Influence of integration formulations on the performance of the fast inertial relaxation engine (FIRE) method

Abstract

As a simple and robust minimization algorithm with high efficiency, the fast inertial relaxation engine (FIRE) method has been widely applied in various fields. Different from the viewpoint of previous reports, the present study found out that integration formulations in the FIRE method have huge influence on its convergence performance and capability of critical analysis. Specifically, the Forward Euler (FE) integration is found ill-suited for all the trial applications due to the outdated velocity update; the Velocity Verlet (VV) integration shows robust convergence and superior efficiency, but loses the ability of critical analysis; the Semi-implicit Euler (SE) integration endows the FIRE method with the critical analysis ability as well as good efficiency, but its convergence is conditional. It is also found that the FIRE method using a modified energy monitor shows more robust convergence than using the original power monitor. Further investigation indicates that the SE integration combined with the energy monitor should be the first choice for the FIRE method in general molecular statics simulations. These findings extend the capability of FIRE and provide practical suggestions for selecting minimization algorithms in molecular simulations.