Stochastic order on metric spaces and the ordered Kantorovich monad (Q2309111): Difference between revisions

The subject of the paper is the notion of \textit{stochastic order}. This is an order on the set of probability distributions on a space which is induced from a given order on the space, such that \(p \leq q\) if \(q\) assigns a higher (or equal) probability to each upper set. Reflexivity and transitivity is usually clear, but antisymmetry of this order is stated to be a hard problem in general. This subject is treated in the context of probability monads. The central contribution of the paper is the extension of the Kantorovich monad, which sends each metric space \(X\) to a certain space of probability measures on \(X\) equipped with the Kantorovich or \(1\)-Wasserstein metric, to a category of ordered metric spaces. To achieve this, the authors introduce the notion of ``\(L\)-ordered metric space'' -- this is an ordered complete metric space where the order can be detected on the short monotonous maps. In other words, if \(f(x) \leq f(y)\) for \textit{every} short monotonous map \(X \to \mathbb{R}\), then \(x \leq y\). This condition holds precisely for those ordered metric spaces which are isomorphic to closed subsets of ordered Banach spaces. With this assumption, the Kantorovich space acquires an order as well - this extends previous results on the antisymmetry of the stochastic order. The paper goes on to explore various elaborations of this idea. The algebras for the ordered Kantorovich monad are precisely the closed \textit{convex} subsets of ordered Banach spaces, with the structure maps given by integrating a measure (i.e taking expectation).