincide in a neighborhood of tt.

9.1.2 (« Flows »). Let W be an open set of X×KX \times K. For every x∈Xx \in X, denote by WxW_x the set of t∈Kt \in K such that (x,t)∈W(x,t) \in W. An integral flow of ξ\xi with domain W is called a mapping ff of class CkC^k (k≤rk \leq r) from W into X such that, for every x∈Xx \in X, the mapping fx:Wx→Xf_x : W_x \to X defined by fx(t)=f(x,t)f_x(t)=f(x,t) is an integral curve of ξ\xi, and such that for (x,0)∈W(x,0) \in W one has f(x,0)=xf(x,0)=x.

For every point x∈Xx \in X, there exists an integral flow of ξ\xi of class CsC^s whose domain is a neighbourhood of (x,0)(x,0) in X×KX \times K. Two integral flows of ξ\xi defined in a neighbourhood of (x,0)(x,0) coincide on a neighbourhood of (x,0)(x,0).

Let x∈Xx \in X and let ff be an integral flow of ξ\xi whose domain W is a neighbourhood of (x, 0). There exists a neighbourhood V of (x, 0, 0) in X × K × K such that, if (x, t_1, t_2) ∈ V, one has

(x,t1)∈W,(f(x,t1),t2)∈W,(x,t1+t2)∈W(x, t_1) \in W,\quad (f(x, t_1), t_2) \in W,\quad (x, t_1 + t_2) \in W
and (2)
f(f(x,t1),t2)=f(x,t1+t2).f(f(x, t_1), t_2) = f(x, t_1 + t_2).
If X is paracompact, or if X is separated and K = R or C, there exists an integral flow of ξ\xi whose domain is a neighbourhood of X × {0}.

9.1.3 (« The real case »). We suppose that K = R and that X is separated. An integral arc of ξ\xi is any integral curve of ξ\xi whose domain is an open interval of R. Two integral arcs f1f_1 and f2f_2 of ξ\xi, defined on intervals I1I_1 and I2I_2, and coinciding at a point of I1∩I2I_1 \cap I_2, coincide in I1∩I2I_1 \cap I_2.

For every point x∈Xx \in X, there exists one and only one integral arc with origin x, fx:Ix→Xf_x : I_x \to X, such that, for every integral arc f:I→Xf : I \to X with origin x, one has I⊂IxI \subset I_x and f=fx∣If = f_x|I. The arc fxf_x is called the maximal integral arc of ξ\xi with origin x, and its domain IxI_x is sometimes called the interval of life of the point x for the field ξ\xi.

If Ix=]−a−,a+[I_x = ] -a_-, a_+ [ and if a+a_+ is finite, the restriction of fxf_x to the interval [0,a+[[0, a_+ [ is a proper mapping of [0,a+[[0, a_+ [ into X. In particular, f(t)f(t) has no limit when t tends to a+a_+.

9.1.4. With the notations and hypotheses of 9.1.3, the set Ω\Omega of pairs (x,t)∈X×R(x, t) \in X \times \mathbf{R} such that t∈Ixt \in I_x is open in X×RX \times \mathbf{R} and the mapping f:Ω→Xf : \Omega \to X defined by f(x,t)=fx(t)f(x, t) = f_x(t) is of class CsC^s; it is an integral flow of ξ\xi with domain Ω\Omega.

The functions α+\alpha_+ and α−:X→]0,+∞]\alpha_- : X \to ]0, +\infty] defined by Ix=]−α−(x),α+(x)[I_x = ] -\alpha_-(x), \alpha_+(x)[ are lower semi-continuous. For (x,t)∈Ω(x, t) \in \Omega and y=f(x,t)y = f(x, t), one has α+(y)=α+(x)−t\alpha_+(y) = \alpha_+(x) - t and α−(y)=α−(x)+t\alpha_-(y) = \alpha_-(x) + t.

If t1∈Ixt_1 \in I_x and if t2∈If(x,t1)t_2 \in I_{f(x, t_1)}, one has t1+t2∈Ixt_1 + t_2 \in I_x and (3)

f(f(x,t1),t2)=f(x,t1+t2).f(f(x, t_1), t_2) = f(x, t_1 + t_2).

If the function α+\alpha_+ (resp. α−\alpha_-) is bounded below on X by a constant > 0, it is constant and equal to +∞+\infty.

One has α+=α−=+∞\alpha_+ = \alpha_- = +\infty (in other words, one has Ω=X×R\Omega = X \times \mathbf{R}) in each of the following cases:

(i) the field ξ\xi has compact support (for example if X is compact); (ii) there exists a group of automorphisms of X, transitive, and preserving ξ\xi; (iii) there exists on X a riemannian structure for which X is complete and ξ\xi bounded.

