Encoding electronic structure information in potentials for multi-scale simulations: SiO2

Potentials generally used in molecular dynamics (MD) simulation of SiO2 properties customarily are calibrated to a combination of computed molecular electronic structure data and experimental crystalline data. The present study tests parametrization to data from high-level, first-principles electronic structure calculations alone. The issue is crucial to the success of multi-scale simulations. They require a consistent embedding of the so-called quantum mechanical region (the region in which the forces come from gradients of quantum mechanical total energies) in a classical inter-ionic potential region. The evident challenge is generation of a quantum mechanically consistent parametrization. A simple probe of the issue is to see how parametrization solely from first-principles data influences the simulation outcomes. We parametrized a widely used form of effective inter-ionic potential for SiO2 and did MD simulations of tensile failure in a 72 formula unit SiO2 nanorod. Separate parametrizations were done to high quality calculated data for H4SiO4 and H6Si2O7 clusters and for α-quartz. The differing parametrizations yield quantitative differences in the prediction of the yield strength and even semi-qualitative differences in the system behavior in that region. Some superficially similar parametrizations do not even provide a stable T = 0 K configuration. These differences highlight the crucial distinction between potential parametrization aimed at replacing realistic quantum mechanical forces entirely in an MD calculation versus a parametrization aimed at embedding an explicitly QM region.