Revealing Electronic Structure–Chemisorption Relationships for Accelerated Discovery of Aqueous Zinc Battery Additives Through Machine Learning

Inferior rechargeability of the metallic zinc anode remains a critical bottleneck for the stable and efficient operation of aqueous Zn-ion batteries (AZIBs). This issue can be considerably addressed by using organic electrolyte additives, with the adsorption energy of additives on the Zn surface playing the dominant role for predicting additive efficacy, particularly in terms of dictating the Zn-anode cycle life. In this study, we present a machine learning (ML) framework for its rapid and accurate prediction, leveraging electronic features derived solely from Density Functional Theory (DFT) calculations of isolated additive molecules. Specifically, we use the statistical moments of the sp-band density of states and the frontier molecular orbital (FMO)-based Highest (Lowest) Occupied (Unoccupied) Molecular Orbitals (HOMO and LUMO) as primary features. For a dataset of 301 diverse organic additives, the Random Forest model (the best-performing model amongst seven ML models evaluated) achieved a root-mean-square error (RMSE) of 0.140 eV and a mean absolute error (MAE) of 0.113 eV on the test set. Our analysis highlights the significance of the molecule's sp-band shape, particularly the third-order moments, and the LUMO in governing adsorption trends. This work not only introduces an efficient framework for high-throughput screening of electrolyte additives for AZIBs, but also uncovers a fundamental relationship between the electronic structure of relatively large organic molecules and their chemical reactivity at metal surfaces. Notably, it draws conceptual parallels to the well-established d-band model, which has been critical in understanding the reactivity of transition metal surfaces.