(508d) Melting Points of Alkali Chlorides Evaluated for a Polarizable and Non-Polarizable Model

The field of molten salts has experienced a recent resurgence due to potential applications in the energy sector. In particular, molten salt reactors are among the leading candidates for next-generation nuclear reactor designs. As such, research from the 1970s and 1980s is being revisited with improved experimental techniques and substantially increased computational power. As part of this effort, we are focused on benchmarking the performance of classical molecular models of molten salts across a range of thermodynamic and transport properties with modern computational methods and resources. Our first step has been to investigate the alkali halides. As 1:1 mixtures of monovalent positive and negative ions, the alkali halides are one of the simplest charged systems found in nature and a natural prerequisite to studying more complex and technologically relevant systems including divalent salts and complex salt mixtures including actinides, metal ions, etc.

Here we report the performance of two models for predicting the melting points of the alkali chlorides. These include the non-polarizable rigid ion model (RIM) and the polarizable ion model (PIM). Though an important thermodynamic property in its own right, the accuracy of a model’s melting point also represents a stringent test of model quality. The melting point arises from a delicate balance of the liquid and solid free energies, and errors of <1 kcal/mol in the solid-liquid free energy difference can result in errors >100 K in the melting point. The melting points for the RIM are calculated via two distinct methods: the pseudo-supercritical path method and the direct coexistence method. We then proceed to calculate the melting points of the PIM with the direct coexistence method alone.

The pseudo-supercritical path and direct coexistence methods show excellent agreement with each other; RIM melting points predicted with the two methods agree within 5 K for all four alkali chlorides. Nonetheless, when compared with experimental results, the RIM only predicts accurate (within 10 K) melting points for NaCl and KCl. The melting point of LiCl (RbCl) is under (over) predicted by ~100 K. The addition of polarizability in the PIM does not yield more accurate results; the predictions deviate from experiment by ~10 K - 140 K. To better understand the origins of the deviations from experiments, we report and discuss the effects of entropy and enthalpy of melting on the predicted melting points. We find that both models nearly always fail to accurately predict these properties. More often, the accuracy of model melting points are improved by a fortuitous cancellation of errors. We close by discussing how targeting an accurate enthalpy of melting during force field parameterization may offer one approach to improving molecular models of the alkali chlorides.