Modeling Intermolecular\nInteractions with Exchange-Hole\nDipole Moment Dispersion Corrections to Neural Network Potentials

Neural network potentials (NNPs) are an innovative approach\nfor\ncalculating the potential energy and forces of a chemical system.\nIn principle, these methods are capable of modeling large systems\nwith an accuracy approaching that of a high-level ab initio calculation,\nbut with a much smaller computational cost. Due to their training\nto density-functional theory (DFT) data and neglect of long-range\ninteractions, some classes of NNPs require an additional term to include\nLondon dispersion physics. In this Perspective, we discuss the requirements\nfor a dispersion model for use with an NNP, focusing on the MLXDM\n(Machine Learned eXchange-Hole Dipole Moment) model developed by our\ngroups. This model is based on the DFT-based XDM dispersion correction,\nwhich calculates interatomic dispersion coefficients in terms of atomic\nmoments and polarizabilities, both of which can be approximated effectively\nusing neural networks.