8 Differential forms
Differential forms on manifolds modeled on locally convex spaces and the exterior derivative, following [ GN26 , Section 3.6 ] . Two features of the finite-dimensional theory are missing: there is no coordinate description of a form by components in a chart, and there are no partitions of unity, even on Banach manifolds. The definitions below only use local representatives, and all statements are local. No completeness of the model space or of the space of values is assumed.
Glöckner and Neeb define the exterior derivative through the Lie algebra differential of vector fields [ GN26 , Proposition 3.6.7, formula (3.29) ] and show that it is given in charts by the formula (3.33) below. We take (3.33) as the definition and check directly that it does not depend on the chart: the terms with second derivatives of the transition maps cancel because these are symmetric. The same kind of computation proves \(\mathrm{d}\circ \mathrm{d}= 0\), as indicated in [ GN26 , Remark 3.6.10 ] .
8.1 Local forms
Let \(E\) and \(F\) be Hausdorff locally convex spaces and \(U \subseteq E\) open.
A local \(F\)-valued \(n\)-form on \(U\) is a map \(\omega \colon U \times E^n \to F\) which is smooth in the sense of Michal–Bastiani and alternating \(n\)-linear in the last \(n\) variables.
For a local \(n\)-form \(\omega \),
where \(\partial _{v_i}\) is the directional derivative in the base point \(y\) [ GN26 , (3.33) ] .
The exterior derivative of a local \(n\)-form is a local \((n+1)\)-form.
Smoothness: each term is the derivative \(\mathrm{d}\omega \) composed with a continuous linear map. For fixed \(y\), the partial derivative \((u, w) \mapsto \partial _u \omega (y)(w)\) is linear in \(u\) and alternating \(n\)-linear in \(w\), because \(\omega (z)\) is so for every \(z\) near \(y\); then \(\mathrm{d}\omega (y)\) is its alternation, which is alternating \((n+1)\)-linear.
For a local form \(\omega \) on \(U\), \(\mathrm{d}(\mathrm{d}\omega ) = 0\) on \(U\).
\(\mathrm{d}(\mathrm{d}\omega )(y)(v_0, \dots , v_{n+1})\) is a double alternating sum of second derivatives \(\partial _{v_i}\partial _{v_j}\omega (y)\), which are symmetric in \(i, j\) (Schwarz’s theorem for Michal–Bastiani maps). The terms cancel in pairs [ GN26 , Remark 3.6.10 ] .
For a smooth map \(\tau \colon U \to V\) between open subsets of locally convex spaces and a local form \(\omega \) on \(V\), \((\tau ^*\omega )(y)(v_1, \dots , v_n) = \omega (\tau y)(\mathrm{d}\tau (y) v_1, \dots , \mathrm{d}\tau (y) v_n)\). It is a local form on \(U\).
\(\mathrm{d}(\tau ^*\omega ) = \tau ^*(\mathrm{d}\omega )\) on \(U\).
By the chain rule, \(\partial _u (\tau ^*\omega )(y)(w)\) is \(\partial _{\mathrm{d}\tau (y) u}\omega (\tau y)(\mathrm{d}\tau (y) w)\) plus the sum over \(k\) of \(\omega (\tau y)(\mathrm{d}\tau (y) w_1, \dots , \mathrm{d}^2\tau (y)(w_k, u), \dots , \mathrm{d}\tau (y) w_n)\); the second part uses that \(\omega (\tau y)\) is multilinear. In the alternating sum, the first terms give \(\tau ^*(\mathrm{d}\omega )\), and the terms with \(\mathrm{d}^2\tau (y)\) cancel because \(\mathrm{d}^2\tau (y)\) is symmetric.
8.2 Forms on manifolds
Let \(M\) be a smooth manifold modeled on \(E\). A tangent vector at \(x\) is given by its coordinates in the preferred chart at \(x\) (Definition 3.1).
An \(F\)-valued \(n\)-form \(\omega \) on \(M\) assigns to each \(x \in M\) an alternating \(n\)-linear map \(\omega _x \colon (T_xM)^n \to F\) such that, for every chart \(\varphi \) of the atlas, the local representative \((y, v_1, \dots , v_n) \mapsto \omega _{\varphi ^{-1} y}(T\varphi ^{-1} v_1, \dots , T\varphi ^{-1} v_n)\) is smooth on \(\varphi (U) \times E^n\) [ GN26 , Definition 3.6.1 ] . We write \(\Omega ^n(M, F)\) for the space of these forms.