If Ω=X×R\Omega = X \times \mathbf{R}, the mapping f:X×R→Xf : X \times \mathbf{R} \to X is a right operation law of class CsC^s of the group variety R\mathbf{R} in the variety X, cf. 5.12.5.

9.1.5 (« The complex case »). One supposes that K = C and that X is separated. For every a∈]0,+∞]a \in ]0, +\infty], denote by DaD_a the open disk of center 0 and radius a in C. For every x∈Xx \in X, there exists one and only one number ρ(x)∈]0,+∞]\rho(x) \in ]0, +\infty], one and only one integral curve fx:Dρ(x)→Xf_x : D_{\rho(x)} \to X with origin xx such that, for every a∈]0,+∞]a \in ]0, +\infty] and every integral curve f:Da→Xf : D_a \to X with origin xx, one has a≤ρ(x)a \leq \rho(x) and f=fx∣Daf = f_x|D_a.

For every λ∈C\lambda \in \mathbf{C}, let αλ(x)\alpha_\lambda(x) be the supremum of the interval of existence of xx for the vector field λξ\lambda \xi on the real variety deduced from XX by restriction of scalars. One has ρ(x)=inf⁡∣λ∣=1αλ(x)\rho(x) = \inf_{|\lambda|=1} \alpha_\lambda(x).

The set Δ\Delta of pairs (x,t)∈X×C(x, t) \in X \times \mathbf{C} such that t∈Dρ(x)t \in D_{\rho(x)} is open in X×CX \times \mathbf{C} and the mapping ff from Δ\Delta into XX defined by f(x,t)=fx(t)f(x, t) = f_x(t) is an integral flow of ξ\xi. The function ρ:X→]0,+∞]\rho : X \to ]0, +\infty] is lower semi-continuous. For (x,t)∈Δ(x, t) \in \Delta and y=f(x,t)y = f(x, t), one has ρ(y)≥ρ(x)−∣t∣\rho(y) \geq \rho(x) - |t|. If ρ\rho is bounded below on XX by a constant >0> 0, one has ρ=+∞\rho = +\infty and Δ=X×C\Delta = X \times \mathbf{C}.

When Δ=X×C\Delta = X \times \mathbf{C}, the mapping f:X×C→Xf : X \times \mathbf{C} \to X is a right operation law of the group variety C\mathbf{C} on the variety XX.

9.1.6. Let φ:X→Y\varphi : X \to Y be a morphism of varieties and η\eta a vector field on YY such that ξ\xi is φ\varphi-related to η\eta (cf. 8.2.6). If f:I→Xf : I \to X is an integral curve of ξ\xi, the mapping φ∘f:I→Y\varphi \circ f : I \to Y is an integral curve of η\eta. If K=RK = \mathbf{R} and if XX and YY are separated, for every x∈Xx \in X, the interval of existence of xx for ξ\xi is contained in the interval of existence of φ(x)\varphi(x) for η\eta. If, moreover, φ\varphi is proper, these intervals are equal.

9.1.7 (“Time-dependent equations”). Let WW be an open subset of X×KX \times K and let Ξ:W→T(X)\Xi : W \to T(X) be a CsC^s lifting of pr1:W→X\mathrm{pr}_1 : W \to X, that is, a mapping of class CsC^s from WW into T(X)T(X) such that Ξ(x,t)\Xi(x, t) belongs to Tx(X)T_x(X) for every (x,t)∈W(x, t) \in W. An integral curve of Ξ\Xi is a mapping ff of class CkC^k (with k∈NKk \in \mathbf{N}_K and k≤rk \leq r), from an open subset II of KK into XX such that, for every t∈It \in I, one has

(f(t),t)∈W and f′(t)=Ξ(f(t),t). (f(t), t) \in W \text{ and } f'(t) = \Xi(f(t), t).

Let η\eta be the vector field of class CsC^s on WW defined by η(x,t)=(Ξ(x,t),(t,1))\eta(x, t) = (\Xi(x, t), (t, 1)). For a mapping ff from an open subset II of KK into XX to be an integral curve of Ξ\Xi, it is necessary and sufficient that t↦(f(t),t)t \mapsto (f(t), t) be an integral curve of η\eta in the sense of 9.1.1.

Let g:I→Wg : I \to W be an integral curve of η\eta and t0t_0 a point of II such that pr2(g(t0))=t0\mathrm{pr}_2(g(t_0)) = t_0. The set of elements t∈It \in I such that pr2(g(t))=t\mathrm{pr}_2(g(t)) = t is a neighbourhood of t0t_0 which coincides with II if II is connected.

