Identifying Adsorption States of OER Intermediates on Single-Atom Catalysts via a Spectral Machine Learning Framework

Identifying the adsorption states of intermediates in the oxygen evolution reaction (OER) is crucial for revealing the potential-determining step and further optimizing catalytic systems. Infrared (IR) spectroscopy serves as an effective tool for probing oxygen-containing intermediates on electrode surfaces. However, extracting spectral characteristics and establishing a quantitative correlation between these features and the adsorption states of intermediates remains a significant challenge. In this letter, we present a machine learning framework tailored for single-atom catalysts to learn from the infrared spectra of OER intermediates and construct a "spectrum-property" relationship. This enables accurate prediction of the adsorption states, namely adsorption free energy and charge of key intermediates (*OH, *O, and *OOH). Notably, the pretrained model demonstrates efficient transferability across commonly reported single-atom OER systems and provides interpretable attention maps of infrared signals based on vibrational mode analysis. By quantitatively linking spectral features to the adsorption states of oxygen-containing intermediates via machine learning, our framework is expected to provide valuable insights for guiding the optimization of single-atom OER catalysts.