Paper detail

Quasi Monte Carlo inverse transform sampling for phase space conserving Lagrangian particle methods and Eulerian-Lagrangian coupling

This article presents a novel and practically useful link between geometric integration, low-discrepancy sampling and code coupling for Lagrangian and Eulerian Vlasov-Poisson solvers. Low-discrepancy sequences, also called quasi-random sequences (Quasi Monte Carlo), provide convergence rates close to $\mathcal{O}( N^{-1})$ which are far superior to (pseudo) random numbers (Monte Carlo) settling in at only $\mathcal{O}(N^{-0.5})$. Lagrangian particle methods such as PIC rely on Monte Carlo integration. The particle distributions are nonlinearly perturbed by the forward flow following the characteristics. Hence it remains the question of whether particle methods can benefit from such quasi-random-sequences. Any nonlinear measure-preserving map does not affect the low-discrepancy of a QMC sequence such that the order of convergence remains. This article shows that the forward flow of phase space-conserving geometric particle methods induces naturally such a measure-preserving map underlying their importance in a new framework. In this context the Hardy Krause Variation is observed to increase in the Vlasov-Poisson system for the first time. with the linear phase. Also the star discrepancy is presented for an entire PIC simulation. On the other hand, Eulerian and Lagrangian solvers have different strengths and weaknesses, such that we present a novel way of transiting from a spectral discretization of the Vlasov--Poisson system to a PIC simulation. This is achieved by higher dimensional inverse transform sampling (Rosenblatt-Mück transform). In this way Markov Chain Monte Carlo techniques are circumvented which allows the use of pseudo and quasi-random numbers. In the latter case better convergence rates can be observed both in the linear and nonlinear phase.