Manifolds modeled on locally convex spaces

14 Local additions and manifolds of mappings

Glöckner and Neeb build the manifold structure of \(C^\ell (M, N)\), for \(M\) compact and \(N\) with a local addition, in [ GN26 , § 3.8, § 4.1, § 4.2 ] . This chapter follows them: local additions and the ways to produce them (open subsets, products, normalization, ambient formulas on submanifolds, real analytic local additions), the manifolds \(C(K, N)\) and \(C^\ell (M, N)\) with their exponential law, evaluation maps and tangent spaces, and the compact-open \(C^\ell \) topology, described both through iterated tangent maps [ GN26 , Def. 4.1.2 ] and through jets on compact sets in charts. The models of \(C^\ell (M, N)\) vary with the point, so these manifolds are atlases with varying models. As in the rest of the library, no completeness of the models is assumed.

14.1 Local additions

A local addition on \(M\) is a smooth map \(\Sigma \colon \Omega \to M\) on an open subset \(\Omega \subseteq TM\) containing the zero section, with \(\Sigma (0_x) = x\), such that \(\theta = (\pi , \Sigma ) \colon \Omega \to M \times M\) is a smooth diffeomorphism onto an open subset of \(M \times M\). It is normalized if \(T_{0_x}(\Sigma |_{T_x M}) = \mathrm{id}\) for all \(x\); in the preferred chart \(\varphi _x\) at \(x\), this reads \(\mathrm{d}(\varphi _x \circ \Sigma (x, \cdot ))(0)(w) = w\). In Lean the data is the map \(\theta \), an open partial homeomorphism \(TM \to M \times M\) with source \(\Omega \), smooth with smooth inverse, whose first component is the bundle projection.

Let \(U\) be an open subset of \(E\). Then \(\Sigma (x, v) = x + v\) on \(\Omega = \{ (x, v) \in TU : x + v \in U\} \) is a normalized local addition on \(U\), and \(\theta (x, v) = (x, x + v)\) maps \(\Omega \) onto \(U \times U\).

Proof ▶

The charts of \(U\) are the inclusion, so \(TU = U \times E\) with the identity as coordinate changes. In these coordinates \(\theta \) and \(\theta ^{-1}(x, y) = (x, y - x)\) are restrictions of continuous linear maps.

Proof ▶

In the preferred chart \(\varphi \) at \(x\), the map \(v \mapsto \varphi (\Sigma (x, v))\) has the local inverse \(z \mapsto \theta ^{-1}(x, \varphi ^{-1} z)\) read in the tangent chart, so \(\alpha _x\) is invertible by the chain rule. In an arbitrary chart, \(\alpha _x\) and \(\alpha _x^{-1}\) are conjugated by the derivative of the transition map and are partial derivatives of the chart representatives of \(\theta \) and \(\theta ^{-1}\); hence \(F\) and \(F^{-1}\) are smooth. Glöckner and Neeb argue instead that \(\theta \) restricts to diffeomorphisms of the fibres; we do not claim this argument is new.

Proof ▶

The coordinate changes of \(T(M_1 \times M_2)\) between product charts are the products of the coordinate changes of \(TM_1\) and \(TM_2\). Hence, in charts, the inverse map \((v_1, v_2) \mapsto (v_1, v_2)\) is the regrouping \(((a, b), (c, d)) \mapsto ((a, c), (b, d))\) of coordinates, which is continuous linear; the map itself is smooth by Theorem 3.5.

If \(\Sigma _1\) and \(\Sigma _2\) are local additions on \(M_1\) and \(M_2\), then, identifying \(T(M_1 \times M_2)\) with \(TM_1 \times TM_2\), \(\Sigma _1 \times \Sigma _2\) is a local addition on \(M_1 \times M_2\).

Proof ▶

The map \(\theta \) is \(\theta _1 \times \theta _2\) composed with the identification of Lemma 14.4 and the regrouping \((M_1 \times M_1) \times (M_2 \times M_2) \cong (M_1 \times M_2) \times (M_1 \times M_2)\), all of which are smooth with smooth inverses.

