Graph-Based High-Throughput Framework for Screening Selective Propane Dehydrogenation Catalysts

Abstract

High-throughput computational screening is a powerful approach in accelerating the identification of highly selective and stable catalysts. However, it is often hindered by lack of generalized descriptors and the complexity of handling numerous multidentate adsorption configurations. In this study, we propose a computational framework integrating graph theory and python-based databasing tools with robust catalytic descriptors to enable high-throughput screening of alloys for nonoxidative propane dehydrogenation. We derive mechanistic Brønsted–Evans-Polanyi (BEP) correlations for C─H and C─C bond breaking, highlighting the role of metastable binding configurations in transition states involving more than three surface atoms. Although activity and stability descriptors exhibit strong scaling, these descriptors are uncorrelated, enabling construction of a pareto-optimal line identifying alloys with the best balance between activity and selectivity. Known optimal catalysts, including PtZn, PdZn, PtSn, and PdIn, lie on this pareto-optimal line validating the framework. Furthermore, Ir and Rh, typically known for hydrogenolysis, can be engineered for high selectivity by site-isolating active ensembles with high promoter compositions. Experimental validation confirms that Ir₁Sn₁ remains highly stable and selective over 15 h. Overall, our approach highlights the power of generalized descriptors combined with high-throughput screening and experimental benchmarking to extract key mechanistic insights and computationally design novel catalysts.