Data-driven decoupling of electronic and geometric descriptors for fluorinated additives in lithium batteries.

Lithium-coordinated LUMO energies of 1432 monofluorinated additives were predicted by interpretable machine learning (R2 = 0.89) to identify key structural descriptors. Decoupled electronic and geometric descriptors of polar accessibility and compact topology in cyclic carbonate scaffolds were revealed by Shapley analysis; one candidate exhibited an orbital energy of 1.76 eV, which is lower than that of conventional fluoroethylene carbonate.