Machine learning accelerated DFT research on platinum-modified amorphous alloy surface catalysts

Abstract

Pt-modified amorphous alloy (Pt@PdNiCuP) catalyst exhibits excellent electro-catalytic activity and high experimental durability for hydrogen evolution reaction (HER). However, the physical origin of the catalytically active remains unclear. In this paper, we constructed a distance contribution descriptor (DCD) for the feature engineering of machine learning (ML) potential, and calculated the Gibbs free energies (ΔGH) of 46,000 *H binding sites on the Pt@PdNiCuP surface by ML-accelerated density functional theory (DFT). The relationship between ΔGH and DCD revealed that in the H-Pt distance region of 2.0–2.5 Å where the parabolic tail and disordered scatters coexist, the H-metal bonding configuration is mainly the bridge- or hollow- bonding type. The contribution analysis of DCD indicates that the joint effect of Pt, Pd and Ni atoms determines the catalytical behavior of amorphous alloy, which agrees well with experimental results. By counting atomic percentages in different energy intervals, we obtained the atomic ratio for the best catalytic performance (Pt:Pd:Ni:Cu:P = 0.33:0.17:0.155:0.16:0.185). Projected density of states (PDOS) show that H 1s orbital, Pt 5d orbital, and Pd 4d orbital form a bonding state at −2 eV. These results provide new ideas for designing more active amorphous alloy catalysts.