9.1.8 (“Parameters and initial conditions”). Let ZZ be a variety of class CrC^r, let WW be an open subset of X×K×ZX \times K \times Z, and let Ξ\Xi be a CsC^s lifting of pr1:W→X\mathrm{pr}_1 : W \to X (8.6.1). If z∈Zz \in Z, we denote by WzW_z the set of (x,t)∈X×K(x, t) \in X \times K such that (x,t,z)∈W(x, t, z) \in W, and we denote by Ξz\Xi_z the mapping (x,t)↦Ξ(x,t,z)(x, t) \mapsto \Xi(x, t, z) from WzW_z into T(X)T(X).

Let (x0,t0,z0)(x_0, t_0, z_0) be a point of WW. There then exists an open neighbourhood X1×I1×Z1X_1 \times I_1 \times Z_1 of (x0,t0,z0)(x_0, t_0, z_0) in WW and a mapping of class CsC^s

f:X1×I1×I1×Z1→X f : X_1 \times I_1 \times I_1 \times Z_1 \to X

such that, for every (x1,t1,z1)∈X1×I1×Z1(x_1, t_1, z_1) \in X_1 \times I_1 \times Z_1, the mapping t↦f(x1,t1,t,z1)t \mapsto f(x_1, t_1, t, z_1) is an integral curve of Ξz1\Xi_{z_1} (in the sense of 9.1.7) satisfying the relation f(x1,t1,t1,z1)=x1f(x_1, t_1, t_1, z_1) = x_1.

Two mappings

f1:X1×I1×I1×Z1→Xandf2:X2×I2×I2×Z2→X f_1 : X_1 \times I_1 \times I_1 \times Z_1 \to X \quad \text{and} \quad f_2 : X_2 \times I_2 \times I_2 \times Z_2 \to X

satisfying the above conditions coincide in a neighbourhood of (x0,t0,t0,z0)(x_0, t_0, t_0, z_0).

9.2. Foliations

In this No., X denotes a variety of class CrC^r, with r∈NKr \in \mathbf{N}_K. Unless otherwise stated, all varieties and all morphisms considered are assumed to be of class CrC^r.

9.2.1. Let S be a variety, and let p:X→Sp : X \to S be a submersion. For each s∈Ss \in S, endow Xs=p−1(s)X_s = p^{-1}(s) with the variety structure induced by that of X (5.10.5); the set X is the disjoint union of the XsX_s. Denote by XpX_p the variety structure on X obtained by gluing the XsX_s, for s∈Ss \in S (5.2.4); it is the unique variety structure on X for which the XsX_s are open subvarieties. The topological space XpX_p is the sum of the spaces XsX_s.

Let V be a variety and ff a mapping from V into X. For ff to be a morphism from V into XpX_p, it is necessary and sufficient that ff be a morphism from V into X and that p∘fp \circ f be locally constant.

9.2.2. A foliation of X is called a variety Y having the same underlying set as X and satisfying the following condition:

(F) For every x∈Xx \in X, there exists an open subvariety U of X containing x, a variety S and a submersion p:U→Sp : U \to S such that the variety UpU_p is an open subvariety of Y.

The pair (X, Y) is also called a foliated variety. If (X, Y) and (X', Y') are foliated varieties, a morphism from (X, Y) into (X', Y') is any mapping f:X→X′f : X \to X' which is both a morphism of the variety X into the variety X' and a morphism of the variety Y into the variety Y'.

Let (X, Y) be a foliated variety. The identity mapping Y→XY \to X is a bijective immersion. A subset U of X is called a leaf if it is an open set of Y; in this case, U is endowed with the topological structure and the variety structure induced by those of Y; for example, U is said to be a connected leaf if it is an open and connected subset of Y. The leaves which are subvarieties of X form a basis of the topology of Y. When K=RK = \mathbf{R} or C\mathbf{C}, the connected components of Y are leaves, called maximal connected leaves of (X, Y).

9.2.3. If s∈NK,s≤rs \in \mathbf{N}_K, s \leq r, a foliation of class CsC^s of X is called a foliation of the variety of class CsC^s underlying X.

