Near convexity, metric convexity, and convexity (Q2478026)

Many aspects of convexity in a normed space, hidden by the classical use of the obscure principle of excluded middle, are only revealed by a constructive approach. This paper studies various types of convexity, introducing several new concepts; it uses methods proposed by \textit{E.\,A.\thinspace Bishop} in Chapter~1 of ``A Constructivist Manifesto'' in [``Foundations of Constructive Analysis'' (Maidenhead, Berksh.: McGraw-Hill) (1967; Zbl 0183.01503)]. The present paper demonstrates that certain implications and equivalences exist between the various concepts; some absolutely, and some classically but not constructively. The types of convexity considered in a real normed space \( X \) or subset \( S \) are: {\parindent6mm \begin{itemize}\item[(a)] Convexity. For any \( x,y \in S \) and any real number \( \lambda \) with \( 0 \leq \lambda \leq 1 \), the point \( z = (1- \lambda )x + \lambda y \) lies in \( S \). \item[(b)] Uniform convexity. For any \( \varepsilon > 0 \) there exists \( \delta > 0 \) such that for any unit vectors \( u \) and \( v \) with \( \| u - v \| > \varepsilon \), the midpoint \( (u+v)/2 \) has norm less than \( 1 - \delta \). \item[(c)] Strict convexity. If the convex hull of the pair \( \{ x,y \} \) is contained in the boundary of the closed unit ball with center 0, then \( x = y \). \item[(d)] Strong strict convexity. Whenever \( u \) and \( v \) are distinct unit vectors, the midpoint \( (u+v)/2 \) has norm less than 1. \item[(e)] Near convexity. For any \( x,y \) and any nonnegative real numbers \( \lambda \) and \( \mu \) such that \( \| x - y \| < \lambda + \mu ,\) there exists \( z \) such that \( \| x - z \| < \lambda \) and \( \| z - y \| < \mu \). \item[(f)] Metric convexity. For any \( x,y \) and any nonnegative real numbers \( \lambda \) and \( \mu \) such that \( \| x - y \| = \lambda + \mu \), there exists \( z \) such that \( \| x - z \| = \lambda \) and \( \| z - y \| = \mu \). \item[(g)] Osculating balls. Any two closed balls of radii \( r_1 \) and \( r_2 \) whose centers are at distance \( r_1 + r_2 \) have exactly one common point. \item[(h)] Nondegenerate osculating balls. As in (g), with \( r_1 + r_2 > 0 \). \item[(i)] Osculating unit balls. As in (g), with \( r_1 = r_2 = 1 \). \item[(j)] Strict triangle inequality. If \( \| x+y \| = \| x \| + \| y \| \not= 0 \), then \( x \) and \( y \) are linearly dependent. \item[(k)] Any metrically convex subset \( S \) of \( X \) is convex. \end{itemize}} Several of these definitions have a purely metric content; thus some results relate the metric to the linear structure. Several interesting questions remain.