Let \(X \subseteq F\) be an embedded submanifold modeled on a closed subspace \(S\). Let \(\sigma \colon \Omega \to F\) and \(\tau \colon V \to F\) be smooth on open subsets of \(F \times F\) such that, for \(p \in X\), \((p, 0) \in \Omega \), \(\sigma (p, 0) = p\), and \(w \mapsto \sigma (p, w)\) is a bijection from \(\{ w \in T_pX : (p, w) \in \Omega \} \) onto \(\{ q \in X : (p, q) \in V\} \) with inverse \(q \mapsto \tau (p, q)\). Then \(\Sigma (v) = \sigma (\pi v, v)\), with \(v\) read in \(F\), is a local addition on \(X\), normalized if \(\mathrm{d}\sigma (p, \cdot )(0) = \mathrm{id}\) on \(T_pX\).

Proof ▶

A map into \(X\) is smooth when it is smooth into \(F\): in a slice chart \(\varphi \) its local representative is the corestriction to \(S\) of \(\varphi \) composed with the map into \(F\). In the same chart, the fiber coordinate of a tangent vector \(v\) at \(p\) is \(\mathrm{d}\varphi (p) w\), where \(w \in T_pX \subseteq F\) is its ambient vector; so the ambient map \(TX \to F \times F\), \(v \mapsto (p, w)\), is smooth, and a map into \(TX\) is smooth when its composition with the ambient map is. Then \(\theta = (\pi , \Sigma )\) and \(\theta ^{-1}(p, q) = (p, \tau (p, q))\) are smooth. Glöckner and Neeb obtain local additions from Lie group structures and from sprays [ GN26 , Prop. 3.8.12, Prop. 3.8.13 ] ; we do not claim this argument is new.

A continuous map \(f \colon X \to Y\) into a Hausdorff space, injective on a compact set \(K\) and locally injective at the points of \(K\), is injective on an open neighborhood of \(K\). Hence a local \(C^k\)-diffeomorphism (\(k \in \mathbb {N}\) or \(k = \infty \)) into a Hausdorff manifold that is injective on a compact set \(K\) is a \(C^k\)-diffeomorphism from an open neighborhood of \(K\) onto an open set.

Proof ▶

The set \(\{ (x, y) : f(x) = f(y) \Rightarrow x = y\} \) is a neighborhood of each point of \(K \times K\): of \((x, x)\) by local injectivity, of \((x, y)\) with \(x \ne y\) because \(f(x) \ne f(y)\) have disjoint neighborhoods. By the Wallace lemma it contains \(A \times B\) for neighborhoods \(A\), \(B\) of \(K\), and \(f\) is injective on the interior of \(A \cap B\). Glöckner and Neeb argue with nets; we do not claim this argument is new. For the second statement, the restriction is open and continuous, hence a homeomorphism onto its open image, and its inverse agrees near each point with a local inverse of \(f\). The real analytic case is Corollary 14.11.

Let \(f \colon X \to Y\) be continuous, \(Y\) Hausdorff with every point having a compact neighborhood, and \(A \subseteq X\) \(\sigma \)-compact such that \(f(A)\) is closed, \(f|_A\) is an embedding and \(f\) is locally injective at the points of \(A\). Then \(f\) is injective on an open neighborhood of \(A\). Hence a local \(C^k\)-diffeomorphism (\(k \in \mathbb {N}\) or \(k = \infty \)) into a Hausdorff, locally compact manifold (for instance a finite-dimensional one) which restricts to such an embedding on \(A\) is a \(C^k\)-diffeomorphism from an open neighborhood of \(A\) onto an open set.

Proof ▶

Take a compact exhaustion \(\emptyset = C_0 \subseteq C_1 \subseteq \cdots \) of a neighborhood of \(f(A)\) with \(C_n \subseteq \operatorname {int} C_{n+1}\). The “annuli” \(O_j = \operatorname {int} C_{j+1} \setminus C_{j-1}\) meet only for \(|i - j| \le 1\). Let \(V_j\) be an open neighborhood of the compact set \(A \cap f^{-1}(C_{j+1})\) on which \(f\) is injective (Proposition 14.7); then \(f\) is injective on \(U = \bigcup _j f^{-1}(O_j) \cap V_j \cap V_{j+1}\). Glöckner and Neeb use their Lemma A.5.5 instead of the annuli, and the Hausdorff property of the domain, which is not needed here. The statement of [ GN26 , Cor. 3.8.21 ] concludes \(K \subseteq U\), a misprint for \(A \subseteq U\); its finite-dimensionality hypothesis is only used through local compactness of the target. We do not claim this argument is new.

14.2 Real analytic local additions

Real analyticity is meant in Milnor’s sense (Chapter 15): local restrictions of complex analytic maps on complexifications, with no completeness of the model.

