A localized version of Ky Fan’s minimax inequality

https://doi.org/10.1016/S0362-546X(97)00700-1Get rights and content

Introduction

Let X denote a real normed space. Then the minimax inequality discovered by Ky Fan in 1972 reads as follows (see [5]).

Theorem A. Let KX be a compact convex set, g:K×K→R. Assume that g fulfills the following properties:

  • 1.

    xg(x,y) is quasiconcave on K for all fixed yK;

  • 2.

    yg(x,y) is lower semicontinuous (lsc) on K for all fixed xK;

  • 3.

    g(x,x)≤0 for all xK.

Then there exists y0K such that g(x,y0)≤0 for all xK.

This result has many interesting applications concerning fixed point theorems, variational inequalities, potential theory, mathematical economy, etc. It is also known that Ky Fan’s minimax inequality is one of the several statements which are equivalent to each other: Brouwer’s fixed point theorem, the lemma of Knaster–Kuratowski–Mazurkiewicz, the theorem of Browder concerning variational inequalities, the fixed point theorem of Kakutani, the fixed point theorem of Ky Fan–Glicksberg, the Nash equilibrium principle, the theorem of Gale–Nikaido–Debreu on Walras equilibrium in mathematical economics, etc. (see [6] Vol. IV, pp. 795–810]).

Consider now the special case of Theorem A, when xg(x,y) is concave (instead of being quasiconcave only) for all yK and g(x,x)=0 for all xK. Then, if y0 satisfies the requirements of the statement,g(y0+t(x−y0),y0)−g(y0,y0)t≤0forallt∈[0,1],x∈K.Therefore, taking the limit t→0+0, we getg′(y0,y0)(h)≤0forallh∈K−y0,where g′ denotes the (existing by the concavity) directional derivative of g with respect to the first variable. By homogeneity of g′ in the h variable, we can extend this inequality to the conical hull of Ky0, that is, we getg′(y0,y0)(h)≤0forallh∈cone(K−y0).On the other hand, if this inequality holds for some y0K and g is concave in the first variable, then (by known properties of concave functions)g(y0+h,y0)−g(y0,y0)≤g′(y0,y0)(h)≤0foh∈K−y0.Thus g(x,y0)≤0 for all xK, therefore the conclusion of Theorem A is valid.

The aim of this note is to state a principle (similar to the above formulated one) for locally Lipschitz functions in terms of a generalized Clarke-type derivative (see Theorem 1 of Section 2). It will also be pointed out that Theorem 1 and a specialized version of Ky Fan’s minimax inequality are directly equivalent, i.e. they are deducible from each other.

In the subsequent two sections, we clarify how this local Ky Fan’s inequality relates to the results mentioned above. Using Theorem 1, we give new and simple proofs for extended versions of Schauder’s and Kakutani’s fixed point theorems for inward maps (see Corollaries 1 and 2 below). In the last section, we deduce a new variational inequality for upper demicontinuous set-valued maps which, in particular, contains that of Browder [2]. In this way, it will be clear that our localized version of Ky Fan’s minimax inequality is equally powerful as the above mentioned fixed point theorems and variational inequalities.

Section snippets

The generalized Clarke-type derivative – Main result

Let X be a real normed space, KX be a nonempty compact convex set, and let f:D×K→R be a given function, where DX is an open set containing K. We say that f is locally Lipschitz on the diagonal of K×K with respect to the first variable if for each xK there exists Lx>0 and a neighbourhood UxD of x such that|f(x′,y)−f(x″,y)|≤Lx‖x′−x″‖for all x′,x″∈Ux and yUxK. The family of all functions f:D×K→R with the above property will be denoted by L(K).

Let f∈L(K). For each xK and hX, the generalized

Applications to fixed point theorems

First we show that, using Theorem 1, one can easily deduce an extended form of Schauder’s fixed point theorem (cf. [6] Vol. I, Theorem 2.6.2.A, p. 57]).

Corollary 1. Let K be a compact convex subset of a real normed space X and let ϕ:KX be a continuous function such that ϕ(x)∈x+TK(x) for all boundary points x of K. Then ϕ admits a fixed point x0K.

Proof. Consider the function f:X×K→R defined byf(x,y)=‖x−ϕ(y)‖.Clearly, xf(x,y) is a convex Lipschitz function with Lipschitz constant L=1. Thus f∈L

Applications to variational inequalities

In the sequel, let X be a real Banach space and X be its dual space.

A set-valued mapping A:D⊂X→2X is said to be upper demicontinuous (udc) at xD if, for any hX, the real-valued functionx↦σ(A(x),h)≔supA(x),h=supxA(x)x,his upper semicontinuous at x.

Clearly, if A is an usc set-valued map, then it is also udc.

If A(x)={A(x)}, that is, if A is a single valued map, then A is udc at xD if and only if the operator A:D⊂X→X is demicontinuous at xD, i.e., if for each sequence xn converging to x

First page preview

First page preview
Click to open first page preview

References (6)

  • J.P. Aubin, H. Frankowska, Set-Valued Analysis, Systems and Control: Foundations and Applications, vol. 2, Birkhäuser,...
  • F. Browder, The fixed point theory of multi-valued mappings in topological vector spaces, Math. Annalen 177 (1968)...
  • F.H. Clarke, Optimization and Nonsmooth Analysis, Canadian Math. Soc. Series of Monographs and Advanced Texts, Wiley,...
There are more references available in the full text version of this article.

Cited by (6)

  • Some applications of variational inequalities in nonsmooth analysis

    2007, Nonlinear Analysis, Theory, Methods and Applications
  • Equilibria and fixed points of set-valued maps with nonconvex and noncompact domains and ranges

    2006, Nonlinear Analysis, Theory, Methods and Applications
    Citation Excerpt :

    The above three results are the source of many investigations and have been generalized in several directions with applications in various branches of mathematics: minimax theorems, minimax inequalities, variational inequalities, fixed point theory, coincidence theory, game theory, mathematical economics, optimization problems, etc. Especially during the last few decades, by applying various new methods (notably based on those introduced by Ky Fan), remarkable progress has been made in these areas of investigations, and also some fascinating new problems have been proposed (see, e.g. [1–14,17–21,23–31,33–41] and many references therein). Most of the work has centered around the equilibrium points, minimax inequalities and fixed points of maps on convex compact sets, but there are also a considerable number of papers devoted to maps on nonconvex or noncompact domains (see, e.g. [10,28,37,39,41]).

  • GABOR KASSAY - IN MEMORIAM

    2022, Studia Universitatis Babes-Bolyai Mathematica
  • On Nash stationary points

    1999, Publicationes Mathematicae Debrecen
View full text