Bourbaki

§ 14. Differential Operators

In this paragraph, X denotes a locally finite-dimensional manifold of class CrC^r, where r∈Nkr \in \mathbf{N}_k; E and F denote two vector bundles of finite rank of class CrC^r and with basis X.

All manifolds and vector bundles considered are assumed to be locally finite-dimensional.

14.1. Differential Operators

14.1.1. Let k be an integer such that 0≤k≤r0 \leq k \leq r, and let Pk(E)P^k(E) be the bundle of jets of sections of order k of E (12.6.1). Put

Dk(E,F)=L(Pk(E);F). D^k(E,F)=\mathscr{L}(P^k(E);F).

This is a vector bundle with basis X and of class Cr−kC^{r-k}. A section of this bundle over an open set U of X is identified with a mapping

D:Pk(E)∣U⟶F∣U D:P^k(E)|U\longrightarrow F|U

which commutes with the projection onto U and is linear on each fibre; such a section is called a differential operator on U, of type E→FE \rightarrow F, and of order ≤k\leq k. Let h∈Nk∪{0}h \in \mathbf{N}_k \cup \{0\}, with 0≤h≤r−k0 \leq h \leq r-k. We denote by DUk,h(E,F)\mathscr{D}^{k,h}_U(E,F) the set of differential operators on U, of type E→FE \rightarrow F and of order ≤k\leq k, which are of class ChC^h (as sections of Dk(E,F)D^k(E,F), or as morphisms of Pk(E)∣UP^k(E)|U into F∣UF|U, this amounts to the same thing). If M=Dk(E,F)M=D^k(E,F), we have DUk,h(E,F)=FMh(U)\mathscr{D}^{k,h}_U(E,F)=\mathscr{F}^h_M(U); this is a module over Ch(U)\mathscr{C}^h(U), cf. 7.4.1.

If 0≤k′≤k0 \leq k' \leq k, the morphism rk,k′:Pk(E)⟶Pk′(E)r^{k,k'}:P^k(E)\longrightarrow P^{k'}(E) (cf. 12.6.6) defines an injection of Dk′(E,F)D^{k'}(E,F) into Dk(E,F)D^k(E,F); every differential operator of order ≤k′\leq k' is thus identified with a differential operator of order ≤k\leq k. A differential operator of order ≤k\leq k which is not of order ≤k−1\leq k-1 is sometimes said to be of order k. If h∈Nk∪{0}h \in \mathbf{N}_k \cup \{0\} and h≤r−kh \leq r-k, and if U is an open set of X, we have:

0⊂DU0,h(E,F)⊂DU1,h(E,F)⊂⋯⊂DUk,h(E,F). 0 \subset \mathscr{D}^{0,h}_U(E,F) \subset \mathscr{D}^{1,h}_U(E,F) \subset \cdots \subset \mathscr{D}^{k,h}_U(E,F).

If r=∞r=\infty or ω\omega, we denote by DUω,h(E,F)\mathscr{D}^{\omega,h}_U(E,F) the union of the DUk,h(E,F)\mathscr{D}^{k,h}_U(E,F) for k≥0k \geq 0; an element of this union is called a differential operator on U of bounded order (of type E→FE \rightarrow F, of class ChC^h) or sometimes simply a differential operator on U.

14.1.2 (« Operators of order zero »). We have P0(E)=EP^0(E)=E (12.5.1) and D0(E,F)=L(E;F)D^0(E,F)=\mathscr{L}(E;F). If U is open in X, and if h∈Nk∪{0}h \in \mathbf{N}_k \cup \{0\} and h≤rh \leq r, an element of DU0,h(E,F)\mathscr{D}^{0,h}_U(E,F) is a morphism of class ChC^h of E∣UE|U into F∣UF|U.

14.1.3 (“Weakening of structure”). Suppose that K=RK = \mathbf{R}, and let r′∈NRr' \in \mathbf{N}_\mathbf{R} with r′≤rr' \leq r. Let X′,E′X', E' and F′F' be the manifold and the vector bundles of class Cr′C^{r'} obtained by weakening of structure from X,E,X, E, and FF respectively. Let kk be an integer such that 0≤k≤r′0 \leq k \leq r'. Then Pk(E′)P^k(E') is the bundle of class Cr′−kC^{r'-k} deduced from Pk(E)P^k(E) by weakening of structure, and there is an analogous result for Dk(E′,F′)D^k(E', F'). In particular, if UU is open in XX, and if 0≤h≤r−k′0 \leq h \leq r - k', one has DUk′,h(E′,F′)=DUk,h(E,F)\mathscr{D}_U^{k',h}(E', F') = \mathscr{D}_U^{k,h}(E, F).

14.1.4. Let DD be a differential operator on XX, of type E→FE \to F, of order ≤k\leq k (where 0≤k≤r0 \leq k \leq r). Let UU be an open set in XX, and let ss be a section, of class CmC^m, of EE over UU, with m∈NK∪{0}m \in \mathbf{N}_K \cup \{0\} and k≤m≤rk \leq m \leq r. Then jk(s)j^k(s) is a section of class Cm−kC^{m-k} of Pk(E)P^k(E) over UU (12.6.1); its image under DD is denoted by DU(s)D_U(s), or D(s)D(s). It is a section of F∣UF|U. Suppose that DD is of class ChC^h, where h=m−kh = m - k. Then DU(s)D_U(s) is of class ChC^h, and the mappings

DU:SEm(U)→SFh(U) D_U : \mathscr{S}_E^m(U) \to \mathscr{S}_F^h(U)
thus obtained enjoy the following properties:

(1) DUD_U is KK-linear. (2) For every s∈SEm(U)s \in \mathscr{S}_E^m(U) and every open set VV of UU, one has DV(s∣V)=DU(s)∣VD_V(s|V) = D_U(s)|V. (3) For every s∈SEm(U)s \in \mathscr{S}_E^m(U) and every x∈Ux \in U such that jxk(s)=0j_x^k(s) = 0, one has DU(s)(x)=0D_U(s)(x) = 0. (3') For every coordinate system ξ=(ξ1,…,ξn)\xi = (\xi^1, \ldots, \xi^n) in UU and every frame s=(s1,…,sd)s = (s_1, \ldots, s_d) of EE over UU (7.4.4), there exist sections ni,α(1≤i≤d,α∈Nn,∣α∣≤k)n_{i,\alpha} (1 \leq i \leq d, \alpha \in \mathbf{N}^n, |\alpha| \leq k) of FF over UU, of class ChC^h and such that, for every family (fi)1≤i≤d(f_i)_{1 \leq i \leq d} of elements of Cm(U)\mathscr{C}^m(U), one has

