A stable solid electrolyte interphase (SEI), formed by the reductive decomposition of electrolytes at the anode surface, is crucial for ensuring the safety and long-term performance of lithium metal and post-lithium metal batteries. Conventionally, the lowest unoccupied molecular orbital (LUMO) energy of an ion-solvent complex has been used as the primary descriptor for predicting reductive decomposition kinetics and electrolyte stability. In this work, we critically evaluate this assumption and show that LUMO energies do not exhibit a direct linear correlation with reductive decomposition kinetics. Instead, the relationship between LUMO levels and reductive stability is inherently nonlinear across diverse electrolyte chemistries, limiting the general applicability of LUMO-based screening. To address this limitation, we developed machine learning (ML) models that use multiple structural and electronic structure parameters as input features and free energy barriers as the target quantity. The models are trained on free energy barriers obtained from density functional theory (DFT) calculations for 200 ion-solvent complexes spanning a wide range of lithium metal battery (LMB) and post-LMB electrolytes. These nonlinear models were found to significantly outperform traditional linear approaches based solely on the LUMO energy, yielding more accurate predictions of the reductive stability of electrolytes. Our findings highlight the need for multifeature, nonlinear models to capture the complexity of electrolyte reactivity and offer a computational framework to accelerate the rational design of stable electrolytes for next-generation battery technologies.
Rathnakumaran et al. (Mon,) studied this question.