9.2.4. Examples

a) The variety X is a foliation of X, called the coarse foliation of X.

b) If p:X→Sp : X \to S is a submersion, XpX_p is a foliation of X, called the foliation of X defined by p. The foliation of X defined by the identity mapping is the set X endowed with its purely variety structure of dimension 0 (5.2.1); it is called the discrete foliation of X.

c) Let E be a Banach space, F a closed vector subspace of E having a topological complement, and pp the canonical projection E→E/FE \to E/F. The foliation EpE_p of E is called the foliation of E defined by F.

d) Let Γ\Gamma be a discrete group acting properly and freely on the topological space XX (TG, III, § 4, n° 4). Let YY be a foliation of XX, and suppose that, for every s∈Γs \in \Gamma, the mapping x↦sxx \mapsto sx is an automorphism of (X,Y)(X, Y). The equivalence relation on XX (resp. YY) whose classes are the orbits of Γ\Gamma is regular (5.9.5). If X/ΓX/\Gamma (resp. Y/ΓY/\Gamma) is denoted the corresponding quotient variety, the pair (X/Γ,Y/Γ)(X/\Gamma, Y/\Gamma) is a foliated variety, called the quotient of (X,Y)(X, Y) by Γ\Gamma.

9.2.5. Let YY be a foliation of XX and UU an open subset of XX. Then UU is open in YY and UU, endowed with the structure of variety induced by that of YY, is a foliation of XX, called the foliation induced by YY.

More generally, let f:X′→Xf : X' \to X be a morphism such that ff and the identity mapping Y→XY \to X form a transversal pair (5.11.1). The fibred product Y′=X′×XYY' = X' \times_X Y identifies canonically with a foliation of X′X' called the inverse image foliation of YY by ff.

9.2.6. Let p:X→Sp : X \to S and p′:X′→S′p' : X' \to S' be two submersions, and let ff be a morphism from XX to X′X'. In order that ff be a morphism from XpX_p to Xp′X_{p'}, it is necessary and sufficient that, for every x∈Xx \in X, there exist an open neighbourhood UU of xx in XX and a morphism gg from p(U)p(U) to S′S' such that g∘p=p′∘fg \circ p = p' \circ f on UU.

9.2.7. Let YY be a foliation of XX. A foliated chart of (X,Y)(X, Y) is called a quadruple (U,φ,E,F)(U, \varphi, E, F) such that (U,φ,E)(U, \varphi, E) is a chart of XX, that FF is a closed vector subspace of the Banach space EE admitting a topological complement and that φ\varphi is an isomorphism of the foliation of UU induced by YY onto the foliation of φ(U)\varphi(U) defined by the canonical mapping of EE onto E/FE/F. For every point x∈Xx \in X, there exists a foliated chart (U,φ,E,F)(U, \varphi, E, F) of (X,Y)(X, Y) such that x∈Ux \in U. Let n=dim⁡xXn = \dim_x X and let m=dim⁡xYm = \dim_x Y. If nn is finite, there exists an open neighbourhood UU of xx and a system of coordinates ζ1,…,ζn\zeta^1, \ldots, \zeta^n of XX on UU such that the foliation of UU induced by YY coincides with the foliation defined by the mapping (ζm+1,…,ζn)(\zeta^{m+1}, \ldots, \zeta^n) of UU into Kn−mK^{n-m}.

9.2.8. Let YY be a foliation of XX. When xx runs through XX, the spaces Tx(Y)T_x(Y) are the fibres of a vector subbundle of class Cr−1C^{r-1} of T(X)T(X) which is called the subbundle of T(X)T(X) tangent to the foliation YY and which is denoted by T(X,Y)T(X, Y). If the foliation YY is defined by a submersion p:X→Sp : X \to S, then T(X,Y)=Ker T(p)T(X, Y) = \mathrm{Ker}\, T(p) is the relative tangent bundle T(X/S)T(X/S) of XX over SS (8.1.3).

Let (X,Y)(X, Y) and (X′,Y′)(X', Y') be two foliated manifolds, and let ff be a morphism of XX into X′X'. For ff to be a morphism of foliated manifolds, it is necessary and sufficient that T(f)T(f) map T(X,Y)T(X, Y) into T(X′,Y′)T(X', Y'). In particular, let f:Z→Xf : Z \to X be a morphism of manifolds; for ff to be a morphism of ZZ into YY, it is necessary and sufficient that T(f)T(f) map T(Z)T(Z) into T(X,Y)T(X, Y). For a submanifold ZZ of XX to be a leaf of (X,Y)(X, Y), it is necessary and sufficient that

Tz(Z)=Tz(X,Y)for all z∈Z. T_z(Z) = T_z(X, Y) \quad \text{for all } z \in Z.

Let YY be a foliation of XX and let UU be a leaf of YY, having a countable basis of open sets. Let ZZ be a manifold and ff a mapping of ZZ into UU. For ff to be a morphism of manifolds from Z into U, it is necessary and sufficient that the composite of f and the canonical injection of U into X be a morphism of manifolds from Z into X. If X is locally compact and countable at infinity, every connected leaf of (X, Y) has a countable basis of open sets (cf. TG, I, 3e éd., §11, No. 7, cor. 2 du th. 1).