DU(∑1≤i≤dfi.si)=∑1≤i≤d,∣α∣≤kΔξα(fi).ni,α,(cf. 13.2.6). D_U \left( \sum_{1 \leq i \leq d} f_i . s_i \right) = \sum_{1 \leq i \leq d, |\alpha| \leq k} \Delta_\xi^\alpha(f_i) . n_{i,\alpha}, \quad \text{(cf. 13.2.6).}
(4) For every function f∈Cm(U)f \in \mathscr{C}^m(U) and every mapping θ\theta of SUm(E)\mathscr{S}_U^m(E) into SUh(F)\mathscr{S}_U^h(F), let ad⁡(f)θ\operatorname{ad}(f)\theta denote the mapping
s↦f.θ(s)−θ(f.s) s \mapsto f . \theta(s) - \theta(f.s)
of SUm(E)\mathscr{S}_U^m(E) into SUh(F)\mathscr{S}_U^h(F). Then, for every function f∈Cm(U)f \in \mathscr{C}^m(U), there exists a differential operator LL on UU, of type E→FE \to F, of order ≤k−1\leq k - 1 and of class ChC^h such that
(ad⁡(f)DU)(s)=LU(s)for every s∈SUm(E). (\operatorname{ad}(f)D_U)(s) = L_U(s) \quad \text{for every } s \in \mathscr{S}_U^m(E).
(5) One has
ad⁡(f0)…ad⁡(fk)DU=0for arbitrary f0,…,fk in Cm(U). \operatorname{ad}(f_0) \ldots \operatorname{ad}(f_k) D_U = 0 \quad \text{for arbitrary } f_0, \ldots, f_k \text{ in } \mathscr{C}^m(U).
If one sets I={0,…,k}I = \{0, \ldots, k\} and fH=∏i∈Hfif_H = \prod_{i \in H} f_i for every subset HH of II, the preceding relation is equivalent to:
∑H⊂I(−1)Card⁡(H)fH.DU(fI−H.s)=0 \sum_{H \subset I} (-1)^{\operatorname{Card}(H)} f_H . D_U(f_{I-H}.s) = 0
for arbitrary f0,…,fkf_0, \ldots, f_k in Cm(U)\mathscr{C}^m(U) and ss in SEm(U)\mathscr{S}_E^m(U).

14.1.5 ("Characterization of differential operators"). Let k,mk, m and h=m−kh = m - k be as in 14.1.4. For every open set UU of XX, let DUD_U be a mapping of SEm(U)\mathscr{S}_E^m(U) into SFh(U)\mathscr{S}_F^h(U). Suppose that conditions 1, 2, 3 (resp. 1, 2, 3') of 14.1.4 are satisfied for every open subset U of X. Then there exists an element D of DXk,h(E,F)\mathscr{D}_X^{k,h}(E, F), and only one, such that the DUD_U are the corresponding mappings.(1)^{(1)} In what follows, D is identified with the family of the DUD_U.

When m=∞m = \infty or ω\omega, we have h=mh = m and one may replace conditions (3) and (3') by either of conditions (4) and (5).

14.1.6 (“Scalar differential operators”). Suppose that E and F are equal to the trivial vector bundle KXK_X. A differential operator of type E→FE \to F is then called a scalar differential operator or simply a differential operator; if it is of order ≤k\leq k, it is a section of the vector bundle L(Pk(X);KX)=Pk(X)∗\mathscr{L}(P^k(X); K_X) = P^k(X)^*, dual of the vector bundle Pk(X)P^k(X); using the isomorphism

i−1:Pk(X)∗→T(k)(X)(cf. 13.2.5), i^{-1} : P^k(X)^* \to T^{(k)}(X) \quad (\text{cf. } 13.2.5),
one sees that a scalar differential operator of order ≤k\leq k is identified with a section of the vector bundle T(k)(X)T^{(k)}(X), i.e. with a field of point distributions of order ≤k\leq k.

Take in particular for X an open subset of KnK^n, where n is an integer ≥0\geq 0. The fields of point distributions

Δα:x↦Δxα(cf. 13.2.6) \Delta^\alpha : x \mapsto \Delta_x^\alpha \quad (\text{cf. } 13.2.6)
are scalar differential operators of class CωC^\omega. For every h∈NKh \in N_K, the Δα\Delta^\alpha for (∣α∣≤k)(|\alpha| \leq k) form a basis of the Ch(X)C^h(X)-module DXk,h(KX,KX)\mathscr{D}_X^{k,h}(K_X, K_X).

14.1.7 (“Complex differential operators on a real manifold”). Suppose K=RK = \mathbf{R} and that the vector bundles E and F are endowed with complex structures (8.8.1); let k be an integer such that 0≤k≤r0 \leq k \leq r. The vector bundle Pk(E)P^k(E) is then endowed with a complex structure (12.6.3); the vector bundle Lc(Pk(E);F)\mathscr{L}_c(P^k(E); F) (7.8.5) of complex linear mappings from Pk(E)P^k(E) into F is a vector subbundle of L(Pk(E);F)=Dk(E,F)\mathscr{L}(P^k(E); F) = D^k(E, F); we denote it by Dck(E,F)D_c^k(E, F). A section of Dck(E,F)D_c^k(E, F) is called a complex differential operator, of type E→FE \to F and of order ≤k\leq k. If h∈NKh \in N_K and h≤r−kh \leq r - k, and if U is an open subset of X, we denote likewise by DUk,h(E,F)c\mathscr{D}_U^{k,h}(E, F)_c the space of sections of class ChC^h of Dck(E,F)D_c^k(E, F) over U; it is a C-vector space. The other definitions and results of this paragraph extend analogously to complex differential operators; we leave their formulation to the reader.

14.1.8 (« Composition »). Let G be a vector bundle of class CrC^r and with base X. Let k′k' and k′′k'' be positive integers with sum k≤rk \leq r, and let D′D' (resp. D′′D'') be a differential operator on X, of type E→FE \to F (resp. of type F→GF \to G), of order ≤k′\leq k' (resp. ≤k′′\leq k''). Suppose D′:Pk′(E)→FD' : P^{k'}(E) \to F is of class Ch′C^{h'}, with h′∈NK∪{0}h' \in N_K \cup \{0\} and k′′≤h′≤r−k′k'' \leq h' \leq r - k'; the mapping

D∗′=Pk′′(D′):Pk′′(Pk′(E))→Pk′′(F) D'_* = P^{k''}(D') : P^{k''}(P^{k'}(E)) \to P^{k''}(F)
is of class Ch′−k′′C^{h'-k''} (12.6.3). Let β\beta be the canonical homomorphism from Pk(E)P^{k}(E) into Pk′′(Pk′(E))P^{k''}(P^{k'}(E)), cf. 12.6.5. By composing the mappings
Pk(E)→βPk′′(Pk′(E))→D∗′Pk′′(F)→D′′G, P^k(E) \xrightarrow{\beta} P^{k''}(P^{k'}(E)) \xrightarrow{D'_*} P^{k''}(F) \xrightarrow{D''} G,
one obtains a differential operator of type E→GE \to G and of order ≤k\leq k. This operator is called the composite of D′′D'' and D′D'; it is denoted by D′′∘D′D'' \circ D'.

1^1 It is in fact sufficient that condition (3') be satisfied for a family (ξλ)(\xi_\lambda) of systems of coordinates and a family of frames (sλ)(s_\lambda) whose domains cover X.

