The central question of this paper is: given two probability distributions and a specific set of points in their combined space, can you find a joint probability distribution that matches both original distributions exactly and places all its probability mass on precisely that set, no more and no less? This is a subtle problem because it is relatively easy to find a joint distribution whose support is contained within a given set, but requiring the support to be exactly that set, meaning every part of the set must actually receive some probability, is much harder.
The paper solves this problem in two main settings. When both distributions are discrete, meaning they assign probabilities to countable collections of points, the problem turns out to be equivalent to a question in network flow theory: does a certain graph, where nodes carry weights and edges carry capacity limits, admit a flow that is positive on every single edge? For continuous distributions on the real line, the authors develop a different characterization. A particularly clean result emerges when the set in question is "regular," meaning it has no isolated thin parts and equals the closure of its interior, like a filled region in the plane with no dangling filaments. The special case where both distributions are uniform on the unit interval connects to the classical theory of doubly stochastic measures, which arise in areas like optimal transport and the study of statistical dependence.
Overall, the paper draws a precise boundary between sets that can serve as exact supports of joint distributions and those that cannot. The results unify ideas from measure theory, combinatorics, and graph theory, and they clarify when a geometric or combinatorial structure is compatible with prescribed probabilistic marginals. This has implications for areas like statistics, where one wants to model dependence structures between variables, and for mathematical economics and optimal transport, where coupling two distributions on a specific region of the product space is a common concern.