If \(M\) is a real analytic manifold modeled on \(E\), its tangent bundle \(TM\) is a real analytic manifold modeled on \(E \times E\). Products of real analytic manifolds are real analytic manifolds, and the projections and pairings of real analytic maps are real analytic.

Proof ▶

The transition maps of \(TM\) are \((y, w) \mapsto (\tau (y), \mathrm{d}\tau (y)\, w)\) for transition maps \(\tau \) of \(M\), and the differential of a real analytic map is real analytic: if \(g\) is a complex analytic extension of \(\tau \), then \(\mathrm{d}g\) is one of \(\mathrm{d}\tau \) (lc-analysis). Products of real analytic maps are real analytic because products of complex analytic extensions are complex analytic extensions.

A local addition \(\Sigma \) on a real analytic manifold \(M\) is real analytic if \(\theta = (\pi , \Sigma )\) is a real analytic diffeomorphism onto its open image. Then \(\Sigma \) is real analytic.

A real analytic manifold is a smooth manifold, so a local addition on it makes sense, and being real analytic is a property of a smooth local addition. The real analytic and the smooth maximal atlases differ; the charted space structure of \(TM\) is built from a real analytic atlas of \(M\).

Real analyticity of maps between real analytic manifolds is a local property. Hence Propositions 14.7 and 14.8 hold for real analytic local diffeomorphisms.

Proof ▶

Locality: real analyticity is checked in charts, and on chart domains it is local in the model [ GN26 , Lemma 2.2.3 ] . The rest is the topological part of the two propositions.

Let \(S \subseteq F\) be a complemented subspace, with a continuous linear projection \(P \colon F \to S\). A subset \(X \subseteq F\) is a split real analytic submanifold modeled on \(S\) if every point of \(X\) lies in a real analytic slice chart: a real analytic diffeomorphism \(\varphi \) between open subsets of \(F\) with \(\varphi (U \cap X) = \varphi (U) \cap S\). The restrictions of these charts form an atlas of \(X\). Such an \(X\) is an embedded submanifold.

A split real analytic submanifold \(X \subseteq F\) is a real analytic manifold modeled on \(S\). A map into \(X\) is real analytic if and only if it is real analytic into \(F\). The ambient map \(TX \to F \times F\), \(v \mapsto (\pi v, w)\), with \(w \in T_p X \subseteq F\) the ambient vector of \(v\), is real analytic, and a map into \(TX\) is real analytic as soon as its composition with the ambient map is.

Proof ▶

As for smooth embedded submanifolds (Proposition 14.6), with real analytic maps in place of smooth ones. The only new point is corestriction: a map with values in \(S\) which is real analytic into \(F\) is real analytic into \(S\) because it equals its composition with \(P\).

Remark 14.14 Why complemented
#

For smooth maps, a map with values in a closed subspace \(S\) is smooth into \(S\). For real analytic maps in Milnor’s sense this is less clear: a complex analytic extension to \(D_{\mathbb {C}}\) of a map with values in \(S\) need not take values in \(S_{\mathbb {C}}\). Assuming \(S\) complemented avoids the issue. In finite dimensions, and for closed subspaces of Hilbert spaces, this is no restriction; for general locally convex (even Banach) spaces it is. The real analytic atlas differs from the atlas of all smooth slice charts, so the two tangent bundles of \(X\) are different types in Lean, although the tangent spaces \(T_pX \subseteq F\) agree.

Let \(X \subseteq F\) be a split real analytic submanifold, and \(\sigma \), \(\tau \) real analytic maps on open subsets of \(F \times F\) satisfying the hypotheses of Proposition 14.6. Then \(\Sigma (v) = \sigma (\pi v, v)\) is a real analytic local addition on \(X\).

Proof ▶

By Theorem 14.13, \(\theta = (\pi , \Sigma )\) and \(\theta ^{-1}(p, q) = (p, \tau (p, q))\) are real analytic, because their compositions with the inclusion and with the ambient map are. Glöckner and Neeb do not discuss ambient formulas; we do not claim this argument is new.

14.3 Manifolds of continuous maps

Let \(K\) be a compact Hausdorff space and \(N\) a manifold with a local addition \(\Sigma \colon \Omega \to N\), \(\theta = (\pi , \Sigma )\).