If U is an open subset of X, and if s ∈ SEk(U)\mathscr{S}_E^{k}(U), one has

(D′′∘D′)(s)=D′′(D′(s)) (D'' \circ D')(s) = D''(D'(s))
(where the two members are sections of G over U); this justifies the terminology and notation adopted.

Suppose now that D'' is of class Ch′′C^{h''}, with h′′∈NK∪{0}h'' \in \mathbf{N}_K \cup \{0\} and h′′≤r−k′′h'' \leq r - k''. Then D′′∘D′D'' \circ D' is of class ChC^h, with h=inf⁡(h′−k′′,h′′)h = \inf(h' - k'', h'').

Suppose that one has E=F=GE = F = G, and that k′≤h′′k' \leq h'', in which case D′∘D′′D' \circ D'' is defined. Then D′∘D′′−D′′∘D′D' \circ D'' - D'' \circ D' is a differential operator of type E→EE \to E and of order ≤k−1\leq k - 1; it is denoted by [D′,D′′][D', D''].

14.1.9 (« Associativity »). With the notation and hypotheses of 14.1.8, let H be a vector bundle of class CrC^r and with base X, and let D′′′D''' be a differential operator on X, of type G→HG \to H and of order ≤k′′′\leq k''', with k′+k′′+k′′′≤rk' + k'' + k''' \leq r. Suppose that D′D' is of class Ch′C^{h'}, with h′∈NK∪{0}h' \in \mathbf{N}_K \cup \{0\} and k′′+k′′′≤h′≤r−k′k'' + k''' \leq h' \leq r - k' and that D′′D'' is of class Ch′′C^{h''}, with h′′∈NK∪{0}h'' \in \mathbf{N}_K \cup \{0\} and k′′′≤h′′≤r−k′′k''' \leq h'' \leq r - k''. Then the composites

D′′′∘(D′′∘D′)and(D′′′∘D′′)∘D′ D''' \circ (D'' \circ D') \quad \text{and} \quad (D''' \circ D'') \circ D'
are defined and are equal.

14.1.10. Examples

a) Let U be an open subset of X, let D∈DUk,h(E,F)D \in \mathscr{D}_U^{k,h}(E, F) with h∈NK∪{0}h \in \mathbf{N}_K \cup \{0\} and h≤r−kh \leq r - k, and let f∈Ch+k(U)f \in \mathscr{C}^{h+k}(U) (resp. f∈Ch(U)f \in \mathscr{C}^h(U)). Multiplication by f in E (resp. F) is a differential operator on U of order ≤0\leq 0, and of type E→EE \to E (resp. F→FF \to F), cf. 14.1.2. The composites D∘fD \circ f and f∘Df \circ D are defined and belong to DUk,h(E;F)\mathscr{D}_U^{k,h}(E; F); we put

ad⁡(f)D=f∘D−D∘f∈DUk−1,h(E,F), \operatorname{ad}(f)D = f \circ D - D \circ f \in \mathscr{D}_U^{k-1,h}(E, F),
cf. 14.1.4, (4). Thus DUk,h(E,F)\mathscr{D}_U^{k,h}(E, F) is endowed with a Ch+k(U)\mathscr{C}^{h+k}(U)-module structure on the right (resp. a Ch(U)\mathscr{C}^h(U)-module structure on the left). The left module structure coincides with that of 14.1.1.

b) Suppose r=∞r = \infty or r=ωr = \omega. The differential operators of type E→EE \to E and of class CrC^r form an associative K-algebra with unit element.

c) Take for X an open subset of KnK^n. The scalar differential operators Δα\Delta^\alpha (14.1.6) satisfy the formulas

Δα∘Δβ=((α,β))Δα+β. \Delta^\alpha \circ \Delta^\beta = ((\alpha, \beta)) \Delta^{\alpha+\beta}.
They commute with one another. Denote by DiD_i the operator Δεi\Delta^{\varepsilon_i} (« i-th partial derivative »). If K is of characteristic 0, we have
Δα=1α!D1α1…Dnαn. \Delta^\alpha = \frac{1}{\alpha!} D_1^{\alpha_1} \ldots D_n^{\alpha_n}.
If K is of characteristic p≠0p \neq 0, we have Dip=0D_i^p = 0 for every ii.

14.1.11 (« Multidifferential operators »). Let E1,…,EnE_1, \ldots, E_n be vector bundles of class CrC^r and with base X, and let k,k1,…,knk, k_1, \ldots, k_n be integers belonging to [0,r][0, r]. Put

L(k1,…,kn)=L(Pk1(E1)⊗⋯⊗Pkn(En);F). L(k_1, \ldots, k_n) = \mathscr{L}(P^{k_1}(E_1) \otimes \cdots \otimes P^{k_n}(E_n); F).

If ki≤kk_i \leq k for all ii, the projections rki,ki:Pk(Ei)→Pki(Ei)r^{k_i, k_i} : P^k(E_i) \to P^{k_i}(E_i) make it possible to identify the bundle L(k1,…,kn)L(k_1, \ldots, k_n) with a sub-bundle of the bundle L(k)=L(k,…,k)L(k) = L(k, \ldots, k). There exists a smallest sub-bundle M(k)M(k) of L(k)L(k) containing all the sub-bundles L(k1,…,kn)L(k_1, \ldots, k_n), for k1+⋯+kn=kk_1 + \cdots + k_n = k. A section HH of M(k)M(k) is called an nn-differential (or multidifferential) operator of type (E1,…,En)→F(E_1, \ldots, E_n) \to F and of order ≤k\leq k. If sis_i (1≤i≤n1 \leq i \leq n) is a section of class CmC^m (m∈NK,m≥km \in \mathbf{N}_K, m \geq k) of EiE_i over an open set UU of XX, one defines HU(s1,…,sn)=H(s1,…,sn)H_U(s_1, \ldots, s_n) = H(s_1, \ldots, s_n) as the image under HH of the section jk(s1)⊗⋯⊗jk(sn)j^k(s_1) \otimes \cdots \otimes j^k(s_n) of the bundle

Pk(E1)⊗⋯⊗Pk(En). P^k(E_1) \otimes \cdots \otimes P^k(E_n).
It is a section of F∣UF|U. If HH is of class ChC^h, this section is of class CpC^p with p=inf⁡(m−k,h)p = \inf(m - k, h).

When E1,…,EnE_1, \ldots, E_n and FF are equal to KXK_X, one says that HH is a scalar nn-differential (or multidifferential) operator.

Example

Let JJ be a finite set, and XX an open set of KJK^J. Let HH be a scalar nn-differential operator of order ≤k\leq k on XX. There exists a unique family of scalar functions

c(α(1),…,α(n))(α(1),…,α(n))∈NJ,∑i=1n∣α(i)∣≤k c(\alpha(1), \ldots, \alpha(n))_{(\alpha(1), \ldots, \alpha(n)) \in \mathbf{N}^J}, \quad \sum_{i=1}^n |\alpha(i)| \leq k
on XX such that
HX(f1,…,fn)=∑α(1),…,α(n)c(α(1),…,α(n))Δα(1)(f1)…Δα(n)(fn) H_X(f_1, \ldots, f_n) = \sum_{\alpha(1), \ldots, \alpha(n)} c(\alpha(1), \ldots, \alpha(n)) \Delta^{\alpha(1)}(f_1) \ldots \Delta^{\alpha(n)}(f_n)
for every family (f1,…,fn)(f_1, \ldots, f_n) of functions of class CkC^k on XX. In order that HH be of class ChC^h, it is necessary and sufficient that the functions c(α(1),…,α(n))c(\alpha(1), \ldots, \alpha(n)) be of class ChC^h.

