Applications of Strassen's theorem and Choquet theory to optimal transport problems, to uniformly convex functions and to uniformly smooth functions (Q6097522)

An extensive account of classical optimal transport can be found in [\textit{C. Villani}, Topics in optimal transportation. Providence, RI: American Mathematical Society (AMS) (2003; Zbl 1106.90001)] and in [\textit{C. Villani}, Optimal transport. Old and new. Berlin: Springer (2009; Zbl 1156.53003)]. The martingale optimal transport is presented in [\textit{M. Beiglböck} and \textit{N. Juillet}, Ann. Probab. 44, No. 1, 42--106 (2016; Zbl 1348.49045)] and in [\textit{A. Galichon} et al., Ann. Appl. Probab. 24, No. 1, 312--336 (2014; Zbl 1285.49012)]. One of the aims of this article is to introduce a novel, unifying approach to various optimal transport problems. A new, natural formulation of the primal and the dual problems for martingale optimal transport is provided. The author introduces the martingale triangle inequality, and uses it to prove an analogue of the Kantorovich-Rubinstein duality formula in the martingale optimal transport setting. He also simplifies the formulation of the dual problem for martingale optimal transport, under the assumption that the considered cost function satisfies the martingale triangle inequality. The martingale optimal transport for cost functions that satisfy the martingale triangle inequality is closely related to classes of uniformly convex and of uniformly smooth functions. Their continuity properties are studied and a new characterization of such functions is provided. The paper comprises new proofs of the duality formulae for the optimal transport problem, including the Kantorovich-Rubinstein formula and also the duality formula for multi-marginal optimal transport. The main tool employed to prove the aforementioned results is a new variant of Strassen's disintegration theorem. Another tool is Choquet theory, which is used to infer the duality formulae from representations of appropriate sets of extreme points. In particular, the set of extreme points of pairs of probabilities in convex order is computed.