Structural insights into sodium chloride solutions from state-of-the-art neural network potentials

Sodium chloride (NaCl) solutions, also known as saltwater, are ubiquitous in nature. Understanding the effect of the dissolved Na⁺ and Cl^- ions on the tetrahedral hydrogen-bond network of water is essential to uncover the mechanisms underlying various physical, chemical, biological, and geological processes. Although the effect of ions on water has long been the focus of scientific interest during the past century, the way and the extent of ionic effects on the H-bond network of water is still an unsolved problem. In this work, we study the effect of increasing NaCl solute concentration on the structure of solvent water and compare it with the effect of increasing pressure on pure water using deep potential molecular dynamics (DPMD). In particular, the deep neural network potential is trained with the density functional theory data based on the strongly constrained and appropriately normed functional. Therefore, our DPMD simulations preserve the quantum mechanics accuracy with computational costs comparable to that of empirical force fields, which enables efficient simulations of different concentrations with large simulation cells and long simulation time. The computed reciprocal-space structure factors agree quantitatively with experimental neutron diffraction data. The detailed analyses suggest that ion-induced modifications to the structure of water are mainly restricted to the ionic first solvation shells.