Computing equilibrium properties by a dissipative non-equilibrium process

A dissipative algorithm called quench was proposed that allows one to compute an equilibrium density of states (DOS) using a non-equilibrium estimator; this method was previously successfully applied to a mean field spherical Ising model. Multiple equilibrium starting points are drawn from a high temperature distribution that is easy to sample and quenched independently to low temperature regions that are usually difficult to sample in plain simulations. Here, we extend this method to molecular systems to a compute free energy surface (FES) rather than a DOS at any intermediate temperature. We implement this method in LAMMPS, which makes it possible for almost every molecular system and run simulations in parallel using Parsl to allow these independent quenches to be run simultaneously on high performance computing (HPC) resources. We investigate the influence of simulation parameters on our results and suggest heuristic choices of simulation parameters for general application. Our results demonstrate that sampling is focused properly in low free energy regions. In order to get an accurate FES at all energy levels, we couple quenching to umbrella sampling, and find that it can approximately accelerate convergance by 50 times for some model molecular systems.