A transferrable range-separated force field for water: Combining the power of both physically-motivated models and machine learning techniques.

An accurate, transferrable, and computationally efficient potential energy surface is of paramount importance for all molecular mechanics simulations. In this work, by using water as an example, we demonstrate how one can construct a reliable force field by combining the advantages of both physically motivated and data-driven machine learning methods. Different from the existing water models based on molecular many-body expansion, we adopt a separation scheme that is completely based on distances, which is more convenient for generic molecular systems. The geometry dependence of atomic charges and dispersion coefficients are also introduced to improve the accuracy of the long-range part of the potential. The new potential provides a physically interpretable energy decomposition, and it is more accurate than the conventional physically motived potentials. Most importantly, through this study, we show that the information we learn from small clusters can be extrapolated into larger systems, thus providing a general recipe for the intermolecular force field development at the coupled-cluster singles and doubles plus perturbative triples level of theory in the future.