Simplified optimality conditions for minimizing the difference of vector-valued functions (Q5942220): Difference between revisions

This article is a valuable contribution to the axiomatic theory on a problem (P) of nonconvex minimization of a vector-valued function. The objective function \(g(x)- h(x)\) to be minimized is defined by \(g,h: X\to Z\), \(Z\) being an ordered topological vector space endowed with a partial ordering given by a convex cone \(P\subset Z\) and artificial elements \(\pm\infty\) adjoined. By this setting, the authors extend results from the case \(Z= \mathbb{R}\). As analytical tools, for an arbitrary function \(f\), a nonempty family \(\Phi\) of functions on \(X\) and some \(\alpha\in Z\), the \(\varepsilon\)-\(\Phi\)-subdifferential and the \(\alpha\)-\(\Phi\)-minorant of \(f\) are provided. Herewith, for \((f:=)h\) assumptions (H1,2) are made, saying that \(h\) is almost \(\Phi\)-supported or \(\Phi\)-supported, respectively. Several examples are given in which \(X\) is, e.g., a locally convex topological vector space or a metric space, \(\Phi\) consists of continuous (linear or real-valued) functions, and \(h\) is lower semicontinuous, proper and/or convex. A first main result says that under (H1) a point \(x^0\) from the common domain of \(g\), \(h\) solves (P) if and if only if for all \(\varepsilon\geq 0\) at \(x^0\) the subdifferential of \(h\) is lying in the one of \(g\). Under (H2) and stability of \(\Phi\) with respect to dilatation and finite suprema, a second one makes the same characterization by referring to \(\varepsilon= 0\) only. A third one equivalently represents the forgoing two characterizing conditions by duality conditions involving \(\Phi\)-conjugate functions. The authors conclude by related results, e.g., on representation of the \(\varepsilon\)-subdifferential of a proper \(h\) referring to the graph of the \(0\)-subdifferential of \(h\). The article is clearly written and carefully proved.