Machine Learning-Guided Solvation Engineering of Chiral Viologens for Durable Neutral Aqueous Organic Flow Batteries

Conventional N-alkylated viologen electrolytes in neutral aqueous organic redox flow batteries (AORFBs) undergo irreversible nucleophilic S_N2 dealkylation degradation. Moreover, trial-and-error molecular design often fails to resolve the solubility-stability trade-off in high-concentration systems. Here we report a machine learning (ML) strategy using large language models (LLMs) trained on over 1300 AORFB studies to predict chiral viologens with ortho-dihydroxy motifs. This bonding network forms a dynamic, pH-adaptive "solvation armor" that stabilizes the viologen structure. The R-/S-enantiomers (2.75/2.76 M) exhibit 1.66 times higher solubility versus RS-racemate. Molecular simulations and in situ spectroscopy confirm that the dihydroxy groups protect reactive C─N bonds via a solvation structure (unrelated to chiral effect), enhancing stability to pH 11. The 1 M R²⁺/R^+• redox couple sets a new record by achieving 99.42% capacity retention over 3652 cycles. The 1 M R-based AORFB shows 100% retention over 533 cycles, outperforming quaternary ammonium- ([(NPr)₂ V]Cl₄, 94.92%) and sulfonate-modified viologen ((SPr)₂ V), 65.49%). Stable cycling across 0.1 ∼ 2.5 M demonstrates decoupling of degradation from concentration. This strategy is validated by 2.5 kg-scale synthesis and Ah-class stack testing (98.65% retention over 77 cycles), demonstrating industrial scalability. This work establishes a generalizable, ML-enabled platform for electrolyte development, bridging molecular design and practical AORFB deployment.