Martingale transports and Monge maps (Q7034150)

Martingale optimal transport is a subject of extensive research. In the classical transport theory (without the martingale constraint), much attention has been devoted to transport plans given by so-called Monge maps (transport maps). Under some natural requirements, it is known that the optimizers for numerous important optimal transport problems are indeed Monge maps. The present paper is dedicated to the theory of Monge martingale transports (MMT) as well as their implications. Given marginal probability distributions \(\mu\) and \(\nu\) that are in a convex order, it is not obvious if an MMT exists, apart from the trivial fact that atoms in \(\nu\) often preclude the existence of Monge transport (martingale or not) from \(\nu\) to \(\mu\). Assuming that \(\nu\) is atomless, the authors prove that the set of (backward) MMTs is never empty: moreover, they construct a particular MMT that is called the barcode transport, a name derived from its pictorial representation. The basic idea is to decompose the marginals \(\mu\) and \(\nu\) into countably many pieces (the bars of the barcode) that can be coupled by MMTs more easily, and then aggregate. As an auxiliary result, the authors provide a novel structural description of the left-curtain transport. More generally, it is proved that ``backward Monge'' martingale transports are dense in the set of all martingale couplings, paralleling the classical denseness result for Monge transports in the Kantorovich formulation of optimal transport. Various properties and applications of this fact are presented, including a refined version of Strassen's theorem and a mimicking theorem, where the marginals of a given martingale are reproduced by a ``backward deterministic'' martingale.