Site specific descriptors for oxygen evolution reaction activity on single atom catalysts using QMML

Abstract

Descriptors are properties or parameters of a material that are used to explain any catalytic activity both computationally and experimentally. Such descriptors aid in designing the material's properties to obtain an efficient catalyst. For transition metals, the d-band center is a well-known descriptor that shows a Sabatier type relationship for several catalytic reactions. However, it fails to explain the activity when considering the same metal active site with a varying local environment. To address this, density functional theory was used for single atom catalysts (SACs) embedded on armchair and zigzag graphene nanoribbons (AGNR and ZGNR). By varying the anchoring nitrogen atoms' orientation and considering pristine and doped cases, 432 active sites were used to test the oxygen evolution reaction (OER) activity. It was observed that S and SO₂ dopants help in reducing the overpotential on Co-SAC (η = 0.28 V). Along with the d-band center, a total of 105 possible descriptors were individually tested and failed to correlate with the OER activity. Furthermore, PCA was employed to narrow down unique descriptors, and machine learning algorithms (MLR, RR, SVR, RFR, BRR, LASSO, KNN and XGR) were trained on the two obtained descriptors. Among the models, SVR and RFR models showed the highest performances with R² = 0.89 and 0.88 on test data. This work shows the necessity for a multi-descriptor approach to explain OER catalytic activity on SACs and the approach would help in identifying similar descriptors for other catalytic reactions as well.