Coupled-Cluster Theory for Water Empowered by Quantum Fragmentation and Deep Learning.

Water exhibits many anomalous properties, which have given rise to uninterrupted interest in it. A large number of theoretical models have been proposed to elucidate the unique behavior of water, yet the long-sought-after "universal water model" capable of describing the properties of water across diverse thermodynamic conditions remains elusive. Herein, combining advanced fragment-based quantum mechanical calculation and deep learning approach, we present a Deep neural network Potential at the Coupled Cluster considering single, double, and perturbative triple excitations (CCSD(T)) level of theory (DP-CC) to represent the first-principles potential energy surface of water. Utilizing the DP-CC potential, our simulations demonstrate overall good agreement with experimental observations regarding the structural, dynamical, and thermodynamic properties of liquid water and ice Ih. The DP-CC potential provides a reliable way for efficient coupled-cluster-level simulations of liquid water and ice Ih, which may contribute to a better understanding of the origin of the anomalous water properties.