Predicting Properties of Key Ingredients of Battery Solvents with Density Functional Theory and Adaptive Force Matching.

The performance of batteries is heavily influenced by the properties of their solvents, which play a crucial role in ion solvation, conductivity, and electrochemical stability. However, traditional force field models often fall short in accurately predicting these properties. This study investigates the potential of adaptive force matching (AFM) to predict a range of solvent-related properties using only electronic structure theory. To demonstrate this approach, we developed AFM models based on B3LYP-D3(BJ) for three common molecules present in battery solvents: dimethylamine (DMA), dimethyl carbonate (DMC), and tetrahydrofuran (THF). Our results reveal that AFM models significantly outperform traditional force fields, including OPLS-AA (CM1A), GAFF2, and GROMOS 54A7, in predicting key properties such as density, heat of vaporization, viscosity, diffusion constant, boiling temperature, and free energy of vaporization. Notably, the average percentage error of AFM models is approximately 13%, substantially lower than that of traditional force fields (21% for GAFF2, the best-performing empirical model). These findings underscore the promise of AFM in predicting physical properties of battery solvents, with far-reaching implications for chemistry, materials science, and related fields where accurate predictions are essential for understanding complex phenomena and designing innovative materials and systems.