Duality on gradient estimates and Wasserstein controls (Q971830): Difference between revisions

There are several ways to formulate a quantitative estimate on rate of convergence to equilibrium. An \(L^ q\)-gradient estimate \(|\nabla P_ tf|(x)\leq e^{-kt}P_ t\left(|\nabla f|^ q\right)(x)^{1/q}\) for a heat semigroup \(P_ t\) is a very powerful tool. It implies several functional inequalities such as Poincaré inequalities, when \(q=2\), and logarithmic Sobolev inequalities, when \(q=1\), which quantify convergence rates. As a different approach to this problem, there is a contraction of \(L^ p\)-Wasserstein distance \(d^ W_ p(\mu_ t,\nu_ t)\leq e^{-kt} d^ W_ p(\mu_ 0,\nu_ 0)\) for two diffusions \(\mu_ t\) and \(\nu_ t\) of masses when \(p=2\). The main result of this paper is to establish a duality between \(L^ p\)-Wasser\-stein control and \(L^ q\)-gradient estimate in a general framework for \(p\), \(q\in[1,\infty]\) with \(p^{-1}+q^{-1}=1\). This result extends a known result for a heat flow on a Riemannian manifold. The author derives a Wasserstein control of a heat flow directly from the corresponding gradient estimate of the heat semigroup without using any other notion of lower curvature bound. Also, they apply their result to a sub-elliptic heat flow on a Lie group to obtain a coupling of heat distributions which carries a good control of their relative distance.