Sion's minimax theorem

Sion's minimax theorem is a fundamental result in functional analysis and game theory establishing equality between a supremum of infima and an infimum of suprema for certain functions on product spaces. The theorem generalizes classical results such as the Von Neumann minimax theorem and interacts with results from convex analysis, topology, and measure theory. It is widely used in the study of zero-sum games, variational inequalities, and saddle point problems.

Statement of the theorem

Sion's minimax theorem considers a function f defined on the Cartesian product of two topological vector spaces X and Y. Under hypotheses of convex set structure in X, concave structure in Y, and semicontinuity conditions, the theorem asserts sup_{x in X} inf_{y in Y} f(x,y) = inf_{y in Y} sup_{x in X} f(x,y). Typical hypotheses require X to be a compact convex subset of a locally convex topological vector space and Y to be a convex subset of a topological vector space, with f convex and lower semicontinuous in one variable and concave and upper semicontinuous in the other. The theorem thereby gives a sufficient criterion for existence of saddle points and equilibrium values analogous to the Minimax theorem in finite dimensions.

Historical background and motivation

Sion proved the result in 1958 against a backdrop of developments initiated by John von Neumann's 1928 minimax result for finite zero-sum games and subsequent extensions by researchers working in functional analysis and convexity theory. Motivations came from problems in economics where equilibrium concepts advanced by scholars at institutions such as Princeton University and University of Chicago interacted with methods from topology and measure theory. Parallel work by mathematicians like Ky Fan, L. A. Lyusternik, and contributors to the theory of topological vector spaces informed the semicontinuity and compactness conditions that Sion adopted. The theorem consolidated strands from variational principles studied by researchers associated with Institute for Advanced Study and provided tools applicable in both abstract analysis and applied game theory.

Proof outline

The proof proceeds by reduction to separation and fixed-point arguments common in functional analysis and uses convexity and semicontinuity hypotheses to apply supporting hyperplane techniques. One constructs auxiliary convex sets in dual or product spaces and employs the Hahn–Banach theorem and compactness arguments to show the existence of approximate saddle points. Upper semicontinuity in one variable and lower semicontinuity in the other permit passage from approximate to exact equalities using limiting arguments reminiscent of proofs of the Ky Fan inequality and the Brouwer fixed-point theorem or its generalizations such as the Schauder fixed-point theorem. The strategy mirrors finite-dimensional minimax proofs by John Nash and John von Neumann while relying on infinite-dimensional topology tools developed in schools at Université de Paris and Princeton University.

Applications and examples

Sion's result is applied in analyses of zero-sum games where strategy sets are infinite-dimensional, for instance in continuous-time control problems studied in Stanford University and Massachusetts Institute of Technology research. It underlies existence proofs for equilibria in optimization problems arising in Bellman-type dynamic programming, variational inequalities in mechanics and physics contexts, and saddle point formulations in constrained optimization at institutions such as ETH Zurich and University of Cambridge. Concrete examples include bilinear forms on product spaces of functions where one side is compact in the weak topology, and payoff functions in stochastic games researched at Columbia University and University of California, Berkeley. The theorem also facilitates duality results in convex programming linked to work at INRIA and Courant Institute.

Generalizations and related results

Sion's theorem is part of a web of minimax and saddle point results. Notable relatives include the Von Neumann minimax theorem, Ky Fan minimax inequality, results by John von Neumann and Oskar Morgenstern in game theory, and extensions treating noncompact or nonconvex settings via measurable selection theorems linked to C. R. Rao and researchers in measure theory at University of Chicago. Further generalizations relax compactness using coercivity conditions or employ variational methods from the Direct method in the calculus of variations and techniques from nonlinear functional analysis developed by groups at IHES and Max Planck Institute for Mathematics. Connections to modern developments appear in research on equilibrium in infinite games by scholars affiliated with Yale University and Harvard University and in algorithmic game theory in work at Google and Microsoft Research.

Category:Theorems in functional analysis