9.2.9. Suppose that K = R or C. Let Y be a foliation of X. The following conditions are equivalent:

a) There exists a manifold S and a submersion p:X→Sp : X \to S such that Y=XpY = X_p.

b) For every x∈Xx \in X, there exists a submanifold SxS_x of X possessing the following two properties: b₁) The tangent space Tx(Sx)T_x(S_x) to SxS_x at xx is a topological supplement of Tx(X,Y)T_x(X, Y) in Tx(X)T_x(X). b₂) Every connected leaf of (X, Y) meets SxS_x in at most one point.

Suppose these conditions are satisfied and let R{x,y}R \{x, y\} be the relation on X “x and y belong to the same connected leaf”; then R is a regular equivalence relation (5.9.5) on X, and, if p denotes the canonical projection X→X/RX \to X/R, one has Y=XpY = X_p.

Example

Let K=RK = R and X=R2X = R^2. Let the discrete group Γ=Z2\Gamma = Z^2 operate on X by translations. Let m∈Rm \in R and let (X,Ym)(X, Y_m) be the foliation defined by the submersion p:(x1,x2)↦x2−mx1p : (x_1, x_2) \mapsto x_2 - mx_1 of X onto R. Put X′=X/ΓX' = X/\Gamma and Ym′=Ym/ΓY'_m = Y_m/\Gamma; then Ym′Y'_m is a foliation of the torus X′X' (cf. 9.2.4, example d)). If m is rational, there exists a submersion p′p' of X′X' onto R/ZR/Z such that Ym′=Xp′′Y'_m = X'_{p'}. If m is irrational, every maximal connected leaf of (X′,Ym′)(X', Y'_m) is dense in X′X', and there does not exist a submersion p′p' of X′X' into a manifold S such that Ym′=Xp′′Y'_m = X'_{p'}.

9.2.10. Let Y be a foliation of X, and let π:X→S\pi : X \to S be a morphism of manifolds such that the composite Y→X→πSY \to X \xrightarrow{\pi} S is étale. If S’ is an open subset of S, a section σ:S′→X\sigma : S' \to X of π\pi over S’ is said to be horizontal (with respect to the foliation Y) if it is a morphism from S’ into Y; it amounts to saying that σ(S′)\sigma(S') is a leaf of (X, Y), or again that T(σ)T(\sigma) maps T(S′)T(S') into T(X,Y)T(X, Y). For every s0∈Ss_0 \in S and every x0∈π−1(s0)x_0 \in \pi^{-1}(s_0), there exists a horizontal section defined in a neighbourhood of s0s_0 and taking the value x0x_0 at s0s_0; two such sections coincide on a neighbourhood of s0s_0.

More generally, let s0∈Ss_0 \in S, let T be a manifold, let f be a morphism from T into the submanifold π−1(s0)\pi^{-1}(s_0) of X, and let t0∈Tt_0 \in T. There then exists an open neighbourhood T’ (resp. S’) of t0t_0 in T (resp. of s0s_0 in S), and a morphism

F:S′×T′→X F : S' \times T' \to X
satisfying the following condition: for every t∈T′t \in T', the mapping s↦F(s,t)s \mapsto F(s, t) is a horizontal section of π\pi over S’ taking the value f(t)f(t) at the point s0s_0. Two such mappings F coincide on a neighbourhood of (s0,t0)(s_0, t_0).

9.3. Integrable subbundles

In this no., X denotes a manifold of class CrC^r and F a vector subbundle of class CsC^s of T(X)T(X), with r,sr, s in NK,s≤r−1N_K, s \leq r - 1.

9.3.1. Let V be a manifold of class CkC^k (k∈NK,k≤rk \in \mathbf{N}_K, k \leq r) and let ff be a morphism of class CkC^k from V into X. One says that ff is an integral of F if T(f)T(f) maps T(V)T(V) into F.

A submanifold Z of class CkC^k of X is called an integral submanifold of F if the injection Z→XZ \to X is an integral of F in the above sense, i.e. if one has Tx(Z)⊂FxT_x(Z) \subset F_x for all x∈Zx \in Z.

Let x∈Xx \in X. If Z1Z_1 and Z2Z_2 are two integral submanifolds of X containing xx and if Tx(Z1)=FxT_x(Z_1) = F_x, there exists a neighbourhood U of xx such that U∩Z2⊂Z1U \cap Z_2 \subset Z_1. If in addition Tx(Z2)=FxT_x(Z_2) = F_x, the germs of Z1Z_1 and Z2Z_2 at xx coincide.