The local representatives of a form are local forms. On chart overlaps they are related by pullback along the transition map: \(\omega _i = \tau ^*\omega _j\) for \(\tau = j \circ i^{-1}\).
The coordinate changes of tangent vectors are the derivatives of the transition maps, and they compose by the chain rule.
For \(\omega \in \Omega ^n(M, F)\), \((\mathrm{d}\omega )_x\) is the local exterior derivative of the local representative of \(\omega \) in the preferred chart at \(x\), evaluated at the image of \(x\).
In every chart of the atlas, the local representative of \(\mathrm{d}\omega \) is the local exterior derivative of the local representative of \(\omega \). In particular, \(\mathrm{d}\omega \) is a smooth \((n+1)\)-form [ GN26 , Proposition 3.6.7 ] .
Near a point, \(\omega _i = \tau ^*\omega _j\) with \(\tau \) the transition map from the chart \(i\) to the preferred chart \(j\) at the point; apply Theorem 8.6.
For every \(\omega \in \Omega ^n(M, F)\), \(\mathrm{d}(\mathrm{d}\omega ) = 0\) [ GN26 , Proposition 3.6.9 ] .
For a smooth function \(f\), \((\mathrm{d}f)_x(v) = \mathrm{d}(f \circ \varphi ^{-1})(\varphi x)\, v\), where \(\varphi \) is the preferred chart at \(x\): the exterior derivative of \(f\) is its differential [ GN26 , Remark 3.6.12(a) ] .
For \(n = 0\) the alternating sum of Definition 8.2 has a single term.
8.3 The vector space of forms
Sums and real multiples of forms are taken pointwise; their local representatives are the corresponding combinations of local representatives, hence smooth. This makes \(\Omega ^n(M, F)\) a real vector space.
The exterior derivative \(\mathrm{d}\colon \Omega ^n(M, F) \to \Omega ^{n+1}(M, F)\) is linear.
In the preferred chart at each point, directional derivatives are additive and homogeneous in the function.
8.4 Pullbacks
For a smooth map \(\varphi \colon N \to M\) and \(\omega \in \Omega ^n(M, F)\), \((\varphi ^*\omega )_x(v_1, \dots , v_n) = \omega _{\varphi (x)}(T_x\varphi \, v_1, \dots , T_x\varphi \, v_n)\) [ GN26 , Definition 3.6.6 ] . In charts, the local representative of \(\varphi ^*\omega \) is the local pullback of the local representative of \(\omega \) along the local representative of \(\varphi \); in particular \(\varphi ^*\omega \) is a smooth form.
For a smooth map \(\varphi \colon N \to M\) and \(\omega \in \Omega ^n(M, F)\), \(\mathrm{d}(\varphi ^*\omega ) = \varphi ^*(\mathrm{d}\omega )\) [ GN26 , Lemma 3.6.13 ] .
In the preferred charts at \(x\) and \(\varphi (x)\), this is Theorem 8.6 for the local representative of \(\varphi \).
The pullback \(\varphi ^*\) is linear, \(\mathrm{id}^*\omega = \omega \) and \(\psi ^*(\varphi ^*\omega ) = (\varphi \circ \psi )^*\omega \) [ GN26 , (3.27) ] .
The vectors in \(\varphi ^*\omega \) are images under the tangent map, so this is the chain rule \(T(\varphi \circ \psi ) = T\varphi \circ T\psi \) and \(T\mathrm{id}= \mathrm{id}\).
For a continuous linear map \(\alpha \colon F \to G\), \((\alpha _*\omega )_x = \alpha \circ \omega _x\) defines a \(G\)-valued form, and \(\alpha _*(\mathrm{d}\omega ) = \mathrm{d}(\alpha _*\omega )\) [ GN26 , Exercise 3.6.4 ] .
A continuous linear map commutes with directional derivatives.
8.5 The wedge product
Let \(\beta \colon E_1 \times E_2 \to E_3\) be a continuous bilinear map between locally convex spaces.
For \(\omega \in \Omega ^p(M, E_1)\) and \(\eta \in \Omega ^q(M, E_2)\),
[ GN26 , Definition 3.6.3 ] . It is alternating, being the alternation of a multilinear map, and its local representatives are finite sums of compositions of \(\beta \) with local representatives of \(\omega \) and \(\eta \), hence smooth.
Direct from the definitions; for \(p = q = 1\) the symmetric group has two elements.