For \(f \in C(K, N)\), \(\Gamma _f = \{ \tau \in C(K, TN) : \pi \circ \tau = f\} \), with the pointwise linear structure and the initial topology with respect to the fiber coordinates \(\tau \mapsto (x \mapsto \tau (x) \text{ in the chart } i) \in C(L, E)\), for charts \(i\) of \(N\) and compact \(L \subseteq K\) with \(f(L)\) in the domain of \(i\). It is a Hausdorff locally convex space, its topology is the one induced by \(C(K, TN)\), and the fiber coordinates embed it as a closed subspace of a product.

The map \(\varphi _f \colon O'_f \to O_f\), \(g \mapsto \theta ^{-1} \circ (f, g)\), from \(O'_f = \{ g : (f, g)(K) \subseteq \theta (\Omega )\} \) (open in the compact-open topology) onto \(O_f = \Gamma _f \cap C(K, \Omega )\) (open in \(\Gamma _f\)) is a homeomorphism, with inverse \(\tau \mapsto \Sigma \circ \tau \).

Proof ▶

Postcomposition with maps that are continuous on open sets is continuous in the compact-open topology (\(K\) is Hausdorff), and \(\Gamma _f\) carries the topology induced by \(C(K, TN)\).

Proof ▶

A map into \(\Gamma _g\) is smooth if its fiber coordinates are, because these form a closed linear embedding into a product. The coordinates over a compact set are determined by those over the pieces of a finite cover on which \(f\) and \(g\) lie in charts \(i\) and \(j\) of \(N\), again through a closed embedding. Over such a piece, the transition map \(\varphi _g \circ \varphi _f^{-1}\) is the superposition operator \(\gamma \mapsto h \circ \gamma \) of the smooth map \(h(y, z, v) = \theta ^{-1}(j^{-1} z, \Sigma (i^{-1} y, v))\) (in the chart \(j\)), applied to a curve that depends affinely on the section; superposition operators are smooth by [ GN26 , Prop. 1.7.1 ] (formalized in lc-analysis). Glöckner and Neeb deduce the smoothness from the exponential law for sections instead.

14.4 Manifolds of \(C^\ell \) maps

Let \(M\) be a compact smooth manifold, \(N\) a manifold with a local addition \(\Sigma \colon \Omega \to N\), \(\theta = (\pi , \Sigma )\), and \(\ell \in \mathbb {N}_0 \cup \{ \infty \} \).

Lemma 14.19 Compact manifolds are finite-dimensional
✓

If \(M\) is nonempty and compact, its model is locally compact, hence finite-dimensional by Riesz’s theorem; so \(M\) is modeled on some \(\mathbb {R}^n\), and only \(N\) may be infinite-dimensional. The results below use only the local compactness of the model, inside the proofs (the case \(M = \emptyset \), where \(C^\ell (M, N)\) is a point, is treated separately), so that their statements carry only the hypothesis that \(M\) is compact.

Proof ▶

A chart sends a compact neighborhood of a point onto a compact neighborhood in the model, and a topological group with a compact neighborhood is locally compact.

Let \(Z\) be locally compact, \(U \subseteq Z\) open, \(r \in \mathbb {N}_0\), and \(h \colon E \to F\) smooth on an open set \(O\). If \(g \colon X \to C^r(U, E)\) is \(C^k\) on an open set \(W\) and \(g(w)(U) \subseteq O\) for \(w \in W\), then \(w \mapsto h \circ g(w) \in C^r(U, F)\) is \(C^k\) on \(W\).

Proof ▶

Induction on \(r\). Since \(C^r(U, F)\) is a closed subspace of \(\prod _{j \le r} C(U \times Z^j, F)\) [ GN26 , Lemma 1.7.15 ] , it suffices to treat each \(d^{(j)}\). For \(j = 0\), restriction to the compact subsets of \(U\) (a closed embedding, \(Z\) being locally compact) reduces the claim to superposition operators on \(C(C, E)\), \(C\) compact, which are smooth [ GN26 , Prop. 1.7.1 ] . For \(j \ge 1\), \(d^{(j)}(h \circ \gamma )\) depends continuously and linearly on \(d(h \circ \gamma ) = dh \circ (\gamma \circ \mathrm{pr}_1, d\gamma )\), a superposition by \(dh\) on \(C^{r-1}(U \times Z, E \times E)\), and the induction hypothesis applies. Glöckner and Neeb obtain the smoothness they need from the exponential law [ GN26 , Prop. 4.1.34 ] instead.

