Machine learning-assisted computational screening of high-entropy alloy catalysts for HCOOH decomposition

Abstract

Catalytic activities have been found to correlate with the adsorption energies of key intermediates in many catalytic reactions, and computational screening based on the adsorption energies of such intermediates has been widely used for catalyst discovery. High-entropy alloys (HEAs) offer an expansive configuration space, leading to a near-continuous distribution of adsorption energies for intermediates, thereby facilitating the identification of promising catalysts with optimal adsorption energies. However, comprehensive DFT calculations of adsorption energies on HEAs are hindered by the vast number of surface arrangements. Using the SISSO approach and DFT calculations, the adsorption energies of CO on the HEA AuCuIrPdPt(111) surfaces are predicted to identify improved catalysts for HCOOH decomposition, which provides a potential solution to hydrogen storage. We identify Au₈Cu₅IrPd₂₁Pt₁₃, Pd₂Au₂ and Pd₂Au/Pd as promising candidates, and DFT calculations and microkinetic modeling show that the systems present superior activity to conventional Pd catalysts by three orders of magnitude at 400 K while maintaining high H₂ selectivity. HCOOH decomposition proceeds through HCOO* and COOH* species on Au₈Cu₅IrPd₂₁Pt₁₃, Pd₂Au₂ and Pd₂Au/Pd, with the rate-determining steps being HCOO* and COOH* dehydrogenation, respectively. Various PdAu surface alloys with varied Pd/Au ratios also exhibit salient activities, which agrees well with the superior performance of PdAu catalysts widely observed in experimental studies. This work highlights the importance of the ensemble effect in HCOOH decomposition, and the combination of a data-driven approach, DFT calculations and microkinetic modeling provides a powerful tool for fast catalyst discovery.