Proper Thermal Equilibration of Simulations with Drude Polarizable Models: Temperature-Grouped Dual-Nosé-Hoover Thermostat.

An explicit treatment of electronic polarization is critically important to accurate simulations of highly charged or interfacial systems. Compared to the iterative self-consistent field (SCF) scheme, extended Lagrangian approaches are computationally more efficient for simulations that employ a polarizable force field. However, an appropriate thermostat must be chosen to minimize heat flow and ensure an equipartition of kinetic energy among all unconstrained system degrees of freedom. Here we investigate the effects of different thermostats on the simulation of condensed phase systems with the Drude polarizable force field using several examples that include water, NaCl/water, acetone, and an ionic liquid (IL) BMIM+/BF4-. We show that conventional dual-temperature thermostat schemes often suffer from violations of equipartitioning and adiabatic electronic state, leading to considerable errors in both static and dynamic properties. Heat flow from the real degrees of freedom to the Drude degrees of freedom leads to a steady temperature gradient and puts the system at an incorrect effective temperature. Systems with high-frequency internal degrees of freedom such as planar improper dihedrals or C-H bond stretches are most vulnerable; this issue has been largely overlooked in the literature because of the primary focus on simulations of rigid water molecules. We present a new temperature-grouped dual-Nosé-Hoover thermostat, where the molecular center of mass translations are assigned to a temperature group separated from the rest degrees of freedom. We demonstrate that this scheme predicts correct static and dynamic properties for all the systems tested here, regardless of the thermostat coupling strength. This new thermostat has been implemented into the GPU-accelerated OpenMM simulation package and maintains a significant speedup relative to the SCF scheme.