Machine Learning Accelerated Screening Advanced Single-Atom Anchored MXenes Electrocatalyst for Nitrogen Fixation

Abstract

Electrochemical nitrogen reduction reaction (NRR) has garnered significant attention as an alternative to the energy-intensive Haber–Bosch process. However, the vast search space for electrocatalysts and unclear structure–activity relationships limit the rational design of electrocatalysts for NRR. Herein, we present a machine learning-driven catalyst screening process to achieve high-throughput screening of two-dimensional material (MXene) supported single-atom catalysts (MXene-SACs) targeting NRR performance. Utilizing a database from density functional theory calculations as input, we identified four top-performing catalysts from 3146 MXene-SACs and recognized an effective intrinsic descriptor to accelerate high-throughput screening without additional computations, which was further validated experimentally by 10 synthesized MXene-SACs. In-depth study of the descriptor revealed a NRR mechanism: electron transfer from the single atom to coordinating atoms, causing crystal field splitting into eg and t2g bands. The interaction between the empty eg band of the single atom and the N2H π* orbital, activated by electrons from the t2g occupied orbital, facilitates the N≡N bond, promoting a smooth reaction under weak N2 adsorption. Impressively, the screened Mo2CO2–Zr, with strong atomic interactions, achieved a superior Faradaic efficiency of 31.7%. This work can not only provide a deeper understanding of the catalytic processes but also offer a foundation for future catalyst design and synthesis.