14.2. Symbols

14.2.1. Let kk be a positive integer ≤r\leq r. Consider the exact sequence

0→Pk(T(X);E)→i′Pk(E)→j′Pk−1(E)→0 0 \to P_k(T(X); E) \xrightarrow{i'} P^k(E) \xrightarrow{j'} P^{k-1}(E) \to 0
of No. 12.6.8, where j=rk,k−1j = r^{k, k-1}. Applying the vector bundle functor M↦L(M;F)M \mapsto \mathscr{L}(M; F) to this sequence, we obtain the exact sequence
0→L(Pk−1(E);F)→j′′L(Pk(E);F)→σkL(Pk(T(X);E);F)→0 0 \to \mathscr{L}(P^{k-1}(E); F) \xrightarrow{j''} \mathscr{L}(P^k(E); F) \xrightarrow{\sigma_k} \mathscr{L}(P_k(T(X); E); F) \to 0
which can be written:
0→Dk−1(E,F)→j′′Dk(E,F)→σkSk(E,F)→0 0 \to D^{k-1}(E, F) \xrightarrow{j''} D^k(E, F) \xrightarrow{\sigma_k} S^k(E, F) \to 0
by putting
Sk(E,F)=L(Pk(T(X);E);F). S^k(E, F) = \mathscr{L}(P_k(T(X); E); F).
The homomorphism j′=L(j;IdF)j' = \mathscr{L}(j; \mathrm{Id}_F) is the canonical inclusion of Dk−1(E,F)D^{k-1}(E, F) in Dk(E,F)D^k(E, F); the homomorphism σk\sigma_k is, by definition, equal to L(i;IdF)\mathscr{L}(i; \mathrm{Id}_F). If DD is a differential operator of type E→FE \to F and of order ≤k\leq k, its image under σk\sigma_k is a section σk(D)\sigma_k(D) of Sk(E,F)S^k(E, F) which is called the kk-symbol (or simply the symbol) of DD; we have σk(D)=0\sigma_k(D) = 0 if and only if DD is of order ≤k−1\leq k-1. If DD is of class ChC^h, the same is true of its symbol.

14.2.2 (« Identifications of the symbol bundle »). The bundle Sk(E,F)=L(Pk(T(X);E);F)S^k(E, F) = \mathscr{L}(P_k(T(X); E); F) can be written in various ways. First, if x∈Xx \in X, an element of Pk(Tx(X);Ex)P_k(T_x(X); E_x) is identified (13.1.2) with an element of L(TSk(Tx(X));Ex)\mathscr{L}(TS^k(T_x(X)); E_x). We thus obtain identifications of vector bundles

Pk(T(X);E)=L(TSk(T(X));E)=(TSk(T(X)))∗⊗E P_k(T(X); E) = \mathscr{L}(TS^k(T(X)); E) = (TS^k(T(X)))* \otimes E

and, applying the vector bundle functor M↦L(M;F)=M∗⊗FM \mapsto \mathscr{L}(M; F) = M^* \otimes F, we obtain:

Sk(E,F)=TSk(T(X))⊗E∗⊗F=TSk(T(X))⊗L(E;F). S^k(E, F) = TS^k(T(X)) \otimes E^* \otimes F = TS^k(T(X)) \otimes \mathscr{L}(E; F).

The preceding expression can be transformed further. If MM and NN are two vector bundles, let Symk(M;N)\mathrm{Sym}^k(M; N) denote the subbundle of L(M,…,M;N)\mathscr{L}(M, \ldots, M; N) formed by symmetric kk-linear mappings. We have a canonical isomorphism

Symk(M;N)→Symk(M;KX)⊗N; \mathrm{Sym}^k(M; N) \to \mathrm{Sym}^k(M; K_X) \otimes N;

on the other hand, the duality between ⊗kM\otimes^k M and ⊗kM∗\otimes^k M^* identifies Symk(M;KX)\mathrm{Sym}^k(M; K_X) with TSk(M∗)TS^k(M^*). We thus obtain an identification

Symk(M;N)=TSk(M∗)⊗N. \mathrm{Sym}^k(M; N) = TS^k(M^*) \otimes N.

Applying this formula to M=T(X)∗M = T(X)^* and N=L(E;F)N = \mathscr{L}(E; F), we have:

Sk(E,F)=TSk(T(X))⊗L(E;F)=Symk(T(X)∗;L(E;F)). S^k(E, F) = TS^k(T(X)) \otimes \mathscr{L}(E; F) = \mathrm{Sym}^k(T(X)^*; \mathscr{L}(E; F)).

Suppose KK is of characteristic 0 or >k> k. If uu is a symmetric kk-linear mapping, let u~\tilde{u} denote the polynomial mapping

x↦1k!u(x,…,x) x \mapsto \frac{1}{k!} u(x, \ldots, x)

corresponding to it. The mapping u↦u~u \mapsto \tilde{u} defines an isomorphism

Sk(E,F)=Symk(T(X)∗,L(E;F))→Pk(T(X)∗;L(E;F)). S^k(E, F) = \mathrm{Sym}^k(T(X)^*, \mathscr{L}(E; F)) \to P_k(T(X)^*; \mathscr{L}(E; F)).

Thus, every element σ\sigma of Sk(E,F)xS^k(E, F)_x, with x∈Xx \in X, may be identified with a homogeneous polynomial mapping σ~\tilde{\sigma} of degree kk from Tx(X)∗T_x(X)^* into L(Ex;Fx)\mathscr{L}(E_x; F_x).

14.2.3 (“Calculation of the symbol of a differential operator”). Let x∈Xx \in X and let DD be a differential operator of type E→FE \to F, of order ≤k\leq k, defined in an open neighbourhood of xx and of class Cr−kC^{r-k}. We shall make explicit the value σk(D)(x)\sigma_k(D)(x) of the symbol of DD at xx. Consider it as an element of Symk(Tx(X)∗;L(Ex;Fx))\mathrm{Sym}^k(T_x(X)^*; \mathscr{L}(E_x; F_x)) (cf. 14.2.2). Let ω1,…,ωk\omega_1, \ldots, \omega_k be covectors at xx and choose functions f1,…,fkf_1, \ldots, f_k of class CrC^r in an open neighbourhood UU of xx such that dxfi=ωid_x f_i = \omega_i for i=1,…,ki = 1, \ldots, k. Put

D′=(−1)kad (fk)…ad(f1)D(cf. 14.1.10,a). D' = (-1)^k \mathrm{ad}\,(f_k) \ldots \mathrm{ad}(f_1)D \quad (\text{cf. } 14.1.10, a).

