Abstract
Langevin diffusion processes and their discretizations are often used forsampling from a target density. The most convenient framework for assessing thequality of such a sampling scheme corresponds to smooth and stronglylog-concave densities defined on $\mathbb R^p$. The present work focuses onthis framework and studies the behavior of Monte Carlo algorithms based ondiscretizations of the kinetic Langevin diffusion. We first prove the geometricmixing property of the kinetic Langevin diffusion with a mixing rate that is,in the overdamped regime, optimal in terms of its dependence on the conditionnumber. We then use this result for obtaining improved guarantees of samplingusing the kinetic Langevin Monte Carlo method, when the quality of sampling ismeasured by the Wasserstein distance. We also consider the situation where theHessian of the log-density of the target distribution is Lipschitz-continuous.In this case, we introduce a new discretization of the kinetic Langevindiffusion and prove that this leads to a substantial improvement of the upperbound on the sampling error measured in Wasserstein distance.