Parallel approaches to long-time atomistic simulations: decomposition, replication, and speculation

Molecular Dynamics (MD) is a workhorse of computational materials science. Indeed, MD can in principle be used to obtain any thermodynamic or kinetic quantity for a given interatomic potential. This enviable quality however comes at a steep computational price, limiting the system sizes and simulation times that can be achieved in practice. While the size limitation can be efficiently addressed with massively parallel implementations of MD based on spatial decomposition strategies, the same approach cannot extend the timescales much beyond microseconds. This is a significant issue, as this implies that the constant increase in the computing power delivered by leadership-scale machines cannot be leveraged to make long-time predictions.
In this talk, I discuss additional parallelization strategies, namely replication and speculation, that can be used to address the timescale limitation of MD for systems that evolve through rare transitions, concentrating primarily on the Parallel Trajectory Splicing (ParSplice) method and its recent developments. Using as an example the evolution of the shape of metallic nanoparticles, we show how, taken together, these ideas can significantly extend the simulation space accessible to MD in a way that can unlock the potential of current massively-parallel computing platforms.