Dealing with moment measures via entropy and optimal transport (Q285820)

The author studies some properties of the moment measure of a convex function. The moment measure is defined as \(\mu =(\nabla u)_\sharp \rho\), where \(d\rho =e^{-u(x)}dx\). In the paper [\textit{D. Cordero-Erausquin} and \textit{B. Klartag}, J. Funct. Anal. 268, No. 12, 3834--3866 (2015; Zbl 1315.32004)], the authors studied the conditions for a measure \(\mu\) to be the moment measure for a convex function. They proved that some continuity properties of the function \(u\) on \(\partial K \) (the boundary of a proper convex set \(K \subset \mathbb{R}^d\)) are necessary for the existence of \(\mu\). In the present paper, the author proves the same existence result with a different method, replacing functional inequalities techniques with some ideas from optimal transport. First he presents the main tools from optimal transport theory such as Wasserstein distances, geodesic interpolation, etc. and provides some estimates and semicontinuity results for the entropy term. Next, he studies the strict displacement convexity properties of the entropy. In section 3, the author studies the transport cost \(\rho \mapsto \mathcal{T}(\rho, \mu)\) and its estimates and semicontinuity in a \(\mathcal{P}_1(\mathbb{R}^d)\) framework. He presents the variational problem that must be solved in order to find the log-concave measure \(\rho =e^{-u}\). He proves that \(u\) is convex and essentially continuous. Next, he shows that the condition \((\nabla u)_\sharp e^{-u}=\mu\) is equivalent to the fact that \(e^{-u}\) minimizes the considered functional, using the notion of displacement convexity. At the end, the author compares his approach to that used in [loc. cit.], explaining why they are equivalent and how to pass from one to the other.