For an open cover \((V_a)\) of \(U\), the restrictions \(C^r(U, F) \to \prod _a C^r(V_a, F)\) form a linear topological embedding with closed range (compare [ GN26 , Lemma 1.7.43, Lemma 4.1.9 ] ).

For charts \(j\) of \(M\) and \(i\) of \(N\), a compact set \(K\) in the target of \(j\) (no interior is required) and \(m \le \ell \), the \(m\)-jet on \(K\) of \(g \in C^\ell (M, N)\) is \(\mathrm{d}^{(m)}(i \circ g \circ j^{-1})|_{K \times H^m} \in C(K \times H^m, E)\), defined on the open set of maps with \(g(j^{-1}(K))\) in the domain of \(i\). The compact-open \(C^\ell \) topology of \(C^\ell (M, N)\), or jet topology, is the coarsest topology finer than the compact-open topology of \(C(M, N)\) that makes all the jets continuous, with the compact-open topology on \(C(K \times H^m, E)\). A map into \(C^\ell (M, N)\) is continuous if and only if it is continuous into \(C(M, N)\) and its jets are continuous where defined. For \(\ell = \infty \), \(C^\infty (M, N)\) carries the initial topology with respect to the inclusions into the \(C^\ell (M, N)\) [ GN26 , Remark 4.1.3(b) ] . This definition makes sense for every manifold \(M\) (compact or not) modeled on any Hausdorff locally convex space \(H\); for Hausdorff \(M\) it is the topology of [ GN26 , Def. 4.1.2 ] (Theorem 14.41).

This is the weak (compact-open) \(C^r\) topology of [ Hir76 , Ch. 2, § 1, pp. 34–35 ] , with the uniform bounds on the derivatives over \(K\) replaced by the compact-open topology of \(C(K \times H^m, E)\), as in the compact-open \(C^k\) topology of [ GN26 , Def. 1.7.10 ] for maps on open sets.

The chart topology of \(C^\ell (M, N)\) is the coarsest topology finer than the compact-open topology for which the local representatives \(g \mapsto i \circ g \circ j^{-1} \in C^\ell (V, E)\) are continuous on the open sets \(\{ g : g(j^{-1}(L)) \subseteq \mathrm{dom}\, i\} \), for charts \(j\) of \(M\), \(i\) of \(N\), compact \(L\) and open \(V \subseteq L\); for \(\ell = \infty \), the initial topology with respect to the finite orders. It is not the topology of \(C^\ell (M, N)\): if the model of \(M\) is not locally compact, compact sets have empty interior, every such \(V\) is empty, and the chart topology collapses to the compact-open \(C^0\) topology.

Proof ▶

On a compact set \(K \subseteq V\) the jets of the local representative on \(V\) are the jets of \(g\) on \(K\), and the compact-open topology of \(C(V \times H^m, E)\) is determined by the compact subsets of \(V \times H^m\); this gives the first inclusion. Conversely, if the model is locally compact, a compact set \(K\) in the open domain of \(i \circ g_0 \circ j^{-1}\) lies in the interior \(V\) of a compact set \(L\) in that domain; for \(g\) near \(g_0\) the jets of \(g\) on \(K\) are restrictions of the differentials of the local representative on \(V\). A nonempty compact manifold has a locally compact model (Lemma 14.19).

The space \(\Gamma _f\) of \(C^\ell \) sections of \(TN\) along \(f\) carries the initial topology with respect to the local representatives in charts; it is a Hausdorff locally convex space, embedded as a closed subspace of a product of spaces \(C^\ell (D, E)\). For \(\ell = \infty \) it carries the initial topology with respect to the finite orders [ GN26 , Remark 4.1.3(b) ] . Glöckner and Neeb define this topology through \(C^\ell (M, TN)\); the chart description used here corresponds to their Definition 4.1.16, Proposition 4.1.26 and the proof of Proposition 4.1.34.

Proof ▶

Near each point, \(f\) and \(g\) lie in charts \(i\) and \(j\) of \(N\), and the local representative of \(\varphi _g \circ \varphi _f^{-1}\) is the superposition operator of the smooth map \((y, z, v) \mapsto \theta ^{-1}(j^{-1} z, \Sigma (i^{-1} y, v))\) applied to \((i \circ f, j \circ g, \tau )\), which depends affinely on \(\tau \) (Proposition 14.20); Lemma 14.21 glues the local statements. The continuity of the charts is proved the same way, over compact neighborhoods in charts of \(M\), in the chart topology, which is the topology of \(C^\ell (M, N)\) for compact \(M\) (Proposition 14.24). The case \(\ell = \infty \) follows from the finite orders.

