Weak convergence of empirical Wasserstein type distances

We estimate contrasts between two continuous distributions and on such that the set is a finite union of intervals, possibly empty or . The non-negative convex cost function is not necessarily symmetric and the sample may come from any joint distribution on with marginals and having light enough tails with respect to . The rates of weak convergence and the limiting distributions are derived in a wide class of situations including the classical Wasserstein distances and . The new phenomenon we describe in the case involves the behavior of near , which we assume to be regularly varying with index ranging from to and to satisfy a key relation with the behavior of near through the common tails. Rates are then also regularly varying with powers ranging from to also affecting the limiting distribution, in addition to .
View on arXiv