For a smooth function \(f \colon M \to \mathbb {R}\), \((f\omega )_x = f(x)\, \omega _x\) defines a smooth form, and these products satisfy the module identities over \(C^\infty (M, \mathbb {R})\). The form \(f\omega \) is the wedge product of \(\omega \) with the \(0\)-form \(f\) for the scalar multiplication [ GN26 , Remark 3.6.5 ] .
Smoothness follows from the product rule for the scalar multiplication \(\mathbb {R}\times F \to F\). For the wedge product with a \(0\)-form, each permutation contributes \(\operatorname {sgn}(\sigma )^2 f(x)\, \omega _x(v) = f(x)\, \omega _x(v)\), and there are \(p!\) of them.
For \(\omega \in \Omega ^p(M, E_1)\) and \(\eta \in \Omega ^q(M, E_2)\),
(The degree \((p + 1) + q\) of \(\mathrm{d}\omega \wedge _\beta \eta \) is identified with \((p + q) + 1\).)
The rule is local. By the product rule for \(\beta \), the derivative of \(z \mapsto \omega (z) \wedge _\beta \eta (z)\) at \(y\) in the direction \(u\) is \(L(u) \wedge _\beta \eta (y) + \omega (y) \wedge _\beta L'(u)\), where \(L(u)\), \(L'(u)\) are the partial derivatives of the local representatives. By (3.33), it remains to prove two identities between alternating maps,
where \((\mathrm{Alt}L)(v_0, \dots , v_p) = \sum _i (-1)^i L(v_i)(\dots \hat v_i \dots )\). Write \(\mathrm{alt}f(v) = \sum _\sigma \operatorname {sgn}(\sigma ) f(v \circ \sigma )\) for any function \(f\) of \(k\) vectors. Then \(\mathrm{alt} f(v \circ \sigma ) = \operatorname {sgn}(\sigma )\, \mathrm{alt}f(v)\), \(\mathrm{alt} A = k!\, A\) for alternating \(A\), and alternating a partial alternation \(\sum _\tau \operatorname {sgn}(\tau ) g(w \circ \tilde\tau )\) over a subgroup of order \(a!\) gives \(a!\, \mathrm{alt}g\). The wedge product is \((p!\, q!)^{-1}\, \mathrm{alt}\) of \(w \mapsto \beta (f(w_1, \dots , w_p), g(w_{p+1}, \dots ))\), and \(\mathrm{Alt} L = (n!)^{-1} \mathrm{alt}\bigl(w \mapsto L(w_0)(w_1, \dots , w_n)\bigr)\), because its \(i\)-th term is a reindexing by the cycle \((0\, 1 \cdots i)\), of sign \((-1)^i\). With these rules both sides of each identity become \((p!\, q!)^{-1}\) times the alternation of the same function, up to reindexing; in the second identity the reindexing is the cycle \((0\, 1 \cdots p)\), which gives the sign \((-1)^p\).
8.6 Insertion and the Lie derivative
For a smooth vector field \(X\) and \(\omega \in \Omega ^{n+1}(M, F)\), \((i_X\omega )_x(v_1, \dots , v_n) = \omega _x(X(x), v_1, \dots , v_n)\) [ GN26 , Definition 3.6.15 ] . In every chart, its local representative is \((y, v) \mapsto \omega (y)(\tilde X(y), v)\).
We define \(L_X = \mathrm{d}\circ i_X + i_X \circ \mathrm{d}\), with \(i_X = 0\) on functions. Then \(L_X\) is a linear map of \(\Omega ^n(M, F)\), and Cartan’s formula [ GN26 , Proposition 3.6.16(c) ] holds by definition. Theorem 8.26 shows that this agrees with the definition [ GN26 , Definition 3.6.14 ] .
In every chart,
Differentiating \(z \mapsto \omega (z)(\tilde X(z), w)\) gives a partial derivative of \(\omega \) and the term \(\omega (y)(\mathrm{d}\tilde X(y)\, u, w)\). In \(\mathrm{d}(i_X\omega ) + i_X(\mathrm{d}\omega )\) the partial derivatives \(\partial _{v_j}\omega (y)(\tilde X(y), \dots )\) cancel, and moving \(\mathrm{d}\tilde X(y)\, v_j\) to the \(j\)-th place removes the signs.
For smooth vector fields \(X_0, \dots , X_{n+1}\) and \(\omega \in \Omega ^{n+1}(M, F)\),
[ GN26 , Proposition 3.6.7, (3.29) ] , and \((L_Y\omega )(X_1, \dots , X_n) = Y.\omega (X_1, \dots , X_n) - \sum _j \omega (X_1, \dots , [Y, X_j], \dots , X_n)\) [ GN26 , (3.35) ] .
