The aim of this two-hour lecture is to present the mathematical underpinnings of some common numerical approaches to compute average properties as predicted by statistical physics. The first part provides an overview of the most important concepts of statistical physics (in particular thermodynamic ensembles). The aim of the second part is to provide an introduction to the practical computation of averages with respect to the Boltzmann-Gibbs measure using appropriate stochastic dynamics of Langevin type. Rigorous ergodicity results as well as elements on the estimation of numerical errors are provided. The last part is devoted to the computation of transport coefficients such as the mobility or autodiffusion in fluids, relying either on integrated equilibrium correlations à la Green-Kubo, or on the linear response of nonequilibrium dynamics in their steady-states.[-]

Material properties of soft matter are governed by a delicate interplay of energetic and entropic contributions. In other words, generic universal aspects are as relevant as local chemistry specific properties. Thus many different time and length scales are intimately coupled, which often makes a clear separation of scales difficult. This introductory lecture will review recent advances in multiscale modeling of soft matter. This includes different approaches of sequential and concurrent coupling. Furthermore problems of representability and transferability will be addressed as well as the question of scaling of time upon coarse graining. Finally some new developments related to data driven methods will be shortly mentioned.[-]

Deep generative models parametrize very flexible families of distributions able to fit complicated datasets of images or text. These models provide independent samples from complex high-distributions at negligible costs. On the other hand, sampling exactly a target distribution, such as Boltzmann distributions and Bayesian posteriors is typically challenging: either because of dimensionality, multi-modality, ill-conditioning or a combination of the previous. In this talk, I will review recent works trying to enhance traditional inference and sampling algorithms with learning. I will present in particular flowMC, an adaptive MCMC with Normalizing Flows along with first applications and remaining challenges.[-]