Designing optimal materials for redox flow batteries is challenging due to the multi-criteria requirement and large number of potential candidates. Multi-objective Bayesian optimization (MBO) was developed and demonstrated to significantly accelerate (15-fold) the identification of anolyte molecules of desired reduction potential, solubility, and fluorescence from a large chemical space.
Significance and Impact
Our MBO approach shows a 15-fold improvement in efficiency over random selection and identified 16 new anolytes after evaluating only 100 out of 1 million candidates.
- Density functional theory (DFT) was carried out on chemist-proposed 1400 data set of 2,1,3-benzothiadiazole (BzNSN) derivatives to compute the reduction potentials (Ered), solvation free energies (Gsolv), and absorption wavelengths (λabs) .
- MBO was first developed, tuned, and benchmarked on the DFT-computed data set, and then employed on new molecular dataset of 1 million BzNSN molecules to identify anolytes with desired design parameters.