Molecular dynamics simulations of LiCl ion pairs in high temperature aqueous solutions by deep learning potential

Abstract

Molecular dynamics simulation is an efficient method to study ion-pair association in high temperature supercritical fluid. Interatomic potentials based on neural-network machine learning shows outstanding ability of balancing the accuracy and the efficiency in molecular dynamics (MD) simulations. In present study, a neural-network potential (NNP) model for LiCl ion-pair in high temperature aqueous solutions was developed using database obtained by the first-principles density functional theory (DFT) calculations. With this NNP model, the structures of LiCl solution and the dissociation pathway of LiCl dissociation process were investigated. The results show that deep learning molecular dynamic (DPMD) simulations can accurately reproduce the radial distribution functions of ab initio MD simulations. And several metastable states were clearly identified from the dissociation 2D energy surfaces. In addition, the potential of mean force (PMF) profiles and corresponding association constants (Ka) were extensively investigated under a wide range of temperature-density (T = 330–1273 K and ρ = 0.45–1.0 g/cm3) conditions. The association constants calculated from DPMD are satisfactory compared with the experimental data. The study indicates that deep learning potential exhibits good capabilities to describe the association behavior of metal complex in high temperature aqueous solutions. This work also provides the microstructures and LiCl association constants in temperature aqueous solutions for which no experimental data exist.