Abstract
Grid-scale energy storage technologies are critical for achieving NetZero by 2050, which is a roadmap that requires a complete shift away from energy generation using fossil fuels and towards the use of renewable energy sources that are inherently intermittent. Non-aqueous redox flow batteries (NRFBs) are a promising candidate that offers scalability suitable for grid installations. However, current lead-candidate redox-active materials (RAMs) for NRFBs have limitations such as low solubility, impaired transport properties, and low reduction potentials. To address this issue, this work presents a comprehensive computational framework for in silico screening of these properties to rapidly optimize lead RAM candidates using quantum mechanics (QM), molecular dynamics (MD), and machine learning (ML) methods. This framework can be applied to any lead RAM candidate. In this study, the bioinspired vanadiumᴵⱽ bis-hydroxyiminodiacetate, [VBH]²−, was used as the model system. Results showed that accurate solubility prediction could be achieved by utilizing both lattice and solvation free energies, revealing that solvation free energy alone is insufficient. In addition, the MD simulations for transport properties showed congruence with the experiment, particularly for the viscosity, which offers a platform for in silico tuning by modifying the solution composition. Moreover, it was demonstrated that QM molecular design could finely adjust the reduction potential. Furthermore, significant design features were identified using descriptor-based ML algorithms, enabling rapid solubility prediction and providing guidance for improving solubility. The presented computational workflow, complemented with experimental validation, offers an effective strategy for enhancing the performance of lead RAM candidates in NRFBs. This method shows promise in accelerating the development of economical and efficient grid-scale energy storage solutions, essential for the transition toward a fully sustainable energy future.