Breathing New Life into Classical Force Fields

Abstract

Classical force fields form the foundation of molecular dynamics simulations. Because classical force fields are limited to pairwise-additive Lennard-Jones potentials and fixed-charge electrostatics, they cannot accurately capture explicit polarization and charge-transfer effects, metal coordination bonds, or weak, directional nonbonded interactions such as $π$–$π$ stacking. Machine-learning force fields and polarizable force fields, which have been claimed to accurately characterize these types of interactions, have developed rapidly in recent years, raising an ongoing debate over whether further improvements to classical force fields are still necessary. Keeping this question in mind, we review various strategies proposed in recent years to improve the description of nonbonded interactions for classical force fields. These advances include the development of atomic charge models that accurately reflect molecular dipoles, water models that capture electrostatic screening and solvation effects, atom-pair-specific van der Waals parameters to mimic polarization, refined atom types that better represent chemical environments, and the introduction of virtual sites for accurately modeling of lone pairs. We also discuss the applicable scope of each strategy. Furthermore, by comparing classical force fields with polarizable and machine-learning-based force fields, we identify three major advantages of classical force fields: high computational efficiency, strong transferability, and ease of parameterization. Therefore, we join the ongoing debate and believe that the community will greatly benefit from further development of classical force fields. Through the improvement of the accuracy of characterizing nonbonded interactions, classical force fields can be widely applied in areas such as million-atom simulations and high-throughput screening.