Top-down optimization of aqueous electrolyte force fields to model chemical potentials and solubilities

Electrolyte solutions are vital to the development of technologies such as batteries and carbon sequestration methods. Accurate but efficient simulation models are crucial in guiding the development of such technologies. In this work, we investigate how the inclusion of atomic polarizability, by means of Drude oscillators, affects the ability of efficient, classical force fields to model the temperature dependence of the aqueous solubility and activity coefficients of alkali-halide salts. To achieve this, we propose a new method to efficiently and accurately compute derivatives of the salt chemical potential with respect to force field parameters, enabling gradient-based fitting directly to chemical potential data. Using this method, we attempt to refine polarizable models to better predict solid-solution solubility limits as functions of temperature. We find that while solubility predictions can be improved, polarizable models are incapable of reproducing the slope of the solid-solution coexistence line for NaCl. This implies that classical polarizability (with constant atomic charges) alone is an insufficient description of system many-body interactions, and “first-principles” descriptions of the energy landscape are necessary to achieve true predictive power in electrolyte modeling.