Accelerating atomic catalyst discovery by theoretical calculations-machine learning strategy

Abstract

Atomic catalysts (AC) are emerging as a highly attractive research topic, especially in sustainable energy fields. Lack of a full picture of the hydrogen evolution reaction (HER) impedes the future development of potential electrocatalysts. In this work, the systematic investigation of the HER process in graphdyine (GDY) based AC is presented in terms of the adsorption energies, adsorption trend, electronic structures, reaction pathway, and active sites. This comprehensive work innovatively reveals GDY based AC for HER covering all the transition metals (TM) and lanthanide (Ln) metals, enabling the screening of potential catalysts. The density functional theory (DFT) calculations carefully explore the HER performance beyond the comparison of sole H adsorption. Therefore, the screened catalysts candidates not only match with experimental results but also provide significant references for novel catalysts. Moreover, the machine learning (ML) technique bag-tree approach is innovatively utilized based on the fuzzy model for data separation and converse prediction of the HER performance, which indicates a similar result to the theoretical calculations. From two independent theoretical perspectives (DFT and ML), this work proposes pivotal guidelines for experimental catalyst design and synthesis. The proposed advanced research strategy shows great potential as a general approach in other energy-related areas.