Prediction of induced fluxes in reverse nonequilibrium molecular dynamics

Abstract

Reverse nonequilibrium molecular dynamics (RNEMD) simulations impose a flux by swapping the velocities of two particles. This method allows for the calculation of transport coefficients, such as thermal conductivity and viscosity. The relation between the induced fluxes and the control parameters of RNEMD (such as the time interval between successive swap events) is not clear. Thus, trial-and-error is required to realize the desired fluxes in RNEMD simulations. In this study, we develop a theoretical framework using extreme value statistics to estimate the relation between the time interval and the resulting induced fluxes. Our RNEMD simulations, conducted with varying time intervals, confirm that the theoretical predictions are quantitatively consistent with the simulation results when the time interval exceeds the momentum relaxation time. Our RNEMD simulations also show that our theoretical predictions, which are valid for a large number of particles for swap candidates, work well even for a relatively small number of particles for swap candidates. These findings demonstrate that the induced fluxes can be reliably estimated, providing a valuable tool for selecting appropriate RNEMD parameters for simulations.