Section: New Results

Numerical analysis for the ARPS method

Participants : Zofia Trstanova, Gabriel Stoltz, Stephane Redon.

Previous works have led to understanding of the choice of optimal parameters for the ARPS dynamics. The interest lies in achieving the highest percentage of restrained particles, while minimizing the modification of the variance and the systematic error. We study discretization schemes of the ARPS-Langevin dynamics, such that the systematic error remains of second order in the time step size and we introduce a Metropolis step in order to stabilize the simulations and hence to allow "a sharper" choice of the ARPS parameters, which lead to better algorithmic speed-ups.