9.3.2. We say that F is integrable if there exists a foliation Y of X of class CsC^s such that F=T(X,Y)F = T(X, Y) (9.2.8). Such a foliation is then unique; it is called the integral foliation of F. In order that a morphism f:V→Xf : V \to X of class CsC^s be an integral of F, it is necessary and sufficient that ff be a morphism of V into Y.

Example

If F has rank 1 at every point, F is integrable, and defines on X a pure foliation of dimension 1. Suppose in addition that K=RK = \mathbf{R}, that X is separated, and that F has as a frame a vector field ξ\xi everywhere nonzero. Then, for every x∈Xx \in X, the maximal connected leaf containing xx is the image of the maximal integral arc of ξ\xi with origin xx (9.1.3).

9.3.3 (« Integrability Criteria »). The following conditions are equivalent:

(i) The vector subbundle F of T(X)T(X) is integrable.

(ii) For every x∈Xx \in X, there exists an integral submanifold Z of class CsC^s of F such that x∈Zx \in Z and Tx(Z)=FxT_x(Z) = F_x.

(iii) Whatever the open set U of X and the vector fields ξ\xi and η\eta belonging to JFs(U)\mathscr{J}_F^s(U) (7.4.1), one has [ξ,η]∈JFs−1(U)[\xi, \eta] \in \mathscr{J}_F^{s-1}(U).

Let (ξi)i∈I(\xi_i)_{i \in I} be a family of sections of F of class CsC^s such that, for every x∈Xx \in X, the set of the ξi(x)\xi_i(x) is a total subset of the Banach space FxF_x (EVT, I, § 2, No. 1). The conditions (i), (ii), (iii) are then equivalent to:

(iv) for every pair (i,j)(i, j) of elements of I, and every x∈Xx \in X, one has [ξi,ξj](x)∈Fx[\xi_i, \xi_j](x) \in F_x.

When I is finite, and the family (ξi)(\xi_i) is a frame of F, the preceding conditions are also equivalent to:

(v) there exists a family (cijk)(i,j,k)∈I×I×I(c_{ij}^k)_{(i, j, k) \in I \times I \times I} of functions on X with values in K such that [ξi,ξj]=∑kcijkξk[\xi_i, \xi_j] = \sum_k c_{ij}^k \xi_k for all i,ji, j in I.

(If this is the case, the functions ci,jkc_{i,j}^k are of class Cs−1C^{s-1}.)

9.3.4. Let s′∈NKs' \in \mathbf{N}_K, with s′≤ss' \leq s, and let F′F' be the vector bundle of class Cs′C^{s'} deduced from F by weakening of structure (8.7.1). In order that F be integrable, it is necessary and sufficient that F′F' be so. In this case, the integral foliation of F′F' is deduced from that of F by weakening of structure.

9.3.5. Let E be a Banach space and pp an integer ≥0\geq 0. For every x∈Xx \in X, let us denote

N(p, E)_x the vector subspace of Alt^p(T_x(X); E) formed by the elements u such that u(v1,…,vp)=0u(v_1, \ldots, v_p) = 0 for v1,…,vpv_1, \ldots, v_p in FxF_x. The spaces N(p, E)_x, for x∈Xx \in X, are the fibres of a vector subbundle of class C^s of Alt^p(T(X); E). Let us denote it by N(p, E). If F is integrable, we have:

(vi) for every open U of X and every form ω∈SN(p,E)s(U)\omega \in \mathscr{S}_{N(p, E)}^s(U), one has dω∈SN(p+1,E)s−1(U)d\omega \in \mathscr{S}_{N(p+1, E)}^{s-1}(U).

Conversely, if (vi) is satisfied for p=1p = 1 and for every Banach space E, the bundle F is integrable; when K = R or C, it is even enough to verify (vi) for p=1p = 1 and E = K.

Suppose that the dual of T(X)/F admits a frame (ω1,…,ωn)(\omega_1, \ldots, \omega_n). The integrability of F is then equivalent to the following condition:

(vii) dωi∧ω1∧⋯∧ωn=0d\omega_i \wedge \omega_1 \wedge \cdots \wedge \omega_n = 0 for 1≤i≤n1 \leq i \leq n.

If in addition there exists a vector subbundle G of class C^s of T(X) such that T(X) is the direct sum of F and G, condition (vii) is equivalent to:

(viii) there exist differential forms αij\alpha_i^j (i, j in I) of degree 1 on X, of class C^{s-1}, such that dωi=∑jαij∧ωjd\omega_i = \sum_j \alpha_i^j \wedge \omega_j for all i∈Ii \in I.

