Machine learning-guided deep eutectic electrolytes for high energy density and safe lithium–air batteries

Lithium–air batteries (LABs) are considered promising candidates for next-generation energy storage systems due to their ultrahigh energy density. However, their electrolytes still face high volatility, interfacial instability, and insufficient transport. Here, we design a deep eutectic electrolyte (DEE) composed of lithium bis(trifluoromethane sulfonimide), dimethylurea, succinonitrile (SN), and tetraethylene glycol dimethyl ether (G4), and optimize the compositional ratio using machine learning. This electrolyte significantly weakens ion–pair interactions through amide/nitrile with Li + multi-point coordination. Furthermore, the strong SN polarity enhances H-bonding, improving Li + migration efficiency, while G4 provides electron donor sites that coordinate with α-H in SN, reducing the reactivity between SN and lithium metal and enhancing interfacial stability. H-bonds and Li-bonds jointly construct an ordered ion transport network, reducing the free solvent fraction and inducing a dense and stable inorganic-rich SEI film formation at the interface. The assembled LABs exhibit a wide electrochemical stability window (5.12 V), achieve stable 274 cycles at 0.5 A g −1 test under 41% relative humidity, and have high safety. Furthermore, pouch cell demonstrates 671.5 Wh kg −1 energy density based on total cell weight. This study reveals the structure–performance relationship of DEE and provides alternative ideas for molecular design of high-safety, high-energy-density electrolytes. • Synergistic H-bonds and Li-bonds construct an efficient ion transport network. • Machine learning guides the optimization of the deep eutectic electrolyte. • An inorganic-rich SEI ensures interfacial stability and enhanced safety. • The pouch cell achieves 274 cycles and an energy density of 671.5 Wh kg −1 .