A concurrent multiscale method based on smoothed molecular dynamics for large-scale parallel computation at finite temperature

Abstract

A concurrent atomic-to-continuum multiscale method for finite-temperature simulation with large-scale parallel computation is developed. Seamless and stable coupling between molecular dynamics (MD) method and continuum-based material point method (MPM) is achieved with the aid of mesoscopic smoothed molecular dynamics (SMD) method. Novel techniques are proposed to filter spurious high-frequency reflections and bridge different thermal descriptions in atomistic and continuum models by adopting the efficient Markovian generalized Langevin equation (GLE) with newly designed drift matrix. Besides, an effective relaxation strategy is developed to fully relax the multiscale model to thermodynamic equilibrium state. The concurrent multiscale method can be effectively parallelized within the parallel computation framework of particle-grid dual discretization, and its efficiency when executing on large-scale computation clusters is analyzed in detail. The contact-sliding and particle impact examples are calculated with the proposed method to demonstrate its superiority as an effective numerical tool in the practical applications.