9.3.6. Let L be a Banach space and let ω\omega be a differential form of degree 1 on X with values in L, of class C^s. We make the following two hypotheses:

a) for every x∈Xx \in X, ωx\omega_x is a surjective homomorphism from Tx(X)T_x(X) into L, and the kernel of ωx\omega_x is FxF_x; b) there exists a vector subbundle G of T(X) such that T(X) is the direct sum of F and G.

The integrability of F is then equivalent to:

(ix) there exists a differential form α\alpha of degree 1 on X with values in End(L) such that dω=α∧ωd\omega = \alpha \wedge \omega, the exterior product being defined by the canonical pairing End(L) ×\times L →\to L (7.8.2 and 8.3.2).

9.3.7 (« Equations in total differentials »). Suppose that X is the product of two manifolds A and B of class C^r; denote by p1:X→Ap_1 : X \to A and p2:X→Bp_2 : X \to B the two projections. Let f be a morphism of class C^s from p1∗T(A)p_1^*T(A) into p2∗T(B)p_2^*T(B). The graphs of the mappings f(a,b):Ta(A)→Tb(B)f_{(a, b)} : T_a(A) \to T_b(B) are the fibres of a vector subbundle of class C^s of T(X); denote it by F^f.

Let A’ be an open subset of A, and let φ:A′→B\varphi : A' \to B be a morphism of class C^k (k∈NK,k≤rk \in \mathbf{N}_K, k \leq r). We say that φ\varphi is an integral of f if, for every a∈A′a \in A', one has Ta(φ)=fa,φ(a)T_a(\varphi) = f_{a, \varphi(a)}; this is equivalent to saying that the mapping a↦(a,φ(a))a \mapsto (a, \varphi(a)) from A’ into X is an integral of F^f (9.3.1). Such a mapping is of class C^{s+1}. If φ1\varphi_1 and φ2\varphi_2 are two integrals of f and take the same value at a point a∈Aa \in A, they coincide in a neighbourhood of a.

More generally, let Z be a manifold of class C^k, let A’ be an open subset of A, let a∈A′a \in A' and let Φ1,Φ2\Phi_1, \Phi_2 be morphisms of class C^k from Z ×\times A’ into B. Suppose that

Φ₁ and Φ₂ coincide on Z × {a} and that, for every z ∈ Z, the morphisms

a↦Φ1(z,a)anda↦Φ2(z,a) a \mapsto \Phi_1(z, a) \quad \text{and} \quad a \mapsto \Phi_2(z, a)
are integrals of f. Then Φ₁ and Φ₂ coincide in a neighbourhood of Z × {a}.

Suppose that F^f is integrable. Let Z be a manifold of class C^k (k ∈ N_K, k ≤ s), let (z₀, a₀) be a point of Z × A, and let ρ be a morphism of class C^k from Z into B. There exists an open neighbourhood Z' × A' of (z₀, a₀) in Z × A and a morphism Φ : Z' × A' → B of class C^k such that, for every z ∈ Z', the mapping a ↦ Φ(z, a) from A' into B has f as its tangent mapping and takes the value ρ(z) at the point a₀.

9.3.8. Retain the hypotheses and notations of 9.3.7 and suppose that A (resp. B) is an open submanifold of a Banach space E (resp. M). The mapping f is then identified with a morphism of class C^s from X = A × B into the Banach space L(E;M)\mathscr{L}(E; M). Denote by D₁f (resp. D₂f) the first (resp. second) partial derivative of f (1.6.2); this is a morphism of class C^{s-1} from X into L(E;L(E;M))\mathscr{L}(E; \mathscr{L}(E; M)) (resp. into L(M;L(E;M))\mathscr{L}(M; \mathscr{L}(E; M))), a space which is identified in the obvious way with L2(E;M)\mathscr{L}_2(E; M) (resp. with L(M,E;M)\mathscr{L}(M, E; M)). For F to be integrable, it is necessary and sufficient that, for every x ∈ X, the bilinear mapping

Δx:(h1,h2)↦D1f(x)(h1,h2)+D2f(x)(f(x)h1,h2) \Delta_x : (h_1, h_2) \mapsto D_1f(x)(h_1, h_2) + D_2f(x)(f(x)h_1, h_2)
from E × E into M be symmetric. Under this condition, if φ is an integral of f defined in an open subset A' of A, the second derivative of φ at a point a of A' is Δ(a,φ(a))\Delta_{(a, \varphi(a))}.

