Continuous transformations of probability measures and their transport representations

Hugo Lavenant, Giuseppe Savaré

Given a function $F$ transforming a probability measure $μ$ into another one $F(μ)$, we study the existence and regularity of a transport representation of it. That is, we ask whether we can represent the image $F(μ)$ of the input probability measure $μ$ as the push-forward of $μ$ by a map $f(\cdot,μ)$ which may depend on $μ$; and furthermore, how regular $f$ can be chosen depending on $F$. Even if $F$ is continuous and a transport representative exists, it cannot necessarily be chosen in a continuous way; however, if $F$ is Lipschitz continuous with respect to the Wasserstein distance, then $f$ can be chosen continuous. We provide several examples to illustrate the sharpness of our assumptions. This question is motivated by approximation results for transformations of probability distributions with transformers.

Read on ELI