Second order duality for a minimax programming problem (Q1184763): Difference between revisions

The authors obtain the equality of primal and dual costs for a minimax problem under some hypotheses of binvexity on the data. Binvexity is defined by the authors and is an assumption that involves the Hessian of \(f\). Specifically, a function \(f: \mathbb{R}^ n\to\mathbb{R}\) is binvex at \(x^ 0\) if for all \(x\) and \(r\) in \(\mathbb{R}^ n\) there exists \(\eta(x,x^ 0)\) such that \(f(x)-f(x^ 0)\geq\eta(x,x^ 0)^ t[\nabla f(x^ 0)+\nabla^ 2 f(x^ 0)r]-{1\over 2} r^ t\nabla^ 2 f(x^ 0)r\). It is worth doing the following remark. By taking \(\| r\|\) large, we see that this implies that \(\nabla^ 2 f(x^ 0)\geq 0\). Hence, if it is binvex for all \(x\), it is indeed convex. The authors define also other concepts extending convexity such as pseudo-binvexity that seem more general, under which some duality results are obtained.