14.5 Smooth maps between atlases with varying models

A map \(f \colon X \to Y\) between spaces with atlases with varying models is \(C^k\) on an open set \(U\) if it is continuous on \(U\) and all its local representatives \(\psi _j \circ f \circ \varphi _i^{-1}\) are \(C^k\) on their (open) domains. A manifold modeled on a single space \(G\) is an atlas with varying models through all the charts of its atlas.

Being \(C^k\) can be checked in one pair of charts around each point; \(C^k\) maps compose, the identity is smooth, and restrictions of \(C^k\) maps to open subsets are \(C^k\). For manifolds with a single model these are the \(C^k\) maps of Definition 2.5. Charts are smooth, with smooth inverses.

Proof ▶

Near each point, a local representative in arbitrary charts is the composition of the given one with transition maps, which are smooth.

Two atlases with varying models on the same space define the same smooth structure (the identity is smooth in both directions) if and only if all the transition maps between their charts are smooth, that is, if and only if their union is an atlas.

Proof ▶

The local representatives of the identity are the transition maps.

The tangent space at \(x\) is the model of the preferred chart at \(x\); the derivatives of the transition maps at the point are isomorphisms of topological vector spaces satisfying the cocycle condition. The tangent map \(T_x f\) of a \(C^1\) map is the derivative of its local representative in the preferred charts; in other charts it is conjugated by the changes of tangent coordinates. (In infinite dimensions this kinematic tangent space need not agree with the space of derivations.)

14.6 \(C^{k,\ell }\) maps on products of manifolds

Being \(C^{k,\ell }\) on products of open sets is a local property. If \(f\) is \(C^{k,\ell }\) on \(U \times V\), \(\alpha \) is \(C^k\), \(\beta \) is \(C^\ell \) and \(\gamma \) is smooth (with the obvious conditions on the images), and the second factor of the domain of \(\beta \) is locally compact, then \(\gamma \circ f \circ (\alpha \times \beta )\) is \(C^{k,\ell }\) (compare [ GN26 , Props. 1.7.26, 1.7.28 ] ).

Proof ▶

By the exponential law [ GN26 , Thm. 1.7.30 ] , \(x \mapsto f(x, \cdot )\) is \(C^k\) into \(C^\ell (V, F)\); pulling back by \(\beta \) is continuous linear [ GN26 , Prop. 1.7.11 ] , superposition by \(\gamma \) is smooth (Proposition 14.20), and uncurrying needs local compactness. Glöckner and Neeb use Faà di Bruno’s formula instead.

Let \(L\) carry an atlas with varying models (for instance a manifold) [ GN26 , Remark 3.5.28 ] , and \(M\), \(N\) be manifolds without boundary (manifolds with rough boundary are out of scope). A map \(f \colon L \times M \to N\) is \(C^{k,\ell }\) if around each point there are charts \(e\), \(e'\), \(c\) and open sets \(U \ni e(x)\), \(V \ni e'(y)\) with \(f(e^{-1}U \times e'^{-1} V) \subseteq \mathrm{dom}\, c\) and \(c \circ f \circ (e^{-1} \times e'^{-1})\) of class \(C^{k,\ell }\) on \(U \times V\). Such maps are continuous, and, if the model of \(M\) is locally compact, their local representatives on all such boxes, in all charts, are \(C^{k,\ell }\) [ GN26 , Remark 3.5.26(a) ] (this holds when \(M\) is compact, Lemma 14.19).

14.7 The exponential law for manifolds of mappings

Let \(M\) be compact (without boundary) and \(N\) admit a local addition. For every space \(L\) with an atlas with varying models and every \(k \in \mathbb {N}_0\), a map \(g \colon L \to C^\ell (M, N)\) is \(C^k\) if and only if \(g^\wedge \colon L \times M \to N\), \((x, y) \mapsto g(x)(y)\), is \(C^{k,\ell }\) (for \(\ell = \infty \): \(C^{k,\ell '}\) for every \(\ell ' \in \mathbb {N}_0\)).

