The Setting
A finite maximum of affine functions supplies candidates for a transport map through its slopes. The question is which measurable choices of these slopes give the unique quadratic optimal transport plan.
Let with its Euclidean inner product.✦ For a source set , the uniform measure is the restriction of Lebesgue measure to , multiplied by the reciprocal of its volume: (1)✦
Let be a real inner product space, let be a finite nonempty index set, and choose sites and intercepts . The finite affine potential associated with the sites and intercepts is the function given by their maximum; for , the active set consists of the indices whose affine functions attain this maximum: (2) An index is a strict maximizer in (2) at a point when for every .
Quadratic Transport
A coupling of measures and on is a probability measure on with first marginal and second marginal ; write for these couplings.✦ The graph plan of a map is the pushforward of by ✦, and the quadratic transport cost is the extended nonnegative integral of squared distance: (3) A quadratic optimal map from to is a measurable map satisfying (4)✦ A unique quadratic optimal map from to is a quadratic optimal map in the sense of (4) for which every with equals the graph plan in (3).✦ A measure is supported in a target container when .✦
The Claim
Theorem 1. For sites in and every source set , the slopes of (2) satisfy (5): (5)✦ For equipped with its Borel -algebra, a selection exists as in (6): (6)✦ At a strict maximizer at , the potential is Fréchet differentiable, with derivative (7): (7)✦ Suppose the sites belong to and are pairwise distinct, is compact, and is a probability measure. Let be measurable and satisfy for -almost every . Then (8) holds: (8)✦ Let be nonempty, compact and convex, let be compact, and let and be finite measures, with supported in . Let be Lipschitz on and let be measurable with . Suppose that, for -almost every , for every . Then the supporting potential gives (9): (9)✦
In the paper, the result reads: for any positive integer and finite maximum in the general source setting, with repeated slopes allowed, all maximizing indices have the same slope for -almost every , and choosing the smallest maximizing index gives a Borel map with almost everywhere whose graph plan is the unique quadratic optimal plan from to .
What Is Checked
The claims are stated over finite affine families in real inner product spaces and, for transport, over the uniform measure (1) with the compactness and measure hypotheses of (8) and (9).