Minimax theorems and saddling transformations (Q757260): Difference between revisions

Let X, Y be nonempty convex subsets of topological vector spaces, and let L be the extended real numbers. (Actually, L can be more general.) A member \(f\in L^{X\times Y}\) (viz., a function f: \(X\times Y\to L)\) is said to have a saddle value, iff \(\inf_{Y}\sup_{X}f=\sup_{X}\inf_{Y}f\). A function f is called quasi-concave-convex, if for every \((x_ 0,y_ 0)\in X\times Y\) and \(a\in L\), the sets \(\{\) \(y\in Y|\) \(f(x_ 0,y)\leq a\}\) and \(\{\) \(x\in X|\) \(f(x,y_ 0)\geq a\}\) are convex. Let \({\mathcal H}\) be any subfamily of \(L^{X\times Y}\). A mapping S: \(L^{X\times Y}\to L^{X\times Y}\) is called a saddling transformation for \({\mathcal H}\) if (i) every saddle point of \(f\in L^{X\times Y}\) is a saddle point of S(f), (ii) whenever \(f\in L^{X\times Y}\) has a saddle value, S(f) has a saddle value and they are equal; and (iii) for every \(f\in {\mathcal H}\), S(f) has a saddle value. The mapping S is called stable for \({\mathcal H}\) if S(\({\mathcal H})\subset {\mathcal H}\). Let \({\mathcal H}_ 0\) be the family of quasi-concave- convex functions f such that there exists \((x_ 0,y_ 0)\) for which \(\{\) \(x\in X|\) \(f(x,y_ 0)\geq a\}\) is relatively compact for each \(x\in L\) or for which \(\{\) \(y\in Y|\) \(f(x_ 0,y)\leq a\}\) is relatively compact for each \(a\in L\). This paper establishes that specific mappings (given in the paper) are stable saddling transformations for \({\mathcal H}_ 0\), and uses this result to prove a generalization of Sion's minimax theorem.