Extremal doubly stochastic measures and optimal transportation

Abstract

This article connects the theory of extremal doubly stochastic measures to the geometry and topology of optimal transportation. We begin by reviewing an old question (# 111) of Birkhoff in probability and statistics [4], which is to give a necessary and sufficient condition on the support of a joint probability to guarantee extremality among all measures which share its marginals. Following work of Douglas, Lindenstrauss, and Beneš and Štěpán, Hestir and Williams [15] found a necessary condition which is nearly sufficient; we relax their subtle measurability hypotheses separating necessity from sufficiency slightly, yet demonstrate by example that to be sufficient certainly requires some measurability. Their condition amounts to the vanishing of γ outside a countable alternating sequence of graphs and antigraphs in which no two graphs (or two antigraphs) have domains that overlap, and where the domain of each graph / antigraph in the sequence contains the range of the succeeding antigraph (respectively, graph). Such sequences are called numbered limb systems. Surprisingly, this characterization can be used to resolve the uniqueness question for optimal transportation on manifolds with the topology of the sphere.