Calculating Position-Dependent Diffusivity in Biased Molecular Dynamics Simulations

Calculating transition rates and other kinetic quantities from molecular simulations requires knowledge not only of the free energy along the relevant coordinate but also the diffusivity as a function of that coordinate. A variety of methods are currently used to map the free-energy landscape in molecular simulations; however, simultaneous calculation of position-dependent diffusivity is complicated by biasing forces applied with many of these methods. Here, we describe a method to calculate position-dependent diffusivities in simulations including known time-dependent biasing forces, which relies on a previously proposed Bayesian inference scheme. We first apply the method to an explicitly diffusive model, and then to an equilibrium molecular dynamics simulation of liquid water including a position-dependent thermostat, comparing the results to those of an established method. Finally, we test the method on a system of liquid water, where oscillations of the free energy along the coordinate of interest preclude sufficient sampling in an equilibrium simulation. The adaptive biasing force method permits roughly uniform sampling along this coordinate, while the method presented here gives a consistent result for the position-dependent diffusivity, even in a short simulation where the adaptive biasing force is only partially converged.