Quadratic optimal transportation problem with a positive semi definite structure on the cost function

TL;DR

This paper introduces a quadratic variation of the optimal transportation problem with a positive semi-definite cost structure, analyzing its properties and establishing a duality formula.

Contribution

It proposes a quadratic transportation problem with a semi-definite cost function and derives the Kantorovich duality for this new formulation.

Findings

Defined quadratic transportation problem with semi-definite cost

Proved properties of solutions to the quadratic problem

Established Kantorovich duality for squared cost functions

Abstract

Optimal transportation problem seeks for a coupling $\pi$ of two probability measures $\mu$ and $\nu$ which minimize the total cost $\int c d\pi$, which is linear in $\pi$. In this paper, we introduce a variation of optimal transportation problem which we call quadratic transportation problem that considers a total cost $\iint c d\pi d\pi$ which is quadratic in $\pi$. We compare this problem with other variations of optimal transportation problem, and prove some properties of the solutions to the problem. Then, we introduce squared cost function, which let us consider the total cost $\iint c d\pi d\pi$ as a positive semi-definite bilinear operator on probability measures, and show Kantorovich duality formula when we have a squared cost function.