Beyond Fluorination: A Golden Criterion Guided by Chemical Coordination-Informed Machine Learning for High-Voltage Electrolyte Design.

Fluorine chemistry has garnered attention for extending operating voltage limits of electrolytes through robust interfacial passivation owing to fluorine's strong electronegativity. However, confronted with solvent/salt/additive multicomponent induced vast combinatorial space, conventional high-voltage electrolyte recipe design has been confined to reliance on fluorine content adjustments, resulting in inevitable trade-off between oxidation stability and ion transport kinetics. Herein, we develop a Chemical Coordination-Informed Molarity feature parsing approach embedded into machine learning for training adapted models. By building the one-to-one mapping between components and chemical-coordination atomic molarities of a given recipe, the trained gradient boosting regression achieves a prediction of oxidation potential with MAE below 0.36 V. Demonstrating 2808 experiment operational candidates based on a ternary-solvent blend, we reveal the pronounced role of mono-coordinated fluorine and double-bonded oxygen molarity ratio (F1/O1) for breaking the oxidative stability limit, and define a golden design criterion for guiding O1-involved recipes: F1(≥8.19)/O1(≥13.39) ∈ [0.55, 1.10]. Following this, we validate three experimentally reported low-fluoride recipes and identify two promising ones exhibiting oxidation potentials around 6.3 V vs. Li+/Li along with high ion-transport kinetics for further assessments. This work demonstrates customizable feature engineering in yielding intelligent materials design principles for reconciling multiple target performance that are usually mutually exclusive.