Wasserstein space over the Wiener space (Q846747): Difference between revisions

Define on an abstract Wiener space \((X,H,\mu)\) the lower semicontinuous distance \(d_H(x,y)=|x-y|_H\) if \(x-y\in H\) and \(d_H(x,y)=\infty\) otherwise. On the space of probability measures \({\mathcal P}(X)\) on \(X\) one considers the Wasserstein distance \[ W_s(\nu_1,\nu_2)^2:= \inf\left\{ \int_{X\times X} |x-y|_H^2\pi(dx,dy),\;\pi\in{\mathcal P}(X\times X),\;\pi(X\times\cdot)=\nu_2, \pi(\cdot\times X)=\nu_1\right\}. \] The purpose of the present paper is to study the geometry and the structure of gradient flows of the Wasserstein space \(({\mathcal P}_2(X),W_2)\); here \({\mathcal P}_2(X)\) are the probability measures having finite second moments. The paper is based on earlier work by \textit{L. Ambrosio} and \textit{G. Savaré} [``Gradient flows of probability measures'', Handbook of differential equations: Evolutionary equations. Vol. III. Amsterdam: Elsevier/North-Holland. Handbook of Differential Equations, 1--136 (2007; Zbl 1203.35002)] where for the first time the tangent space of the Wasserstein space is introduced and on the paper by \textit{D. Feyel} and \textit{A. S. Üstünel} [Probab. Theory Relat. Fields 128, No.~3, 347--385 (2004; Zbl 1055.60052)] who study the Monge-Kantorovich optimal transportation problem on Wiener space. The present paper studies the connection between the convexity of the entropy functional relative to the Riemannian volume and the lower bound of the Ricci curvature in the framework of Wiener space. The authors introduce the derivative processes associated to absolutely continuous curves which allows to express the distance \(W_2\) as a Riemannian distance and which gives a new interpretation of the Benamou-Brenier formula. The gradient flow associated to a general convex functional is defined through sub-gradients. In this paper the directional derivative is explicitly calculated and it is shown that the gradient of the entropy functional exists at the minimizers in the Jordan-Kinderlehrer-Otto approximation scheme. Finally, it is shown that the solutions of the Ornstein-Uhlenbeck evolution is the gradient flow associated to the entropy functional.