Abstract

We investigate specific ion effects on the structure and dynamics of water in aqueous electrolyte solutions using neural network potentials trained on DFT predictions using the SCAN functional. Ion specificity of bulk electrolyte solutions following the Hofmeister series can induce wide-ranging effects on the dynamics of water – kosmotropic ions mitigate water mobility and chaotropic ions accelerate water molecules. Many existing studies of this phenomena apply conventional empirical models which are inherently limited in transferability and accuracy across a range of concentrations. In this work, in order to explore ion solvation characteristics with a first principles level of accuracy, we train neural networks to learn highly complex and multi-dimensional interactions native to DFT representations. We find that these potentials overcome the limitations of conventional empirical force fields in representing water dynamics with concentration dependence. We then use the potentials to probe the underlying mechanisms of ion-induced water structure and mobility for a series of alkali halide ions (KCl, CsCl, NaBr, NaCl) in bulk solution.