PDMD: Potential-free Data-driven Molecular Dynamics for Variable-sized Water Clusters

Abstract: Conventional molecular dynamics (MD) simulation approaches, such as ab initio MD and empirical force field MD, face significant trade-offs between physical accuracy and computational efficiency. This work presents a novel Potential-free Data-driven Molecular Dynamics (PDMD) framework for predicting system energy and atomic forces of variable-sized water clusters. Specifically, PDMD employs the smooth overlap of atomic positions descriptor to generate high-dimensional, equivariant features before leveraging ChemGNN, a graph neural network model that adaptively learns the atomic chemical environments without requiring a priori knowledge. Through an iterative self-consistent training approach, the converged PDMD achieves a mean absolute error of 7.1 meV/atom for energy and 59.8 meV/angstrom for forces, outperforming the state-of-the-art DeepMD by ~80% in energy accuracy and ~200% in force prediction. As a result, PDMD can reproduce the ab initio MD properties of water clusters at a tiny fraction of its computational cost. These results demonstrate that the proposed PDMD offers multiple-phase predictive power, enabling ultra-fast, general-purpose MD simulations while retaining ab initio accuracy.