Machine Learning-Guided Precise Electrolyte Engineering Enables Stable Lithium Metal Pouch Cells Exceeding 350 Wh/kg at −40 °C

Abstract

High-energy-density lithium metal batteries (LMBs) with wide-temperature adaptability are urgently needed for aerospace and military fields, yet extreme temperature conditions disrupt Li+ transport kinetics and cause failure in solid–electrolyte interphase (SEI) chemistry regulation. To guide molecular selection for wide-temperature electrolyte design, we constructed Electrolyte-65K, a rule-defined library of electrolyte-relevant molecules, and trained D-P Compass using batch-computable molecular descriptors to efficiently predict and rank dipole moment (D) and polarizability (P). Applying the trained models to a literature-derived set of electrolyte molecule candidates, D-P Pareto analysis identified N,N-dimethyltrifluoromethanesulfonamide (DMTMSA) as the candidate at the Pareto knee. The thermally sensitive dipole–dipole interactions between DMTMSA and weakly solvating solvent 1,2-diethoxyethane (DEE) enable a temperature-autoresponsive electrolyte that maintains an anion-dominated solvation structure across a wide temperature range, enabling SEI chemistry engineering regulation from −40 to 60 °C. A 5.8 Ah pouch cell achieves a high energy density of 503.7 Wh kg–1 at 25 °C with a lifespan of 275 cycles and retains 365.5 Wh kg–1 at −40 °C with 99.8% capacity retention over 50 cycles. This work establishes a data-driven molecular screening framework for wide-temperature electrolyte engineering design.