Multi- to one-dimensional transportation
Abstract
Fix probability densities f and g on open sets X ⊂ Rm and Y ⊂ Rn with m n1. Consider transporting f onto g so as to minimize the cost -s(x,y). We give a non-degeneracy condition (a) on s ∈ C1,1 which ensures the set of x paired with [g-a.e.] y∈ Y lie in a codimension n submanifold of X. Specializing to the case m>n=1, we discover a nestedness criteria relating s to (f,g) which allows us to construct a unique optimal solution in the form of a map F:X Y. When s∈ C2 W3,1 and f and g are bounded, the Kantorovich dual potentials (u,v) satisfy v ∈ C1,1loc(Y), and the normal velocity V of F-1(y) with respect to changes in y is given by V(x) = v"(f(x))-syy(x,f(x)). Positivity (b) of V locally implies a Lipschitz bound on f; moreover, v ∈ C2 if F-1(y) intersects ∂ X ∈ C1 transversally (c). On subsets where (a)-(c) can be be quantified, for each integer r 1 the norms of u,v ∈ Cr+1,1 and F ∈ Cr,1 are controlled by these bounds, || f, g, ∂ X ||Cr-1,1, ||∂ X||C1,1, ||s||Cr+1,1, and the smallness of F-1(y). We give examples showing regularity extends from X to part of X, but not from Y to Y. We also show that when s remains nested for all (f,g), the problem in Rm × R reduces to a supermodular problem in R × R.
Turn this paper into a lesson
ArcXiv compiles a structured reading guide from this paper's metadata: plain-English importance, contributions, prerequisite concepts, which sections to read first, flashcards, and a quiz. Grounded in the abstract, never invented.