Gradient-based fitting of empirical potentials in the presence of a distributed-multipole electrostatic model

Abstract

The use of distributed multipoles to describe the electrostatic potential energy of a molecular solid offers the possibility of accurate atomistic simulation of even highly polar molecular crystals. This work is concerned with empirically optimizing exp-6 potential parameters for use in conjunction with this electrostatic model. The use of a potential-gradient-based fitting function is critically examined and is found to perform satisfactorily for the strongest interactions in the data set but fails when parameters for the weaker interactions are fitted.