Proof ▶

Locally, \(g^\wedge \) reads \((a, b) \mapsto c(\Sigma (\tau (a)(b)))\) with \(\tau = \varphi _f \circ g\); as a function of \(a\) this is a superposition operator applied to a \(C^k\) curve in a space of \(C^\ell \) maps, and uncurrying gives \(C^{k,\ell }\). Conversely, the local representatives of \(\varphi _f \circ g\) are superposition operators applied to the curried local representatives of \(g^\wedge \), over a finite cover of \(M\) by compact sets on which \(g(x)\) stays in single charts of \(N\). Glöckner and Neeb argue through the exponential law for sections [ GN26 , Prop. 4.1.34 ] . (Part (b) of [ GN26 , Thm. 4.2.2 ] holds by construction for the chart description of the compact-open \(C^\ell \) topology; its agreement with [ GN26 , Def. 4.1.2 ] is Theorem 14.39.)

The evaluation map \(\varepsilon \colon C^\ell (M, N) \times M \to N\) is \(C^{k,\ell }\) for every \(k\), and the point evaluations \(\varepsilon _x\) are smooth.

Proof ▶

Apply the exponential law to the identity; partial maps of \(C^{k,\ell }\) maps are \(C^k\).

An atlas with varying models on \(C^\ell (M, N)\) satisfying the exponential law defines the same smooth structure as the atlas of the charts \(\varphi _f\).

Proof ▶

Apply the exponential law to the identity in both directions: the evaluation map is \(C^{k,\ell }\) for both atlases.

Proof ▶

The chart \(\varphi _f\) sends \(f\) to the zero section, and in this chart \(\varepsilon _x\) reads \(\tau \mapsto \Sigma (\tau (x))\); by the chain rule and the normalization (3.55), \(T_f \varepsilon _x(v) = v(x)\), so \(\Theta \) is the identity of \(\Gamma _f\), the model of the chart at \(f\), when the local addition is normalized. Glöckner and Neeb compute with geometric tangent vectors and assume the local addition normalized, which is no loss of generality (Proposition 14.3). For an arbitrary local addition \(\Sigma \), the atlases of \(\Sigma \) and of its normalization are compatible, and \(\Theta \) is the change of tangent coordinates between their charts at \(f\); it is again \(v \mapsto (T_f \varepsilon _x(v))_x\), a formula that involves no chart of \(C^\ell (M, N)\).

14.8 The compact-open \(C^\ell \) topology via iterated tangent maps

The iterated tangent bundles \(T^j M\) and tangent maps \(T^j f\) are defined recursively. The compact-open \(C^\ell \) topology on \(C^\ell (M, N)\) is the initial topology with respect to \(f \mapsto T^j f \in C(T^j M, T^j N)\), \(j \le \ell \), compact-open. It is initial with respect to \(f \mapsto f\) and \(f \mapsto Tf \in C^{\ell -1}(TM, TN)\) (as in the proof of [ GN26 , Lemma 4.1.6 ] ), and for \(\ell = \infty \) it is the projective limit of the finite orders [ GN26 , Remark 4.1.3(b) ] .

If the model of \(M\) is locally compact, the chart topology of \(C^0(M, N)\) is the compact-open topology, and that of \(C^{\ell +1}(M, N)\) is initial with respect to \(f \mapsto f \in C(M, N)\) and \(f \mapsto Tf \in C^\ell (TM, TN)\) (chart topology). (The chart topology is meant for locally compact models: otherwise it collapses to the \(C^0\) topology.)

Proof ▶

In self tangent charts, \(d(\mathrm{rep}\, f)\) is a component of a local representative of \(Tf\); in general tangent charts the local representative of \(Tf\) is \((F, dF)\) pulled back along a smooth map and composed with a smooth map. Localizations use compact neighborhoods.

Theorem 14.39 The chart topology and Definition 4.1.2 [ GN26 , Def. 4.1.2 ]
✓

If the model of \(M\) is locally compact, the chart topology of \(C^\ell (M, N)\) is the compact-open \(C^\ell \) topology of [ GN26 , Def. 4.1.2 ] , for \(\ell \in \mathbb {N}_0 \cup \{ \infty \} \).

Proof ▶

Both topologies satisfy the recursion of [ GN26 , Lemma 4.1.6 ] ; induction on \(\ell \), and the projective limit for \(\ell = \infty \).

14.9 The jet topology and Definition 4.1.2

The topology of \(C^\ell (M, N)\) is the jet topology (Definition 14.22), which works for every Hausdorff locally convex model and every manifold \(M\), compact or not. We compare it with the topology of [ GN26 , Def. 4.1.2 ] .

