(662c) Integrating Experimental and Theoretical Data for High Quality Predictions of Material Performance Towards Electrochemical Reactions

The discovery of inexpensive and abundant catalysts that have high activity, selectivity, and stability for oxygen reduction reaction (ORR) is crucial for broader, efficient use of fuel cell devices. However, exploring the vast chemical space of materials and testing them experimentally is a challenging and time-consuming task. Therefore, the authors propose a high-throughput approach to explore untested materials with potential for ORR reactions, by combining computational simulations and experimental data.

The authors focus on transition metal (M) antimonates (MSbO_x) and aim to extrapolate the approach to other classes of materials. The methodology utilizes Density Functional Theory (DFT) calculations to extract electronic and structural descriptors from bulk crystal structures of materials. These descriptors are then combined with experimental ORR catalyst data and machine learning (ML) techniques to efficiently identify the most relevant factors that predict catalytic activity under relevant conditions. The authors have identified both experimental and theoretical descriptors and mechanisms to determine patterns of activity towards ORR. Mathematically simple and human interpretable models (rather than a black-box type approach) built over the descriptors, have been generated and simplified. Resulting models are straightforward and interpretable, and they have practical applications for transfer learning in predicting other active materials based on derived mathematical correlations.

The authors’ approach can result in identification of new and promising catalysts with high ORR activity, and the methodology can be extended to include other materials such as sulfided (MSbS_x) and nitrided antimonates (MSbN_x) by incorporating universal electronic and structural descriptors. Similarities between extracted descriptors for M-O, M-N, and M-S would establish universality. The authors’ models have been benchmarked against standard ML methodologies and found to be accurate and transferable. The authors’ efforts also include an integration of these experimental and theoretical data via CatHub (https://www.catalysis-hub.org/) Python API, which provide valuable data for the discovery of novel electrocatalysts.