High-throughput screening and an interpretable machine learning model of single-atom hydrogen evolution catalysts with an asymmetric coordination environment constructed from heteroatom-doped graphdiyne

Abstract

Exploring high-activity and low-cost electrocatalysts for the hydrogen evolution reaction is the key to developing new energy sources, but it faces major challenges. Herein, a series of single atom catalysts with an asymmetric coordination environment constructed from heteroatom-doped graphdiyne are constructed to study HER activity. Based on DFT calculations, 120 configurations are screened and it is discovered that the ΔG_*H of Cr@N₂B1-GDY, Mn@N1B2-GDY, Fe@N1S2-GDY, Co@N1B4-GDY and Ni@N2-GDY was close to zero. Among them, Cr@N2B1-GDY has the lowest Gibbs free energy change (−0.0046 eV), exhibiting superior HER performance to known SACs of carbon-based supports. The constructed asymmetric coordination environment can reduce adsorption and enhance HER activity by adjusting the electron spin polarization of the central metal and the electronic structure of the D-band so that the e_g orbital is conducive to filling the antibonding orbital. The interpretable machine learning model's feature importance analysis further proves that the spin magnetic moment and the D-band center are highly correlated with the prediction accuracy of ΔG_*H. The hydrogen affinity of metal atoms also highly influences HER activity. RF is considered to be the best predictive machine learning model with R² of 96%. In this study, interpretive machine learning and DFT reveal that asymmetric doping of heteroatoms affects the D-band structure and spin states of the TM to regulate HER activity. It provides a novel way to establish the theory of HER SACs.