Predicting Solubility in Nonaqueous Redox Flow Batteries via Quantum Mechanical Calculations and Machine Learning Approaches.

With increasing concerns over the climate crisis, there is a growing emphasis on sustainable and renewable energy sources. Solar and wind energy have proven competitive with fossil fuels; however, their intermittency remains a significant challenge. Nonaqueous redox flow batteries (NRFBs) offer a promising alternative due to their decoupled power and energy ratings, which provide enhanced flexibility, thermal stability, and safety. A key limitation of NRFBs is their low solubilities, which negatively impact energy density, stability, and cyclability. In this study, we investigate a highly stable bio-inspired octa-coordinated dianionic vanadium complex, vanadium^IV bis-hydoxyiminodiacetate (VBH^2-), which has shown promising solubility when paired with alkylammonium cations. Using quantum chemical methods, we predicted the solubilities of 91 additional alkylammonium cations in 19 different solvents for both [VBH]²⁻ and [VBH]⁻ active materials. Machine learning algorithms were then applied to identify the key molecular features driving solubility. Our results indicate that solubility does not increase linearly with alkyl chain length. For instance, [N_44t3t3]₂[VBH] in methanol exhibited a solubility 4.3 times higher than [N₄₄₄₄]₂[VBH] in acetonitrile. Additionally, cation molecular weight emerged as the most important predictor of solubility. This approach provides a framework for accelerating the development of high-solubility active materials efficiently and cost-effectively.