The operator D′D' is of order ≤0\leq 0; it is a section of L(E;F)∣U\mathscr{L}(E; F)|U. The value of D′D' at xx is an element λ(ω1,…,ωk)\lambda(\omega_1, \ldots, \omega_k) of L(Ex;Fx)\mathscr{L}(E_x; F_x) which depends only on (ω1,…,ωk)(\omega_1, \ldots, \omega_k). The resulting mapping λ\lambda is symmetric. It is equal to the symbol σk(D)(x)\sigma_k(D)(x) of DD at xx.

Suppose that K=RK = \mathbf{R} or C\mathbf{C}, and let us make explicit σk(D)(x)\sigma_k(D)(x) considered as a homogeneous polynomial mapping of degree kk from Tx(X)∗T_x(X)^* into L(Ex;Fx)\mathscr{L}(E_x; F_x) (cf. 14.2.2). Let ω∈Tx(X)∗\omega \in T_x(X)^* and v∈Exv \in E_x; choose a function ff of class CrC^r in an open neighbourhood U of x such that dxf=ωd_x f = \omega, and a section s of class CrC^r of E in U such that s(x)=vs(x) = v. There exists a family (φ0,φ1,…,φk)(\varphi_0, \varphi_1, \ldots, \varphi_k) of sections of F in U, and only one, such that

e−tfD(etfs)=∑j=0ktjφjfor all t∈K. e^{-tf} D(e^{tf}s) = \sum_{j=0}^k t^j \varphi_j \quad \text{for all } t \in K.
Then φk(x)\varphi_k(x) does not depend on the choice of f and s such that dxf=ω,v(x)=sd_x f = \omega, v(x) = s; the mapping v↦φk(x)v \mapsto \varphi_k(x) is an element λ(ω)\lambda(\omega) of L(Ex;Fx)\mathscr{L}(E_x; F_x), and λ\lambda is a homogeneous polynomial mapping of degree k from Tx(X)∗T_x(X)^* into L(Ex;Fx)\mathscr{L}(E_x; F_x). This mapping is equal to the symbol σk(D)(x)\sigma_k(D)(x) of D at x.

14.2.4 ("The scalar case"). Take E=F=KXE = F = K_X, in which case one identifies Dk(E,F)D^k(E, F) with the bundle T(k)(X)T^{(k)}(X) of point distributions of order ≤k\leq k, cf. 14.1.6. One has

Sk(E,F)=TSk(T(X)) S^k(E, F) = TS^k(T(X))
and the symbol
σk:T(k)(X)→TSk(T(X)) \sigma_k : T^{(k)}(X) \to TS^k(T(X))
is none other than the composite
T(k)(X)→T(k)(X)/T(k−1)(X)→ikTSk(T(X)), T^{(k)}(X) \to T^{(k)}(X)/T^{(k-1)}(X) \xrightarrow{i_k} TS^k(T(X)),
where iki_k is the isomorphism defined in 13.3.2.

Take for example X to be an open subset of KnK^n, and let {e1,…,en}\{e_1, \ldots, e_n\} be the canonical basis of KnK^n (identified with the tangent spaces Tx(X)T_x(X)). If α∈Nn\alpha \in \mathbf{N}^n is such that ∣α∣≤k|\alpha| \leq k, the symbol of the operator Δα\Delta^\alpha is given by the formulas (cf. 13.3.3):

σk(Δα)(x)=0if ∣α∣<kσk(Δα)(x)=γα1(e1)⋯γαn(en)if ∣α∣=k, \begin{aligned} \sigma_k(\Delta^\alpha)(x) &= 0 & \text{if } |\alpha| < k \\ \sigma_k(\Delta^\alpha)(x) &= \gamma_{\alpha_1}(e_1) \cdots \gamma_{\alpha_n}(e_n) & \text{if } |\alpha| = k, \end{aligned}
the product of the γαi(ei)\gamma_{\alpha_i}(e_i) being the symmetric product, cf. 13.2.6 and 13.3.3.

14.2.5 ("Symbol of a composite"). The notations and hypotheses are those of 14.1.8. The composition of linear mappings defines a pairing

L(F;G)×L(E;F)→L(E;G). \mathscr{L}(F; G) \times \mathscr{L}(E; F) \to \mathscr{L}(E; G).
On the other hand, the operation of symmetric product defines a pairing
TSk′′(T(X))×TSk′(T(X))→TSk(T(X)). TS^{k''}(T(X)) \times TS^{k'}(T(X)) \to TS^k(T(X)).
Since one has
Sk′′(F,G)=TSk′′(T(X))⊗L(F;G)Sk′(E,F)=TSk′(T(X))⊗L(E;F)Sk(E,G)=TSk(T(X))⊗L(E;G) \begin{aligned} S^{k''}(F, G) &= TS^{k''}(T(X)) \otimes \mathscr{L}(F; G) \\ S^{k'}(E, F) &= TS^{k'}(T(X)) \otimes \mathscr{L}(E; F) \\ S^k(E, G) &= TS^k(T(X)) \otimes \mathscr{L}(E; G) \end{aligned}
cf. 14.2.2, one deduces from this, by tensor product, a pairing (*)
Sk′′(F,G)×Sk′(E,F)→Sk(E,G). S^{k''}(F, G) \times S^{k'}(E, F) \to S^k(E, G).
One then has the formula
σk(D′′∘D′)=σk′′(D′′)⋅σk′(D′), \sigma_k(D'' \circ D') = \sigma_{k''}(D'') \cdot \sigma_{k'}(D'),

where the product occurring in the right-hand term is defined by the pairing (*) above.

Now suppose K of characteristic zero, and let x∈Xx \in X. Denote by σ\sigma (resp. σ′,σ′′\sigma', \sigma'') the k-symbol (resp. the k′k'-symbol, the k′′k''-symbol) of D′′∘D′D'' \circ D' (resp. D′,D′′D', D'') at xx. For every ω∈Tx(X)∗\omega \in T_x(X)^*, one then has (with the notations at the end of 14.2.2):

σ~(ω)∈L(Ex;Gx),σ~′(ω)∈L(Ex;Fx),σ~′′(ω)∈L(Fx;Gx) \tilde{\sigma}(\omega) \in \mathscr{L}(E_x; G_x), \quad \tilde{\sigma}'(\omega) \in \mathscr{L}(E_x; F_x), \quad \tilde{\sigma}''(\omega) \in \mathscr{L}(F_x; G_x)

and

σ~(ω)=σ~′′(ω)∘σ~′(ω). \tilde{\sigma}(\omega) = \tilde{\sigma}''(\omega) \circ \tilde{\sigma}'(\omega).

14.3. Transposition

In this No., it is assumed that X is purely of dimension nn.

14.3. Transposition

14.3.1. Let Ω=det⁡(T(X)∗)\Omega = \det(T(X)^*), cf. 7.9.9; it is a vector bundle of rank 1 at each point, with base X, and of class Cr−1C^{r-1}. One has

Ω=∧nT(X)∗=Altn(T(X);KX). \Omega = \wedge^n T(X)^* = \mathrm{Alt}^n(T(X); K_X).

