Infinite dimensional calculus allowing nonconvex domains with empty interior (Q2640954): Difference between revisions

In this paper, integration of curves serves as the pivotal concept for a new differentiation theory, developed in various settings of separated spaces e.g. complete locally convex spaces or rather a certain closed category of convergence spaces containing them, `complete' compactly generated topological vector spaces, Mackey complete bornological locally convex spaces (the latter for both real and complex scalars). A curve p: \(\Lambda\to U\subset E\) (scalar-to-vector \({\mathcal C}\)-maps) is called a path if there exists a curve \(p'\) such that \(\int^{\beta}_{\alpha}p'(\tau)=p(\beta)-p(\alpha)\) holds identically. A map f: \(U\to F\) is called a \({\mathcal C}^ 1\)-map if there exists a \({\mathcal C}\)-map Df: \(U\to [E,F]\) (canonical space of linear \({\mathcal C}\)- maps) such that \(\int^{\beta}_{\alpha}Df(p(\tau))\cdot p'(\tau)d\tau =f(p(\beta))-f(p(\alpha))\) holds identically for all paths p. The derivatives thus defined are unique. The domain \(U\subset E\) is required merely to have, at each point, enough tangent vectors \(p'(\tau)\) to generate the linear space E. Calculus is developed on these `tangential' domains (a very wide class) with remarkably simple proofs for the usual basic theorems. Certain elementary maps between Fréchet spaces, hitherto excluded from calculus by numerous known differentiation theories old and new (which require open domains), now emerge as smooth maps. An example is the map f: \(U\to E\), \(f(x)=1/x\), where \(E=C({\mathbb{R}},{\mathbb{R}})\) is the familiar Fréchet space of continuous maps (with compact-open topology) and U is the subspace formed by all never- vanishing functions x: \({\mathbb{R}}\to {\mathbb{R}}\). The domain of this map is nonconvex with empty interior while f has no continuous extension to a properly larger domain.