Use the force! Reduced variance estimators for densities, radial distribution functions, and local mobilities in molecular simulations

Abstract

Even though the computation of local properties, such as densities or radial distribution functions, remains one of the most standard goals of molecular simulation, it still largely relies on straightforward histogram-based strategies. Here, we highlight recent developments of alternative approaches leading, from different perspectives, to estimators with a reduced variance compared to conventional binning. They all make use of the force acting on the particles, in addition to their position, and allow us to focus on the non-trivial part of the problem in order to alleviate (or even remove in some cases) the catastrophic behavior of histograms as the bin size decreases. The corresponding computational cost is negligible for molecular dynamics simulations, since the forces are already computed to generate the configurations, and the benefit of reduced-variance estimators is even larger when the cost of generating the latter is high, in particular, with ab initio simulations. The force sampling approach may result in spurious residual non-zero values of the density in regions where no particles are present, but strategies are available to mitigate this artifact. We illustrate this approach on number, charge, and polarization densities, radial distribution functions, and local transport coefficients, discuss the connections between the various perspectives, and suggest future challenges for this promising approach.