Many-Body Expansion on Solvated Ion Pairs

Building on our previous work on improving the accuracy and stability of many-body potential energy functions, we present the development of the MB-nrg framework for solvated ion pairs. This effort represents an extension of the many-body expansion formalism to heterogeneous molecular systems, aiming to accurately describe ion–water and ion–ion interactions in aqueous environments.

The new MB-nrg models combine high-level electronic structure data with the numerically stable permutationally invariant Fourier series (PIFS) representation introduced in our earlier studies. The PIFS-based formulation ensures physically consistent behavior even for out-of-distribution configurations, addressing the hallucinatory instabilities commonly observed in machine-learning-driven molecular potentials.

Preliminary results indicate that the resulting MB-nrg ion-pair models capture key structural and thermodynamic features of solvated ions with near ab initio accuracy while maintaining numerical stability across a range of thermodynamic conditions. These findings suggest that the many-body expansion, combined with PIFS representations, provides a promising path toward systematically extending predictive modeling from neat water to more complex electrolyte solutions.