2 Manifolds and smooth maps
In this chapter we set up the basic objects of the library: manifolds modeled on locally convex spaces and \(C^{k}\) maps between them. The calculus is the Michal–Bastiani calculus of [ LCA ] ; the references for the manifold theory are [ Nee06 , § I.3 ] and [ GN26 , Ch. 3 ] , and [ Lan99 ] for the Banach case. Throughout, \(E\), \(F\), \(G\) denote real Hausdorff locally convex spaces. We stress from the start that no completeness is assumed: nothing in this chapter needs it. Tangent bundles, vector fields and manifolds with rough boundary are treated in the next chapters.
2.1 Manifolds
The groupoid \(\mathcal{G}_{\mathrm{MB}}(E)\) consists of the homeomorphisms \(\varphi \colon U \to V\) between open subsets of \(E\) such that \(\varphi \) is smooth on \(U\) and \(\varphi ^{-1}\) is smooth on \(V\), in the sense of Michal–Bastiani.
That this is a structure groupoid (closed under composition, inversion and restriction, and local in nature) follows from the chain rule and the locality of smoothness.
A manifold modeled on \(E\) is a topological space \(M\) with an atlas of charts with values in \(E\) whose transition maps belong to \(\mathcal{G}_{\mathrm{MB}}(E)\) [ Nee06 , Definition I.3.1 ] . A Banach manifold is a manifold modeled on a normed space which is complete; a Hilbert manifold is one modeled on a complete inner product space.
In Lean, a manifold is a Mathlib ChartedSpace \(E\) \(M\) with HasGroupoid for \(\mathcal{G}_{\mathrm{MB}}(E)\); Mathlib’s charted spaces accept an arbitrary topological model space, so no new infrastructure is needed. Every chart comes with a preferred chart \(c_x = \texttt{chartAt}\ x\) at each point \(x \in M\), and we use it systematically.
We follow the convention of Glöckner–Neeb [ GN26 , § 5.7 ] : the model space of a manifold is not assumed to be complete. The formal theory of this and the following chapters (charts, smooth maps, tangent bundle, vector fields, connections, submanifolds) does not use completeness, since the calculus of [ LCA ] replaces integrals in the model space by Hahn–Banach arguments. Results that need completeness (integrals with values in \(E\), ordinary differential equations, exponential maps, inverse function theorems) say so in their hypotheses. In particular, a manifold modeled on an incomplete normed space is a manifold in our sense, but it is not a Banach manifold. Without completeness things do go wrong: [ GN26 , Example 5.6.8 ] is a Lie group modeled on an incomplete normed algebra that has no exponential function.
On normed spaces, Michal–Bastiani smoothness agrees with Fréchet smoothness, so for complete normed \(E\) a Banach manifold is the classical object of [ Lan99 ] .
If \(M\) and \(N\) are manifolds modeled on \(E\) and \(F\), then \(M \times N\), with the product charts, is a manifold modeled on \(E \times F\). The projections onto \(M\) and \(N\) are smooth, and a pair \((f, g)\) of \(C^{k}\) maps \(P \to M\), \(P \to N\) is a \(C^{k}\) map \(P \to M \times N\).
The transition maps of the product atlas are products \(\varphi \times \psi \) of transition maps, and a product of smooth diffeomorphisms between open sets is again one, since \((\varphi \times \psi ) = (\varphi \circ \pi _1, \psi \circ \pi _2)\) and pairs of smooth maps are smooth. In charts the projections are the linear maps \(\pi _1, \pi _2\).
In Lean the model space of the product is Mathlib’s type synonym ModelProd \(E\) \(F\) of \(E \times F\), which avoids instance diamonds when \(M\) or \(N\) is itself a product.
2.2 Smooth maps
For maps \(f \colon M \to N\), open \(U \subseteq M\) and charts \(e\) of \(M\), \(e'\) of \(N\), we denote by
the domain of the local representative \(e' \circ f \circ e^{-1}\). It is open when \(f\) is continuous on \(U\) (LocallyConvex.Manifolds.isOpen_chartDomain).
A map \(f \colon M \to N\) is \(C^{k}\) on an open set \(U \subseteq M\) if it is continuous on \(U\) and, for every \(x \in U\), the local representative \(c'_{f(x)} \circ f \circ c_x^{-1}\) in the preferred charts is \(C^{k}\) on \(D(f, U, c_x, c'_{f(x)})\). It is smooth on \(U\) if it is \(C^{k}\) on \(U\) for every \(k\).
Only the preferred charts appear in the definition. The next result says that this is no restriction.
If \(f\) is \(C^{k}\) on \(U\), then its local representatives \(e' \circ f \circ e^{-1}\) in arbitrary charts of the atlases (or of the maximal atlases) are \(C^{k}\) on \(D(f, U, e, e')\). Conversely, a continuous map whose local representatives in the preferred charts are \(C^{k}\) near the center of each chart is \(C^{k}\).
The transition maps are smooth by definition of the atlas. Near a point \(y = e(p)\) we write \(e' f e^{-1} = (e' c'^{-1}) \circ (c' f c^{-1}) \circ (c e^{-1})\) with \(c = c_p\), \(c' = c'_{f(p)}\); the three factors are \(C^{k}\) and the chain rule applies. Since \(C^{k}\) is a local property, this suffices.
If \(f\) is \(C^{k}\) on \(U\), \(g\) is \(C^{k}\) on \(V\) and \(f(U) \subseteq V\), then \(g \circ f\) is \(C^{k}\) on \(U\). Identities and constant maps are smooth.
In the preferred charts at \(x\) and \(g(f(x))\), near a point we insert the preferred chart \(d\) at the intermediate point and write the local representative of \(g \circ f\) as the composition of the local representatives of \(g\) and \(f\) in the charts involving \(d\). These are \(C^{k}\) by Theorem 2.6, and we conclude by the chain rule.
For a map \(f \colon E \to F\) between the model spaces themselves (with their identity charts) and \(U \subseteq E\) open, \(f\) is \(C^{k}\) on \(U\) as a map between manifolds if and only if it is \(C^{k}\) on \(U\) in the sense of Michal–Bastiani.
The preferred charts are identities, so \(D(f, U, \mathrm{id}, \mathrm{id}) = U\) and the local representative is \(f\) itself.
The model spaces \(E\), \(F\), \(G\) live in a common universe, because the Michal–Bastiani derivative of a map defined on \(E\) is a map defined on \(E \times E\). The manifolds themselves are unrestricted. A consequence that will reappear below is that real-valued functions on \(M\) are not treated as maps between manifolds (the model space \(\mathbb {R}\) need not be in the universe of \(E\)), but through their local representatives.