First-Principles and Machine Learning-Based Analysis of Sulfur Poisoning of High-Entropy Alloy Catalysts

Mitigation of sulfur poisoning of transition metal catalysts remains a significant challenge for many thermally driven heterogeneous catalytic reactions. Recently, high-entropy alloys (HEAs), a class of multimetallic alloys having a disordered structure, have emerged as highly stable catalysts for a diverse set of reactions. They typically consist of greater than four elements, allowing for greater tunability as compared to conventional bimetallic alloys. HEA catalysts are, therefore, promising candidates for the mitigation of sulfur poisoning. However, the combinatorial space of these multimetallic alloys is exceedingly complex, and therefore, comprehensive experimental investigation of all candidate HEAs, or computational design of HEAs exclusively using first-principles methods, is infeasible. To address these challenges, we introduce a framework combining density functional theory (DFT) calculations and a machine learning model (SlabGCN) that can be utilized for accelerated predictions of the surface stability of HEA catalysts. In this study, we specifically employ the framework to simultaneously predict sulfur adsorption energies and surface energies of binary, ternary, and quaternary alloys comprised of Pt, Pd, Rh, and Cu. Rapid screening of diverse surface structures and varying sulfur coverages using this framework permits construction of detailed surface phase diagrams at arbitrary compositions that elucidate evolution of sulfur poisoning as a function of reaction conditions. We explicitly study the influence of surface segregation on surface sulfidation and demonstrate that this influence decreases as the number of elements increases. Lastly, we illustrate an optimization workflow that can be used to discover more sulfur tolerant alloys, and we discuss potential applications of the broader computational framework in studying more complex chemistries on high-entropy alloys.