Spatio-Temporal Characterization of Water Diffusion Anomalies in Saline Solutions Using Machine Learning Force Field

Conventional classical force fields have historically struggled to accurately model water-salt interactions, often resulting in anomalous water dynamics. The emergence of machine learning force fields (MLFFs) now offers near ab-initio accuracy at reduced computational cost. This study compares Deep Potential Molecular Dynamics (DPMD), a prominent MLFF, with existing methods including ab-initio molecular dynamics and classical force fields such as SPC/Fw+JC, AMOEBA, and MB-Pol. Analysis reveals that spatio-temporal correlation is key to describing accurate water dynamics in saline solutions. This correlation can be summarized through correlation-scaled diffusivity, which provides an effective measure of molecular water movement from the perspective of correlated motion. The results demonstrate DPMD's effectiveness as a computational tool for modeling complex molecular systems, advancing our understanding of fundamental physical principles in water-salt interactions.