Lessons Learned from Semiempirical Methods for the Li-Ion Battery Solid Electrolyte Interphase

Studying the chemical reactivity related to the solid electrolyte interphase (SEI) in lithium-ion batteries is challenging due to system heterogeneity (spatial and compositional). Semiempirical methods have the potential to reduce the computational cost compared to the computationally costly DFT computations. In this study, we have first assessed the performance of four semiempirical methods (GFN-xtb, GFN2-xtb, PM6-D3, and PM7-D3) to model major reactions for SEI formation and growth. We have included the decomposition reactions of the most used solvent (ethylene carbonate), most used salt (lithium hexafluorophosphate), and other electrolyte species like the co-solvent 1,3-dioxolane and the additive vinylene carbonate. We have found that PM7-D3 and GFN-xtb are the two best performing methods for the 32 tested reactions. Finally, we have performed PM7-D3 and GFN-xtb-based molecular dynamics for inorganic/organic interfaces. We have found that LiF is the most rigid salt, which barely reconstructs. In contrast, Li2O is subject to severe reconstruction at the GFN-xtb level of theory, but significantly less when using PM7-D3. Still, even at the PM7-D3 level of theory Li2O readily reacts with alkyl carbonates, leading to CO2 dissociation and thus the formation of surface carbonates. When in contact with Li2O, ethylene carbonate can undergo partial dehydrogenation reactions and ring openings. This suggests that Li2O is overly reactive to be in direct contact with such organic molecules. Rather, it is surrounded by a passivating (mono)layer of Li2CO3. Indeed, our simulations suggest that for such a hybrid system (core of Li2O, shell of Li2CO3, solvated with ethylene carbonate) the organic solvent remains intact. Furthermore, for such a hybrid system GFN-xtb produces physically meaningful results, so that this method can be overall recommended.