The Dynamic Diversity\nand Invariance of Ab Initio\nWater

Comprehending water dynamics is crucial in various fields,\nsuch\nas water desalination, ion separation, electrocatalysis, and biochemical\nprocesses. While ab initio molecular dynamics (AIMD) accurately portray\nwater’s structure, computing its dynamic properties over nanosecond\ntime scales proves cost-prohibitive. This study employs machine learning\npotentials (MLPs) to accurately determine the dynamic properties of\nliquid water with ab initio accuracy. Our findings reveal diversity\nin the calculated diffusion coefficient (<i>D</i>) and viscosity\nof water (η) across different methodologies. Specifically, while\nthe GGA, meta-GGA, and hybrid functional methods struggle to predict\ndynamic properties under ambient conditions, methods on the higher\nlevel of Jacob’s ladder of DFT approximation perform significantly\nbetter. Intriguingly, we discovered that both <i>D</i> and\nη adhere to the established Stokes–Einstein (SE) relation\nfor all of the ab initio water. The diversity observed across different\nmethods can be attributed to distinct structural entropy, affirming\nthe applicability of excess entropy scaling relations across all functionals.\nThe correlation between <i>D</i> and η provides valuable\ninsights for identifying the ideal temperature to accurately replicate\nthe dynamic properties of liquid water. Furthermore, our findings\ncan validate the rationale behind employing artificially high temperatures\nin the simulation of water via AIMD. These outcomes not only pave\nthe path to designing better functionals for water but also underscore\nthe significance of water’s many-body characteristics.