So pleased to get this preprint out. Feel like we’ve finally worked out how to do something I’ve been trying to do for 13 years since the start of my PhD: Build an accurate continuum solvent model of ion-ion interactions in solution.

We combine equivariant neural network potentials (NNP) and density corrected r2SCAN to run highly accurate molecular simulation of lithium chloride solution. The NNP can then also be used to coarse grain out the solvent degrees of freedom, i.e., build a continuum solvent model:

Surprisingly we see the formation of lithium cation pairs! Classical MD shows no sign of this species which should not be possible given lithium ions are the archetypal strongly solvated species.

We can reproduce key structural, thermodynamic and kinetic properties of electrolytes down to infinite dilution. From a really short first principles simulation at one concentration. We also seem to get an almost perfect water model essentially for free.

Most promisingly we can reproduce activity/osmotic coefficients with two different methods. These are one of the most important quantities in physical chemistry. They are key to determining chemical equilibria, solubility, reaction rates etc.

Incidentally the classic model of ion-ion interactions was built by Debye and Hückel 100 years ago this year. Getting these right is a necessary first step towards accurately simulating the many vital systems where electrolyte solutions play a critical role: biology, batteries.