The Poincaré inequality and quadratic transportation-variance inequalities (Q2285794)

Given a connected, complete, Riemannian manifold \(E\) with geodesic distance \(d\), the quadratic Wassserstein distance between probability measures \(\mu\) and \(\nu\) on \(E\) is defined by \[ W_2(\mu,\nu)=\left(\inf_\pi\int_{E\times E}d^2(x,y)\pi(dx,dy)\right)^{1/2}\,, \] where the infimum is taken over all bivariate probability measures with marginals \(\mu\) and \(\nu\). The author proves several characterizations of the Poincaré inequality on \(E\) in terms of inequalities satisfied by the quadratic Wasserstein distance. For example, it is already known that the Poincaré inequality is equivalent to the quadratic transportation-variance inequality \(W_2^2(f\mu,\mu)\leq C_V\mbox{Var}_\mu(f)\), where \(f\mu\) is a probability measure on \(E\). Several other analogous characterizations of the Poinacaré inequality are established in this paper, and some improvements to known results are presented (for example, the estimate of \(C_V\) in the above is sharpened to twice the constant in the Poincaré inequality). Two separate proofs of these characterizations are presented. The other main result of the present paper is an upper bound for \(W_2^2(f\mu,\mu)\) in terms of the centralized version \(W_2^2(f_c\mu,\mu)\), where \[ f_c=\frac{|\sqrt{f}-\mu(\sqrt{f})|^2}{\mbox{Var}_\mu(\sqrt{f})}\,. \] This gives bounds on \(W_2^2(f\mu,\mu)\) in terms of the variance of \(\sqrt{f}\), rather than the variance of \(f\).