In a chart, \(X_i.\omega (\dots , \widehat{X_i}, \dots )\) is a partial derivative of \(\omega \) plus the terms with \(\mathrm{d}\tilde X_k \cdot \tilde X_i\); the partial derivatives add up to \(\mathrm{d}\omega \), and the other terms regroup in pairs into the brackets. The formula for \(L_Y\) follows in the same way from Theorem 8.26.
For smooth vector fields \(X\) and \(Y\) [ GN26 , Proposition 3.6.16 ] :
\([L_X, L_Y] = L_{[X, Y]}\);
\([L_X, i_Y] = i_{[X, Y]}\);
\(L_X \circ \mathrm{d}= \mathrm{d}\circ L_X\);
\(L_X\) maps closed forms to exact forms.
(b) is a computation in a chart with Theorem 8.26: the difference is \(\omega (y)(\mathrm{d}\tilde Y(y)\tilde X(y) - \mathrm{d}\tilde X(y)\tilde Y(y), \dots )\), and the first entry is the local representative of \([X, Y]\). (d) and (e) follow from Cartan’s formula and \(\mathrm{d}\circ \mathrm{d}= 0\), and (a) follows algebraically from (b), (d) and Cartan’s formula. Glöckner and Neeb deduce (a)–(d) from the corresponding identities for Lie algebra cochains.
The first identity holds because \(T\varphi \) maps \(X(x)\) to \(Y(\varphi (x))\); the second follows from it, Cartan’s formula and Theorem 8.17.
8.7 De Rham cohomology
Let \(Z^n_{dR}(M, F) = \ker \mathrm{d}\) and \(B^n_{dR}(M, F) = \mathrm{d}(\Omega ^{n-1}(M, F))\), with \(B^0_{dR}(M, F) = 0\). Then \(B^n_{dR} \subseteq Z^n_{dR}\), and the de Rham cohomology is \(H^n_{dR}(M, F) = Z^n_{dR}(M, F) / B^n_{dR}(M, F)\) [ GN26 , Definition 3.6.11 ] .
A function is closed if and only if it is locally constant; hence \(H^0_{dR}(M, F)\) is the space of locally constant functions \(M \to F\) [ GN26 , Remark 3.6.12(b) ] .
In a chart, a \(C^1\) map with vanishing derivative is constant on convex open sets, and every point of a locally convex space has a convex open neighbourhood in a given open set.
A smooth map \(\varphi \colon N \to M\) induces linear maps \(\varphi ^* \colon H^n_{dR}(M, F) \to H^n_{dR}(N, F)\), \([\alpha ] \mapsto [\varphi ^*\alpha ]\), with \(\mathrm{id}^* = \mathrm{id}\) and \((\varphi \circ \psi )^* = \psi ^* \circ \varphi ^*\) [ GN26 , Section 3.6 ] .
8.8 The Poincaré Lemma
Homotopies are smooth maps \(H \colon \mathbb {R}\times M \to N\), with \(H_t = H(t, \cdot )\); on \(\mathbb {R}\times M\) we use the product charts \(\mathrm{id}_{\mathbb {R}} \times \varphi \). Glöckner and Neeb use \([0, 1] \times M\), a manifold with boundary; a homotopy on \([0, 1] \times M\) can be precomposed with a smooth map \(\mathbb {R}\to [0, 1]\) which is \(0\) near \((-\infty , 0]\) and \(1\) near \([1, \infty )\).
Let \(F\) be Mackey complete. For \(\omega \in \Omega ^{k+1}(\mathbb {R}\times M, F)\), the fiber integral \(I(\omega ) \in \Omega ^k(M, F)\) is
(weak integral; in a chart, the integrand is a smooth curve, whose weak integral exists because \(F\) is Mackey complete). In the chart \(\varphi \), the local representative of \(I(\omega )\) is the same integral of the local representative of \(\omega \) in the chart \(\mathrm{id}\times \varphi \), so \(I(\omega )\) is smooth by the smoothness of parameter-dependent integrals.
With \(i_t(x) = (t, x)\), \(\mathrm{d}I(\omega ) + I(\mathrm{d}\omega ) = i_1^*\omega - i_0^*\omega \) for \(\omega \in \Omega ^{k+1}(\mathbb {R}\times M, F)\), and \(I(\mathrm{d}f) = i_1^* f - i_0^* f\) for functions \(f\).
