The affine separation problem revisited (Q1749014)

Separation theorems play an important role in functional analysis. Typical results provide sufficient conditions for the existence of an affine separating function between a convex and a concave one. The aim of the present paper is to give a new proof of the following characterization result due to \textit{E. Behrends} and \textit{K. Nikodem} [Stud. Math. 116, 43--48 (1995; Zbl 0847.52004)]: Let \(D\) be a convex subset of a \(d\)-dimensional space and \(f,g\) be functions from \(D\) to the real line. Then there exists an affine separator between \(f\) and \(g\) iff \[ \sum_{k=1}^n\lambda_{k}f(x_{k})\leq\sum_{l=1}^m\mu_{l}g(y_{l}) \] for all convex combinations \(\sum_{k=1}^n\lambda_{k}x_{k}=\sum_{l=1}^m\mu_{l}y_{l}\), where the \(x_{k},y_{l}\) lie in \(D\) and \(n+m\leq d+2\). The original proof relies on Helly's theorem, the one presented here uses also the Radon theorem from convex geometry. It avoids the use of set-valued mappings; the affine separator is found as a point in the intersection of suitable subsets of the space of affine functions from \(D\) to the reals.