Let M be a vector bundle with base X. Put

M~=L(M;Ω)=M∗⊗Ω. \tilde{M} = \mathscr{L}(M; \Omega) = M^* \otimes \Omega.

[^1]

If M is of class CnC^n, with h∈NK∪{0}h \in N_K \cup \{0\} and h≤r−1h \leq r - 1, the same is true of M~\tilde{M}. The canonical mapping of M into M~=L(L(M;Ω);Ω)\tilde{M} = \mathscr{L}(\mathscr{L}(M; \Omega); \Omega) is an isomorphism; it is used to identify M with M~\tilde{M}.

14.3.2 ("Definition of the transpose"). Let kk be a positive integer ≤r−1\leq r - 1, and let D be a differential operator on X, of type E→FE \to F, of order ≤k\leq k, and of class CnC^n, with h∈NK∪{0}h \in N_K \cup \{0\} and k≤h≤r−kk \leq h \leq r - k. Then there exists a differential operator tD^tD on X, of type Fˉ→Eˉ\bar{F} \to \bar{E} and of order ≤k\leq k, having the following property:

Let ξ=(ξ1,…,ξn)\xi = (\xi^1, \ldots, \xi^n) be a system of coordinates in an open set U of X and let s=(si)1≤i≤es = (s_i)_{1 \leq i \leq e} (resp. t=(tj)1≤j≤ft = (t_j)_{1 \leq j \leq f}) be a frame of E (resp. of F) on U. Let (si∗)(s_i^*) (resp. (tj∗)(t_j^*)) be the frame of E∗E^* (resp. of F∗F^*) such that ⟨si,sj∗⟩=δij\langle s_i, s_j^* \rangle = \delta_{ij} (resp. ⟨ti,tj∗⟩=δij\langle t_i, t_j^* \rangle = \delta_{ij}), and let (s~i)(\tilde{s}_i) (resp. (t~j)(\tilde{t}_j)) be the frame of Eˉ\bar{E} (resp. Fˉ\bar{F}) on U obtained by forming the tensor product of (si∗)(s_i^*) (resp. of (tj∗)(t_j^*)) with the frame ω=dξ1∧⋯∧dξn\omega = d\xi^1 \wedge \cdots \wedge d\xi^n of Ω\Omega. Let cαijc_{\alpha}^{ij} (1≤i≤e1 \leq i \leq e, 1≤j≤f1 \leq j \leq f, ∣α∣≤k|\alpha| \leq k) be the functions of class CnC^n on U such that one has

D(∑ifisi)=∑i,j,αcαijΔξα(fi)⋅tj D \left( \sum_i f_i s_i \right) = \sum_{i,j,\alpha} c_{\alpha}^{ij} \Delta_{\xi}^{\alpha}(f_i) \cdot t_j

for arbitrary functions fif_i in Ch+k(U)C^{h+k}(U), cf. 14.1.4, (3'). One then has

tD(∑jgjt~j)=∑i,j,α(−1)∣α∣Δξα(cαijgj)s~i {}^tD \left( \sum_j g_j \tilde{t}_j \right) = \sum_{i,j,\alpha} (-1)^{|\alpha|} \Delta_{\xi}^{\alpha}(c_{\alpha}^{ij} g_j) \tilde{s}_i

for arbitrary functions gjg_j in Ch+k(U)C^{h+k}(U).

80 DIFFERENTIAL AND ANALYTIC MANIFOLDS § 14

The preceding property (which must be verified whatever ξ\xi, ss and tt may be) determines tD^{t}D uniquely.[^2] The operator tD^{t}D is called the transpose of DD; it is of class Ch−kC^{h-k}. If h≥2kh\geq 2k, the transpose of tD^{t}D is defined and equal to DD (taking into account the identifications of E~\widetilde{E} and F~\widetilde{F} with EE and FF respectively).

The mapping D↦tDD\mapsto{}^{t}D is KK-linear. If k=0k=0, i.e. if DD is a morphism from EE into FF, tD^{t}D is the morphism L(D;Id⁡Ω)\mathscr{L}(D;\operatorname{Id}_{\Omega}) from F~\widetilde{F} into E~\widetilde{E}.

The symbol of tD^{t}D is deduced from the symbol of DD by means of the isomorphism

TS⁡k(T(X))⊗L(E;F)⟶TS⁡k(T(X))⊗L(F~;E~) \operatorname{TS}^{k}(T(X))\otimes\mathscr{L}(E;F)\longrightarrow \operatorname{TS}^{k}(T(X))\otimes\mathscr{L}(\widetilde{F};\widetilde{E})

which is the tensor product of the automorphism (−1)k(-1)^k of TS⁡k(T(X))\operatorname{TS}^{k}(T(X)) and the isomorphism u↦L(u;Id⁡Ω)u\mapsto\mathscr{L}(u;\operatorname{Id}_{\Omega}) of L(E;F)\mathscr{L}(E;F) onto L(F~;E~)\mathscr{L}(\widetilde{F};\widetilde{E}).

14.3.3 (“Transpose of a Composite”). With the notation and hypotheses of 14.1.8, suppose that D′D' is of class Ch′C^{h'} and D′′D'' of class Ch′′C^{h''}, with h′,h′′h',h'' in NK∪{0}\mathbf{N}_K\cup\{0\}, h′≥k′+2k′′h'\geq k'+2k'' and h′′≥k′′+2k′h''\geq k''+2k'.

Then the transposes of D′D', D′′D'' and D′′∘D′D''\circ D' are defined, as is the composite of tD′^{t}D' and tD′′^{t}D'', and one has

t(D′′∘D′)=tD′∘tD′′. ^{t}(D''\circ D')={}^{t}D'\circ{}^{t}D''.

In particular, with the hypotheses and notation of 14.3.2, and ff a function of class Ch′C^{h'} on XX with 2k≤h′2k\leq h', one has t(D∘f)=f∘tD^{t}(D\circ f)=f\circ{}^{t}D.

Example

Let XX be an open subset of KnK^n, let ξ1,…,ξn\xi^1,\ldots,\xi^n denote the coordinate functions on XX, and identify Ω\Omega with KXK_X by means of the frame ω=dξ1∧⋯∧dξn\omega=d\xi^1\wedge\cdots\wedge d\xi^n. If E=F=KXE=F=K_X, one has

E~=F~=L(KX;Ω)=L(KX;KX)=KX, \widetilde{E}=\widetilde{F}=\mathscr{L}(K_X;\Omega)=\mathscr{L}(K_X;K_X)=K_X,

and the operation D↦tDD\mapsto{}^{t}D transforms scalar differential operators into scalar differential operators. One has, for example,

t(Δα)=(−1)∣α∣Δαfor every α∈Nn. {}^{t}(\Delta^\alpha)=(-1)^{|\alpha|}\Delta^\alpha\qquad\text{for every }\alpha\in\mathbf{N}^n.