Glöckner and Neeb use Cartan’s formula. We compute in a chart. Differentiating under the integral sign (which requires \(C^{2}\) integrands in Mackey complete spaces), \(\mathrm{d}I(\omega )(v_0, \dots , v_k) = \sum _j (-1)^j \int _0^1 \partial _{(0, v_j)} \omega _{(t,x)}((1,0), (0, v_0), \dots , \widehat{(0, v_j)}, \dots )\, dt\). In the formula (3.33) for \((\mathrm{d}\omega )_{(t,x)}((1, 0), (0, v_0), \dots )\), the terms with \(j \geq 1\) are the same with the opposite sign, and the remaining term is \(\partial _t \omega _{(t,x)}((0, v_0), \dots )\). By the fundamental theorem of calculus, its integral is \(\omega _{(1, x)}(\dots ) - \omega _{(0, x)}(\dots ) = (i_1^*\omega - i_0^*\omega )_x(v)\).
Let \(H \colon \mathbb {R}\times M \to N\) be smooth and \(F\) Mackey complete. For a closed \(\alpha \in \Omega ^{k+1}(N, F)\), \(H_1^*\alpha - H_0^*\alpha = \mathrm{d}I(H^*\alpha )\), and \(H_1^* f = H_0^* f\) for closed functions \(f\). Hence \(H_0^* = H_1^* \colon H^n_{dR}(N, F) \to H^n_{dR}(M, F)\) for all \(n\).
Apply the homotopy formula to \(\omega = H^*\alpha \), for which \(\mathrm{d}\omega = H^*\mathrm{d}\alpha = 0\), and use \(i_t^* H^* = (H \circ i_t)^* = H_t^*\).
Let \(M\) be smoothly contractible: there are \(p \in M\) and a smooth \(H \colon \mathbb {R}\times M \to M\) with \(H(0, x) = x\) and \(H(1, x) = p\). Then every closed form of positive degree on \(M\) with values in a Mackey complete space is exact, i.e. \(H^n_{dR}(M, F) = 0\) for \(n \geq 1\).
\(H_0 = \mathrm{id}_M\) and \(H_1\) is constant, so \(H_1^*\alpha = 0\) for \(\deg \alpha \geq 1\). By the Homotopy Lemma, \(-\alpha = \mathrm{d}I(H^*\alpha )\).
Let \(U \subseteq E\) be open and star-shaped with respect to \(p \in U\). Every closed form of positive degree on \(U\) with values in a Mackey complete space is exact.
Open subsets of manifolds are manifolds, because restrictions of smooth diffeomorphisms to open sets are smooth diffeomorphisms. With \(c(t) = (1 - t)^2/((1 - t)^2 + t^2)\), which is smooth with values in \([0, 1]\), \(c(0) = 1\) and \(c(1) = 0\), the map \(H(t, x) = p + c(t)(x - p)\) contracts \(U\) to \(p\).
Let \(G\) be a locally convex space and \(F\) a Mackey complete one. The projection \(p \colon M \times G \to M\) induces isomorphisms \(p^* \colon H^n_{dR}(M, F) \to H^n_{dR}(M \times G, F)\), with inverse \(i_0^*\), where \(i_0(m) = (m, 0)\).
\(p \circ i_0 = \mathrm{id}_M\), and \(H(t, (m, x)) = (m, t x)\) is a smooth homotopy from \(i_0 \circ p\) to \(\mathrm{id}_{M \times G}\), so \(p^* i_0^* = (i_0 \circ p)^* = \mathrm{id}\) by the Homotopy Lemma.
Let \(X\) be a smooth vector field with a local \(C^2\)-flow \(\mathrm{Fl}\). Then \(\frac{d}{dt}\big|_{t=0} ((\mathrm{Fl}_t)^*\omega )_x(v_1, \dots , v_n) = (L_X\omega )_x(v_1, \dots , v_n)\).
In a chart, differentiate \(t \mapsto \omega (\varphi _t(y))(\mathrm{d}\varphi _t(y)v_1, \dots )\), where \(\varphi \) is the local flow: the base point moves with velocity \(X(y)\), and \(\partial _t \mathrm{d}\varphi _t(y) v = \mathrm{d}X(y) v\) at \(t = 0\) by exchanging the order of differentiation in \(\partial _t \varphi = X \circ \varphi \). This is the local formula for \(L_X\). Glöckner and Neeb assume \(C^1\)-flows; we do not claim this argument is new.