A space is compact-localizable if every compact set is covered by finitely many compact sets inside prescribed neighborhoods of its points. Hausdorff spaces and locally compact spaces are compact-localizable. Tangent bundles of Hausdorff manifolds are Hausdorff, so the iterated tangent bundles of a Hausdorff manifold, or of a manifold with a locally compact model, are compact-localizable.

Proof ▶

For Hausdorff spaces this is the finite compact cover lemma used by Glöckner and Neeb [ GN26 , Lemma A.6.11 ] .

The compact-open \(C^\ell \) topology of [ GN26 , Def. 4.1.2 ] is always finer than the jet topology. For every Hausdorff manifold \(M\), compact or not, modeled on any Hausdorff locally convex space, and for every \(M\) with a locally compact model, the two topologies coincide, \(\ell \in \mathbb {N}_0 \cup \{ \infty \} \). In particular, for \(M\) compact the topology underlying the atlases of \(C^\ell (M, N)\) and \(C^\infty (M, N)\) is that of [ GN26 , Def. 4.1.2 ] [ GN26 , Thm. 4.2.2(b) ] .

Proof ▶

Induction on the order through the tangent map, as in the proof of [ GN26 , Lemma 4.1.6 ] . In the tangent charts the local representative of \(Tg\) is \((y, w) \mapsto (F(y), \mathrm{d}F(y)\, w)\), \(F = i \circ g \circ j^{-1}\), and the jets of \(\mathrm{d}F\) and those of \(F\) determine each other by \(\mathrm{d}^{(m+1)}F(x, v_0, v) = \mathrm{d}^{(m)}(\mathrm{d}F)((x, v_0), (v, 0))\) [ GN26 , (1.36) ] and the formula for \(\mathrm{d}^{(m)}(\mathrm{d}F)\), the identities behind [ GN26 , Lemma 4.1.12 ] . Continuity of the jets of \(\mathrm{d}F\) on the non-compact set \(K \times H\) is reduced to compact subsets. Recovering the compact-open topology of \(C(T^n M, T^n N)\) from compact sets inside charts needs Lemma 14.40; for non-Hausdorff manifolds with non-locally compact models only one inclusion is proved. We do not claim these arguments are new.

Corollary 14.42 The jet topology and the chart topology, via Definition 4.1.2
✓

If the model of \(M\) is locally compact, the jet topology is the chart topology.

Proof ▶

This is Proposition 14.24; for a locally compact model it also follows from Theorems 14.41 and 14.39, both topologies being that of [ GN26 , Def. 4.1.2 ] .

Theorems 14.39 and 14.41 hold for a target \(N\) in any universe; only \(M\) and the models share one universe. In particular, the topology of \(C^\ell (M, N)\) is that of [ GN26 , Def. 4.1.2 ] , \(\ell \in \mathbb {N}_0 \cup \{ \infty \} \), for every Hausdorff \(M\) [ GN26 , Thm. 4.2.2(b) ] , and so is the chart topology when the model of \(M\) is locally compact. Moreover, [ GN26 , Def. 4.1.2 ] is natural in the target: a diffeomorphism \(\varphi \colon N \to N'\) induces a homeomorphism \(f \mapsto \varphi \circ f\) between the topologies of [ GN26 , Def. 4.1.2 ] on \(C^\ell (M, N)\) and \(C^\ell (M, N')\).

Proof ▶

The iterated tangent bundles run over bundled manifolds whose carriers live in the universe \(\max (u, v)\) of their tangent bundles, \(u\) that of the models. A target \(N\) in another universe is replaced by \(\mathrm{ULift}\, N\), a manifold with the charts of \(N\); \(f \mapsto \mathrm{up} \circ f\) is a homeomorphism \(C^\ell (M, N) \to C^\ell (M, \mathrm{ULift}\, N)\) for the jet topologies, the jets in corresponding charts being equal. Naturality follows from the chain rule \(T^j(\varphi \circ f) = T^j\varphi \circ T^j f\), with \(T^j\varphi \) a homeomorphism of inverse \(T^j(\varphi ^{-1})\); applied to \(\mathrm{up} \colon N \to \mathrm{ULift}\, N\), it shows that the transported topology is that of [ GN26 , Def. 4.1.2 ] whenever \(N\) is itself a carrier. We do not claim this argument is new.