The Setting
A compactly supported target law need not have finitely many atoms. The question is whether finite target transport maps have a limit in the distance between maps, and whether their stability estimate persists for the limiting laws.
Work in Euclidean space , with coordinates indexed by . For a compact source set , its uniform measure is the restriction of Lebesgue measure to , normalized as in (1): (1)✦ A measure is supported in a target set when .✦
A coupling of measures and is a probability measure on with marginals and .✦ The graph transport plan of a map with source measure is the pushforward in (2).✦ The quadratic transport cost of a measure on the Cartesian square is the nonnegative extended integral in (2): (2) The quadratic Wasserstein distance between measures and is the square root of the real value of the infimum of over couplings of and .
A quadratic optimal map from to is a measurable map with whose graph plan has cost no greater than that of any coupling of and .✦ A unique quadratic optimal map from to is a quadratic optimal map for which every coupling of and with cost at most equals .✦ For maps , the distance between them is the distance in (3): (3)✦
Quantization Data
For the limit statement, let be compact and convex, fix , and let be a probability measure supported in a compact set . Take measurable maps satisfying (4): (4) The displacement tolerances satisfy (5): (5) For a common constant , let be convex and -Lipschitz on , with the normalization and subgradient condition (6): (6) Assume also the pairwise stability bound (7) for a constant : (7)
The Claim
Theorem 1. For a compact convex source with nonempty interior and a nonempty compact target set , every probability measure supported in admits a map as in (8): (8)✦
For the quantization data satisfying (4), (5), (6) and (7), there is a map satisfying (9): (9)✦
Let be compact with nonempty interior, and let be a probability measure supported on a finite set . Suppose for every , where , and fix . There are a potential and a map satisfying (10): (10)✦
Let be compact, and let be probability measures supported in , respectively. For measurable maps and constants , suppose for -almost every and for -almost every . Then (11) holds: (11)✦
Let be compact and convex with nonempty interior, let be nonempty and compact, and fix . Suppose every pair of probability measures supported on finite subsets of admits unique quadratic optimal maps satisfying the bound in (12). For every pair of probability measures supported in , there are unique quadratic optimal maps from to the respective laws satisfying (12)✦
In the paper, the result reads: every , where denotes the probability measures supported in , admits a quadratic optimal map from to , unique up to a -null set, and if are finite -nets with and are Borel nearest-point selections, then in .
What Is Checked
The claims are stated over the uniform measure (1) and target probability laws, with uniqueness referring to the graph coupling and with the limit conditional on the stated approximation hypotheses.