If rr is infinite, transposition is an anti-automorphism of the algebra DKX∞(KX,KX)\mathscr{D}^{\infty}_{K_X}(K_X,K_X).

14.3.4. Let kk be a positive integer such that KK is of characteristic 00 or >k>k. Let Ω1\Omega_1 be the vector bundle ⋀n−1T(X)∗=L(T(X);KX)\bigwedge^{n-1}T(X)^*=\mathscr{L}(T(X);K_X). The notation and hypotheses being those of 14.3.2, let moreover D′D' be a differential operator on XX, from F~→E~\widetilde{F}\rightarrow\widetilde{E}, of order ≤k\leq k; a Green operator for (D,D′)(D,D') is defined to be any multidifferential operator GG (cf. 14.1.11) of type (E,F~)→Ω1(E,\widetilde{F})\rightarrow\Omega_1, of order ≤k−1\leq k-1[^3] and of class ChC^h (with h≥1h\geq1), such that one has the identity

(1)

⟨D(u),v⟩−⟨u,D′(v)⟩=d(G(u,v)) \langle D(u),v\rangle-\langle u,D'(v)\rangle=d(G(u,v))

for every open set UU of XX and every elements u∈SEk(U)u\in\mathscr{S}^k_E(U), v∈SFk(U)v\in\mathscr{S}^k_F(U). Let us specify that, in this formula, dd denotes exterior differentiation (8.3.5) and ⟨D(u),v⟩\langle D(u),v\rangle (resp.

⟨u,D′(v)⟩\langle u, D'(v) \rangle) denotes the section of Ω\Omega on UU obtained from the pair (D(u),v)(D(u), v) (resp. from the pair (u,D′(v))(u, D'(v))) by the canonical pairing of F×F~F \times \tilde{F} (resp. of E×E~E \times \tilde{E}) into Ω\Omega. The existence of GG implies D′=tDD' = {}^tD; one also says that GG is a Green operator for DD.

14.3.5. Let DD be a differential operator on XX, of type E→FE \to F, of order ≤k\leq k, of class ChC^h with h∈NKh \in N_K and k≤h≤r−kk \leq h \leq r - k. There exists a Green operator of class Ch−k+1C^{h-k+1} for DD in each of the following cases:

a) K=RK = \mathbf{R}, r=∞r = \infty and XX is paracompact. b) KK is of characteristic 00 or >k> k, XX is isomorphic to an open set of a space KnK^n, and the bundles EE and FF are isomorphic to trivial bundles. c) The operator DD is of order ≤1\leq 1 (cf. 14.3.7).

14.3.6. The hypotheses and notations of 14.3.3 are retained. If G′G' (resp. G′′G'') is a Green operator for D′D' (resp. D′′D''), there exists one and only one Green operator HH for D′′∘D′D'' \circ D' such that one has

H(u,v)=G′′(D′(u),v)+G′(u,tD′′(v)) H(u, v) = G''(D'(u), v) + G'(u, {}^tD''(v))

for all open sets UU of XX and elements u∈SEk(U)u \in \mathscr{S}_E^k(U) and v∈SG~k(U)v \in \mathscr{S}_{\tilde{G}}^k(U).

14.3.7. One has S1(E,F)=T(X)⊗E∗⊗FS^1(E, F) = T(X) \otimes E^* \otimes F (cf. 14.2.2). On the other hand, the right interior product (ξ,ω)↦i(ξ)ω(\xi, \omega) \mapsto i(\xi)\omega (cf. A, III, p. 158) defines an isomorphism of T(X)⊗ΩT(X) \otimes \Omega onto Ω1\Omega_1; by tensor product with the dual Ω∗\Omega^* of Ω\Omega, one thus obtains an isomorphism of T(X)T(X) onto Ω∗⊗Ω1\Omega^* \otimes \Omega_1, whence identifications of vector bundles

S1(E,F)=Ω∗⊗Ω1⊗E∗⊗F=(E⊗F∗⊗Ω)∗⊗Ω1=(E⊗F~)∗⊗Ω1=L(E⊗F~;Ω1). \begin{align*} S^1(E, F) &= \Omega^* \otimes \Omega_1 \otimes E^* \otimes F = (E \otimes F^* \otimes \Omega)^* \otimes \Omega_1 \\ &= (E \otimes \tilde{F})^* \otimes \Omega_1 = \mathscr{L}(E \otimes \tilde{F}; \Omega_1). \end{align*}

This being set, let DD be a differential operator on XX, of type E→FE \to F, of order ≤1\leq 1, of class ChC^h with h∈NK∪{0}h \in N_K \cup \{0\} and h≤r−1h \leq r - 1. There exists a unique Green operator for DD; it is of order 00; it is a morphism of class ChC^h from E⊗F~E \otimes \tilde{F} into Ω1\Omega_1, and it is deduced from the symbol of DD by the preceding identification.

14.3.8. Suppose that one has K=RK = \mathbf{R}, that XX is separated and oriented (10.2.4) and let AA be a closed piece (11.1.2) of XX. Equip AA with the orientation induced by that of XX, and ∂A\partial A with the corresponding orientation (11.2.1). Let kk be an integer such that 2k≤r2k \leq r, DD a differential operator on XX, of type E→FE \to F, of order ≤k\leq k and of class CkC^k. Let GG be a Green operator for DD. Let UU be an open set of XX, u∈SEk(U)u \in \mathscr{S}_E^k(U), v∈SF~k(U)v \in \mathscr{S}_{\tilde{F}}^k(U) and suppose that the supports of uu and vv meet AA in a compact set. One then has

∫A⟨D(u),v⟩−∫A⟨u,tD(v)⟩=∫∂AG(u,v) \int_A \langle D(u), v \rangle - \int_A \langle u, {}^tD(v) \rangle = \int_{\partial A} G(u, v)

(Green's formula). In particular, if ∂A=∅\partial A = \varnothing, one has

∫A⟨D(u),v⟩=∫A⟨u,tD(v)⟩. \int_A \langle D(u), v \rangle = \int_A \langle u, {}^tD(v) \rangle.

14.4. Examples

Suppose X purely of finite dimension n. Suppose it is also endowed with a structure of manifold of class C^{r+1} compatible with the given structure of class C^r.

14.4.1 (« Infinitesimal transformations »). Let τ\tau be a vector functor in finite dimension for isomorphisms, of class C^r; suppose that, for every finite-dimensional vector space V, τ(V)\tau(V) is finite-dimensional. We denote by EτE_{\tau} the vector bundle τ(T(X))\tau(T(X)); it is of class C^r.

Let ξ\xi be a vector field of class C^r on X. There exists a differential operator D of type Eτ→EτE_{\tau} \to E_{\tau} and of order ≤1\leq 1 such that

DU(s)=θξ.s D_U(s) = \theta_{\xi}.s

for every open set U of X and every s∈SEτ(U)s \in \mathscr{S}_{E_{\tau}}(U), cf. 8.4.3. Such a differential operator D is unique; we denote it by (θξ)τ(\theta_{\xi})_{\tau} or simply θξ\theta_{\xi}. Its symbol is the section ξ⊗IdEτ\xi \otimes \mathrm{Id}_{E_{\tau}} of the vector bundle

S1(Eτ,Eτ)=T(X)⊗L(Eτ;Eτ). S^1(E_{\tau}, E_{\tau}) = T(X) \otimes \mathscr{L}(E_{\tau}; E_{\tau}).

