Convexity inequalities and optimal transport of infinite-dimensional measures (Q1775750)

The mass transportation problem was formulated by Monge in 1781 and consists of moving a good from a location to another one in the Euclidean space \({\mathbb R}^n;\) the amounts at each location are described by densities \(h(x), k(x)\) with \(\int h \,dx = \int k \,dx,\) and the transportation is to be done minimizing a suitable defined cost. This problem was generalized by Kantorovich in 1942 and later years. In his formulation, the densities are probability measures \(\nu_1, \nu_2\) on a measurable space \(X\) and the problem consists of minimizing the functional \[ \int_{X \times X} L(x_1, x_2) \nu(dx_1 dx_2) \] over the set \({\mathcal P}(\nu_1, \nu_2)\) of probability measures \(\nu\) on \(X \times X\) with marginals \(\nu_1, \nu_2,\) where the positive function \(L\) is called the \textit{cost function}; the minimum of the (square root of) the functional is the \textit{cost } \(W_L(\nu_2, \nu_2).\) An inequality due to Talagrand gives the bound \[ W_L(\gamma, \mu)^2 \leq 2 \int_X f(x) \log f(x) \gamma(dx) \] in the particular case where \(X = {\mathbb R}^n,\) \(\gamma\) is the standard Gaussian measure on \({\mathbb R}^n,\) \(\mu\) is a probability measure of the form \(\mu = f \gamma\) and \(L\) is a quadratic function. The author generalizes Talagrand's inequality, shows some applications and discusses the relation of the Monge-Kantorovich problem with other problems in analysis and probability.