Representation theorems for extremal distributions (Q797891): Difference between revisions

In this paper, the authors give a new proof of a result of \textit{G. C. Taylor} [Scand. Actuarial J. 1977, 94-105 (1977; Zbl 0369.62111)] and the second author [Insur. Math. Econ. 1, 109-130 (1982; Zbl 0488.49030)] on the representation of extremal distributions and generalize it as follows: If E is a compact and metrizable topological space, \(f_ 0,f_ 1,...,f_ n\in C(E)\) and \(S_ 1(z)\neq\emptyset \), then there exists a probability distribution \(\nu\) on (E,\({\mathcal B})\) that is supported by at most \(n+1\) points such that \(\nu\in S_ 1(z)\) and \(\int f_ 0d\nu =\sup\{\int f_ 0d\mu\); \(\mu\in S_ 1(z)\}\) where \(z=(z_ 1,...,z_ n)\in R^ n\), \({\mathcal N}^+_ 1\) is an arbitrary subset of \({\mathcal M}^+_ 1\) (the set of all probability measures on (E,\({\mathcal B}))\) that contains all the probability distributions supported by at most \(n+1\) points of E, and \(S_ 1(z)=\{\mu\in {\mathcal N}^+_ 1;\int f_ id\mu =z_ i\), \(i=1,...,n\}.\) The authors also consider the case in which E is a locally compact, \(\sigma\) -compact and metrizable topological space.