If V is a finite-dimensional K-vector space, put

τ~(V)=L(τ(V);Altn(V;K))=τ(V)∗⊗∧nV∗, \tilde{\tau}(V) = \mathscr{L}(\tau(V); \mathrm{Alt}^n(V; K)) = \tau(V)^* \otimes \wedge^n V^*,

and if u:V1→V2u : V_1 \to V_2 is an isomorphism, define τ~(u):τ~(V1)→τ~(V2)\tilde{\tau}(u) : \tilde{\tau}(V_1) \to \tilde{\tau}(V_2) by transport of structure. We thus obtain a vector functor in finite dimension τ~\tilde{\tau}. The bundle Eτ~=τ~(T(X))E_{\tilde{\tau}} = \tilde{\tau}(T(X)) identifies in the obvious way with the bundle (Eτ)∗(E_{\tau})^*; the transpose of (θξ)τ(\theta_{\xi})_{\tau} is −(θξ)τ~-(\theta_{\xi})_{\tilde{\tau}} and formula (1) of no. 14.3.4 takes the form

⟨θξ.u,v⟩+⟨u,θξ.v⟩=d(i(ξ)(u.v)). \langle \theta_{\xi}.u, v \rangle + \langle u, \theta_{\xi}.v \rangle = d(i(\xi)(u.v)).

14.4.2 (« Exterior differential »). For every integer p≥0p \geq 0 put

Ωp=Altp(T(X);KX). \Omega^p = \mathrm{Alt}^p(T(X); K_X).

There exists a differential operator D of type Ωp→Ωp+1\Omega^p \to \Omega^{p+1} and of order ≤1\leq 1 such that

DU(ω)=dω D_U(\omega) = d\omega

for every open set U of X and every ω∈SΩp(U)\omega \in \mathscr{S}_{\Omega^p}(U). Such a differential operator D is unique; we denote it by d (or dpd_p if we wish to specify the integer p).

Its symbol is the element of S1(Ωp,Ωp+1)S^1(\Omega^p, \Omega^{p+1}) obtained as follows: first of all we have

S1(Ωp,Ωp+1)=T(X)⊗L(Ωp;Ωp+1) S^1(\Omega^p, \Omega^{p+1}) = T(X) \otimes \mathscr{L}(\Omega^p; \Omega^{p+1})
=L(Ω1;L(Ωp;Ωp+1))=L(Ω1,Ωp;Ωp+1), = \mathscr{L}(\Omega^1; \mathscr{L}(\Omega^p; \Omega^{p+1})) = \mathscr{L}(\Omega^1, \Omega^p; \Omega^{p+1}),

where L(Ω1,Ωp;Ωp+1)\mathscr{L}(\Omega^1, \Omega^p; \Omega^{p+1}) denotes the bundle of bilinear mappings of Ω1×Ωp\Omega^1 \times \Omega^p into Ωp+1\Omega^{p+1}. Now the exterior product (α,β)↦α∧β(\alpha, \beta) \mapsto \alpha \wedge \beta defines a canonical section of this latter bundle; this section is the symbol of dpd_p.

To determine the transpose of dpd_p, we first remark that the exterior product defines a pairing Ωp×Ωn−p→Ωn=Ω\Omega^p \times \Omega^{n-p} \to \Omega^n = \Omega which permits identifying (Ωp)∗(\Omega^p)^* with Ωn−p\Omega^{n-p}. We identify (Ωp+1)∗(\Omega^{p+1})^* with Ωn−p−1\Omega^{n-p-1} in the same way. With these conventions, the transpose of dpd_p is (−1)p+1dn−p−1(-1)^{p+1} d_{n-p-1} and the corresponding Green operator, of type

(Ωp,Ωn−p−1)⟶Ω1=Ωn−1(\Omega^p,\Omega^{n-p-1})\longrightarrow\Omega_1=\Omega^{n-1}, is simply the exterior product (on each fibre).

14.4.3 (« Laplacian »).

Let K=RK=\mathbf{R}, r=∞r=\infty, X=RnX=\mathbf{R}^n, E=F=KXE=F=K_X; let ξi\xi^i (1≤i≤n)(1\leq i\leq n) denote the coordinate functions on XX; put Di=∂/∂ξiD_i=\partial/\partial\xi^i, and L=∑i=1nDi2L=\sum_{i=1}^n D_i^2. The operator LL is a scalar differential operator of order ≤2\leq 2. If we identify Ω\Omega with KXK_X by means of the frame ω=dξ1∧⋯∧dξn\omega=d\xi^1\wedge\cdots\wedge d\xi^n, we have E~=F~=KX\tilde E=\tilde F=K_X (cf. 14.3.3, example) and L~=L\tilde L=L.

For 1≤i≤n1\leq i\leq n, put ωi=(−1)i−1dξ1∧⋯∧dξi−1∧dξi+1∧⋯∧dξn\omega_i=(-1)^{i-1}d\xi^1\wedge\cdots\wedge d\xi^{i-1}\wedge d\xi^{i+1}\wedge\cdots\wedge d\xi^n. The ωi\omega_i form a frame of Ω1=Ωn−1\Omega_1=\Omega^{n-1}. If f∈C1(X)f\in\mathscr C^1(X), let grad⁡(f)\operatorname{grad}(f) denote the section ∑i=1nDi(f)ωi\sum_{i=1}^n D_i(f)\omega_i of Ω1\Omega_1. There exists a Green operator GG for LL such that

G(u,v)=v.grad⁡(u)−u.grad⁡(v)G(u,v)=v.\operatorname{grad}(u)-u.\operatorname{grad}(v)

for every open set UU of XX and u,vu,v in C1(U)\mathscr C^1(U).

In particular, let AA be a compact piece of Rn\mathbf{R}^n. Orient Rn\mathbf{R}^n, endow AA with the orientation induced by that of Rn\mathbf{R}^n, and ∂A\partial A with the corresponding orientation (11.2.1). We have

∫A(Lu).vω−∫Au.(Lv)ω=∫∂A(v.grad⁡(u)−u.grad⁡(v))\int_A(Lu).v\omega-\int_Au.(Lv)\omega=\int_{\partial A}(v.\operatorname{grad}(u)-u.\operatorname{grad}(v))

for u,vu,v in C1(X)\mathscr C^1(X).

[^1]: Care should be taken not to confuse M~\tilde{M} with the covering of M defined in 10.2.4 and denoted in the same way. [^2]: In fact, it is sufficient to verify the property in question for triples (ξλ,sλ,tλ)(\xi^\lambda,s^\lambda,t^\lambda) such that the domains of the ξλ\xi^\lambda cover XX. [^3]: When k=0k=0, it is agreed that this means that G=0G=0.

Machine translation of the English, by gpt-5-6, gpt-5-6-mini, 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.