Optimal transport is a mathematical framework for finding the most efficient way to move mass from one distribution to another, with applications ranging from economics to machine learning. A "martingale" version of this problem adds the constraint that the transport must respect a fairness condition borrowed from probability theory: on average, the future looks like the present. This is relevant in finance, where it captures the idea that asset prices under certain models should not allow arbitrage (free money). The paper studies a variation of a well-known formula in this field, called the Benamou-Brenier formula, adapted to the martingale setting.
The twist introduced here is that instead of fully specifying where the mass must end up (the second marginal distribution), the authors only require that a finite number of summary statistics, called moment constraints, are satisfied at the endpoint. This is a much weaker requirement and makes the problem harder to analyze. The authors derive a "dual" version of the problem, which is a standard technique in optimization where you reformulate a hard problem into an equivalent one that is often more tractable. Crucially, they show there is no "duality gap," meaning the original problem and its dual have the same value, which is an important technical guarantee that the reformulation is faithful.
The paper also characterizes what the optimal solution looks like. The solutions turn out to be closely related to objects called Bass martingales, which are a well-studied family of stochastic processes that arise naturally in the martingale transport literature. Bass martingales have a clean, structured form, and showing that optimal solutions belong to this class gives useful insight into the geometry of the problem. Overall, the work extends the theoretical toolkit for martingale optimal transport to settings where the endpoint distribution is only partially known, which is a more realistic scenario in many practical applications.