Extended Active Learning-Guided Design of Multi-Component Electrolytes with Opposing Cosolvent Properties for Aqueous Zinc-Ion Batteries

Multi-component electrolytes offer expanded tunability over electrochemical properties, yet their rational design is constrained by a lack of generalizable principles. Intricate intercomponent interactions obscure the relationship between the molecular properties and battery performance, rendering material selection and compositional optimization intractable. Here, we present an extended active learning-guided framework to efficiently navigate the vast materials space and refine formulations for tri-solvent-based electrolytes in aqueous zinc-ion batteries with minimal experimental workload. Specifically, co-solvent candidates are clustered by molecular properties; co-solvents are then selected, and their pairwise combinations undergo rigorous formulation optimization. Our data-driven approach discovered a nonintuitive electrolyte composition: combining co-solvents with fundamentally opposing properties. This unique formulation, which disrupts the hydrogen-bonded network while reconstructing the solvation structure, enables Zn||NH4V4O10 full-cells to retain 90.2% capacity after 1,000 cycles. This study unveils the possibility of using active learning to discover unconventional electrolyte formulations that would otherwise be infeasible to identify through human insights.