Some applications of mass transport to Gaussian-type inequalities (Q1349262)

The author makes use of two ``fine properties'' of transport mappings to give new proofs of some geometric inequalities. The paper begins by recalling that mapping through a convex gradient gives the optimal transport in \({\mathbb R}^N\) [see \textit{Y. Brenier}, Commun. Pure Appl. Math. 44, No. 4, 375-417 (1991; Zbl 0738.46011)]. More precisely, let \(f,g\) be two nonnegative functions on \({\mathbb R}^N\) with \(\int f=\int g=1\) and let \(\nabla \varphi\) be the Brenier mapping pushing \(f(x)dx\) forward to \(g(y)dy\), then by definition of transport \[ \int b(y)g(y) dy =\int b(\nabla\varphi(x))f(x) dx \] for any nonegative Borel function \(b\) on \({\mathbb R}^N\). The last formula is a weak form of the Monge-Ampere equation \[ f(x)=\det (H_x \varphi)g (\nabla\varphi(x)). \] The first ``fine property'' used by the author is that \(\varphi\) actually solves this last equation pointwise a.e. [see \textit{R. J. McCann}, Adv. Math. 128, No. 1, 153-179 (1997; Zbl 0901.49012)] where the Hessian is interpreted in the sense of Alexandrov. The use of the equation leads the author to prove the log-Sobolev, transport and HWI inequalities involving entropies, Fisher information and Wasserstein distances. The second ``fine property'' was discovered by \textit{L. A. Caffarelli} [Commun. Math. Phys. 214, No. 3, 547-563 (2000; Zbl 0978.60107)] and it says that if \(\rho\) is a Gaussian probability measure and \(\mu\) is another probability measure of the form \(f(x)d\rho\) with \(\log f \) concave, then the Brenier mapping \(\nabla \varphi\) is 1-Lipschitz. This property is used to prove that if \(K\) is a convex set and \(C\) is a strip in \({\mathbb R}^N\), then \[ \gamma_N (K\cap C) \geq \gamma_N(K) \gamma_N(C) \] where \(\gamma_N\) is the \(N\)-dimensional Gaussian measure. This is a particular case of the so-called Gaussian correlation conjecture which states the inequality for \(K\) and \(C\) symmetric convex sets.