Benchmarking thermostat algorithms in molecular dynamics simulations of a binary Lennard-Jones glass-former model

Abstract

A systematic comparison was carried out to assess the influence of representative thermostat methods in constant-temperature molecular dynamics simulations. The thermostat algorithms considered include the Nosé–Hoover thermostat and its chain generalization, the Bussi velocity rescaling method, and several implementations of the Langevin dynamics. Using a binary Lennard-Jones liquid as a model glass former, we investigated how the sampling of physical observables, such as particle velocities and potential energy, responds to changes in time step across these thermostats. While the Nosé–Hoover chain and Bussi thermostats provide reliable temperature control, a pronounced time-step dependence was observed in the potential energy. Among the Langevin methods, the Grønbech-Jensen–Farago scheme provided the most consistent sampling of both temperature and potential energy. Nonetheless, Langevin dynamics typically incurs approximately twice the computational cost due to the overhead of random number generation and exhibits a systematic decrease in diffusion coefficients with increasing friction. This study presents a broad comparison of thermostat methods, offering practical guidance for the choice of thermostats in classical molecular dynamics simulations. These findings provide useful insights for diverse applications, including glass transition, phase separation, and nucleation.