Machine-learning enables nitrogen reduction reaction on transition metal doped C3B by controlling the charge states

Abstract

Transition metal (TM)-doped monolayer semiconductors have attracted significant attention as electrocatalysts for various applications. However, conventional density functional theory calculations often yield inaccurate predictions due to the omission of charge states, due to which extensive efforts to explore promising electrocatalysts are in vain. Here, we report a computational pipeline for high-throughput screening that combines charge-state-aware DFT calculations for stability and activity predictions with machine learning (ML)-enabled feature and mechanism analysis. Applying this pipeline to a TM-doped C₃B monolayer (TM@C₃B) to search for potential nitrogen reduction reaction (NRR) electrocatalysts, we initially identified 92 types of stable charge states of TM@C₃B under B-rich conditions. By considering both activity and selectivity, we identified V_C@C₃B (V-doped at the C site in either the 0 or +1 charge state) as a promising candidate, which exhibited both low limiting potentials and excellent selectivity for the NRR. Further ML analysis of the N₂ adsorption energy and the first and last hydrogenation steps of TM@C₃B revealed that charge transfer and the d-band center are critical factors governing NRR performance, both of which can be modulated by the different charge states. This study highlights the necessity of charge state calculations in electrochemical reaction modeling, paving a new pathway for the rational design of high-performance NRR electrocatalysts.