Unveiling high-performance single-atom alloy catalysts for alkane dehydrogenation through DFT and machine learning synergy

Abstract

The recent surge in shale gas production has renewed interest in efficient alkane dehydrogenation for the synthesis of fuels and high-value chemicals. However, developing cost-effective catalysts that exhibit high catalytic activity while minimizing over-dehydrogenation and carbon deposition remains a significant challenge. Here, we integrate density functional theory (DFT) calculations with machine learning (ML) to design single-atom alloy (SAA) catalysts for efficient alkane dehydrogenation. Using DFT, we calculate 92 CH bond disassociation energy barriers to construct a dataset, which is used to train eight ML algorithms with 12 features. The top-performing Bagging Regression (BAR) model is then employed to predict CH bond activation energy barriers on the surfaces of 53 SAA candidates, enabling rapid screening of methane dehydrogenation activity. Among these, the Ru₁Cu SAA catalyst exhibits outstanding activity, outperforming pure Pt. Detailed DFT calculations confirm that Ru₁Cu(111) not only exhibits superior performance in methane dehydrogenation, but also exceptional activity in the dehydrogenation of propane, ethane, and isobutane. Moreover, microkinetic simulations further confirm the high selectivity of the Ru₁Cu(111) surface toward propylene during propane dehydrogenation. Feature engineering analyses reveal the critical roles of dehydrogenation steps and the surface energy of the single-atom metal in influencing CH bond activation. These findings underscore the effectiveness of the DFT–ML framework for catalyst discovery and highlight Ru₁Cu SAA as a highly active, selective, and stable catalyst with strong resistance to over-dehydrogenation and carbon deposition, making it a highly promising candidate for alkane dehydrogenation.