If E = K^n, and if we denote by (x¹, ..., xⁿ) the coordinate functions on K^n, the mapping f is defined by a family (f₁, ..., fₙ) of mappings from A × B into M, and the integrability condition is written:

∂fi∂xj+(D2fi)⋅fj=∂fj∂xi+(D2fj)⋅fi \frac{\partial f_i}{\partial x^j} + (D_2f_i) \cdot f_j = \frac{\partial f_j}{\partial x^i} + (D_2f_j) \cdot f_i
whatever the integers i, j in {1, n}. A mapping φ from an open subset A' of A into B is an integral of f if and only if one has
∂φ∂xi=fi(x,φ(x))pour tout x∈A′ et tout i∈{1,n}, \frac{\partial \varphi}{\partial x^i} = f_i(x, \varphi(x)) \quad \text{pour tout } x \in A' \text{ et tout } i \in \{1, n\},
in other words, if one has
dφ=∑1≤i≤nfi(x,φ(x)) dxi. d\varphi = \sum_{1 \leq i \leq n} f_i(x, \varphi(x)) \, dx^i.

9.4. Integrable bundles in characteristic p ≠ 0

In this no., K is assumed to have characteristic p ≠ 0. Let X be a K-analytic manifold locally of finite dimension.

9.4.1 (« p-th Powers »). Let ξ be a vector field on an open subset U of X. There exists one and only one vector field ξ^p on U such that

Dξp(f)=(Dξ)p(f)=Dξ(Dξ(…(Dξ(f))…))⏟p times D_{\xi^p}(f) = (D_\xi)^p(f) = \underbrace{D_\xi(D_\xi(\ldots(D_\xi(f))\ldots))}_{p \text{ times}}
for every analytic function f defined in an open subset of U.

If φ\varphi is an analytic function on UU, one has

(φξ)p=φpξp+(Dφξ)p−1(φ)⋅ξ (\varphi \xi)^p = \varphi^p \xi^p + (\mathrm{D}_{\varphi \xi})^{p-1}(\varphi) \cdot \xi
and if η\eta is a vector field on UU, one has
[ξp,η]=ad(ξ)p(η)=[ξ,[ξ,…,[ξ,η]…]]. [\xi^p, \eta] = \mathrm{ad}(\xi)^p(\eta) = [\xi, [\xi, \ldots, [\xi, \eta]\ldots]].

9.4.2 (« Jacobson Identity »). Let LL be the free FpF_p-Lie algebra (LIE, II, § 2, No. 2) on a set {x,y}\{x, y\} with two elements. There exists one and only one element Λp(x,y)\Lambda_p(x, y) of LL such that one has

(x+y)p=xp+yp+Λp(x,y) (x + y)^p = x^p + y^p + \Lambda_p(x, y)
in the enveloping algebra of LL. Examples:
Λ2(x,y)=[x,y];Λ3(x,y)=[x,[x,y]]−[y,[x,y]]. \Lambda_2(x, y) = [x, y]; \quad \Lambda_3(x, y) = [x, [x, y]] - [y, [x, y]].
If ξ\xi and η\eta are vector fields on XX, one has
(ξ+η)p=ξp+ηp+Λp(ξ,η). (\xi + \eta)^p = \xi^p + \eta^p + \Lambda_p(\xi, \eta).

9.4.3. Example. — Let us take X=KX = K (resp. K∗K^*) and denote by xx the canonical mapping X→KX \to K. Let ξ\xi (resp. η\eta) be the vector field ∂/∂x\partial / \partial x (resp. x⋅∂/∂xx \cdot \partial / \partial x); it is invariant under the translations of the additive (resp. multiplicative) group XX. One has

ξp=0andηp=η. \xi^p = 0 \quad \text{and} \quad \eta^p = \eta.

9.4.4 (« Integrable bundles »). Let FF be a vector subbundle of T(X)T(X). We say that FF is integrable if, for every x∈Xx \in X, there exists a system of coordinates (ζ1,…,ζn)(\zeta^1, \ldots, \zeta^n) of XX at xx and an integer m≤nm \leq n such that (∂/∂ζ1,…,∂/∂ζm)(\partial / \partial \zeta^1, \ldots, \partial / \partial \zeta^m) is a frame of FF in a neighbourhood of xx. For FF to be integrable, it is necessary and sufficient that it satisfy the following condition: For every open set UU of XX, SFω(U)\mathscr{S}_F^\omega(U) is a Lie subalgebra of ST(X)ω(U)\mathscr{S}_{T(X)}^\omega(U) stable under the operation ξ↦ξp\xi \mapsto \xi^p.

Machine translation of the English, by gpt-5-6, against glossary version 34. Not checked by a person.