Minimax theorems with one-sided randomization (Q1912671)

In an earlier paper [Manuscr. Math. 29, 277-294 (1979; Zbl 0425.90089)], the author proved the following theorem which generalizes a classical result of \textit{J. E. L. Peck} and \textit{A. L. Dulmage} [Can. J. Math. 9, 450-458 (1957; Zbl 0078.32704)]: Theorem. Let \(Y\) be a countably compact topological space and \(F\) a nonvoid set of lower-semicontinuous functions \(f: Y\to \mathbb{R}^*:= \mathbb{R}\cup \{\infty\}\) such that for the arithmetic mean \(\varphi(\alpha, \beta)= {1\over 2} (\alpha+ \beta)\) we have \[ \forall f_1, f_2\in F \exists f_0\in F: f_0\geq \varphi(f_1, f_2).\tag{\(*\)} \] Then there exists a Borel probability measure \(\nu\) such that \[ \sup_{f\in F} \int fd\nu= \sup_{f\in F} \inf_{y\in Y} f(y). \] This theorem can be proved by applying an appropriate integral representation theorem to \textit{H. König's} generalization [Arch. Math. 19, 482-487 (1968; Zbl 0179.21001)] of \textit{Ky Fan's} minimax theorem [Proc. Natl. Acad. Sci. USA 39, 42-47 (1953; Zbl 0050.06501)]. In the present paper, a previous generalization of the Ky Fan-König minimax theorem of the author [Arch. Math. 55, 573-583 (1990; Zbl 0701.49017)] is used to prove: Main result (simplified version). The above Theorem remains true for arbitrary nondecreasing convex functions \(\varphi: \mathbb{R}^*\times \mathbb{R}^*\to \mathbb{R}^*\) which satisfy \(\varphi(a, \infty)= \varphi(\infty, \alpha)= \infty\) and \(\varphi(\alpha, \beta)> \min\{\alpha, \beta\}\) for \(\alpha\neq \beta\). In a subsequent paper [Colloq. Math. 68, 179-186 (1995)] applications in statistical decision theory and, in particular, in the theory of convex metric spaces are discussed.