On modular approximants in sequential convergence spaces (Q2657116)

A convex modular on a real vector space \(X\) is a mapping \(\rho:X\to [0,\infty]\) such that:\(\;\) (1) \(\;\rho(x)=0\iff x=0\),\(\;\)(2)\(\;\rho(x)=\rho(-x)\), and (3)\(\; \rho(\alpha x+\beta y)\le \alpha \rho(x)+\beta\rho(y) \) for all \(\alpha,\beta\ge 0\) with \(\alpha+\beta=1\). The vector space \(X_\rho=\{x\in X:\lim_{\lambda\to 0}\rho(\lambda x)=0\}\) is called a modular space. One considers also a convergence structure \(\zeta\) on a vector space \(X\). The pair \((X,\zeta)\) is called an \(L\)-space if the uniqueness of limits is assured and an \(LTI\)-space if, further, the convergence \(\zeta\) is translation invariant. If \(\tau\) is a Hausdorff vector topology on \(X\), then \(X\) with the sequential convergence structure \(C(\tau)\) corresponding to \(\tau\) is an \(LTI\)-space. A modulated topological vector space is a triple \((X_\rho,\rho,\tau)\), where \(\rho\) is a convex modular, \(\tau \) is a Hausdorff vector topology on \(X\), and some compatibility relations between \(\rho\) and \(\tau\) (as, e.g. the \(\tau\)-lsc of \(\rho\)) are satisfied. A modulated \(LTI\) space \((X_\rho,\rho,\zeta)\) is defined similarly. Several non-equivalent notions of uniform convexity can be defined for a convex modular \(\rho\), all agreeing with the usual one if \(\rho\) is a norm. For a comprehensive study of modular spaces, see [\textit{ M. A. Khamsi} and \textit{ W. M. Kozlowski}, Fixed point theory in modular function spaces. Cham: Birkhäuser/Springer (2015; Zbl 1318.47002)]. The author studies the best approximation problem within this context -- existence, uniqueness and the continuity of the modular projection. For instance, every nonempty \(\zeta\)-closed convex subset of a uniformly convex \(\rho\)-complete LTI-space \((X_\rho,\rho,\zeta)\) is a Chebyshev set, i.e., the existence and uniqueness of best approximation is guaranteed (Theorem 3.1). If further, \(\rho\) satisfies the \(\Delta_2\)-condition (\(\rho(2x)\le K\rho(x)\) for some \(K>0\)) and \(C\) is also \(\rho\)-bounded, then the modular projection \(P:X\to C\) is \((\rho,\zeta)\)-usc (Theorem 3.2).