Stability of optimal transference mappings (Exercise 2.17 in Villani's Topics in Optimal Transportaiton)

136 Views Asked by At

I'm really stuck on the first part of this exercise. I've tried to type it up in a fashion that is as self-contained as possible, but some background in optimal transport is likely required to read this.

Let $P_{ac,2}(\mathbb{R}^n)$ denote the space of absolutely continuous probability measures with finite second moments. Suppose $\sigma \in P_{ac,2}(\mathbb{R}^n)$ is given, as well as a family $\rho_k \in P_{ac,2}(\mathbb{R}^n)$ that weakly converges to some $\rho \in P_{ac,2}(\mathbb{R}^n)$.

Let $\nabla \varphi_k$ be the optimal mapping of the Monge problem with quadratic cost between $\sigma$ and $\rho_k$, i.e. $$\int_{\mathbb{R}^n} |x-\nabla \varphi_k(x)|^2 d\sigma(x) = \inf_{T \# \sigma = \rho_k} \int_{\mathbb{R}^n} |x - T(x)|^2 d\sigma(x)$$ Here, the infimum is taken over functions $T : x \mapsto T(x)$ satisfying the pushforward constraint. Similarly, we can define $\nabla \varphi$ between $\sigma$ and $\rho$.

Let $d\pi_k(x,y) = d\sigma(x) \delta(y = \nabla \varphi_k(x))$, and similarly define $\pi$ between $\sigma$ and $\rho$.

How can I show that $\pi_k$ weakly converges to $\pi$? (The hint given is to note that the optimal transference plan is unique.)