Machine learning-driven shortening the screening process towards high-performance nitrogen reduction reaction...

Area of Science:

• Materials Science
• Catalysis Science
• Computational Chemistry

Background:

• The growing demand for clean energy necessitates environmentally friendly and efficient ammonia production methods.
• Electrocatalytic nitrogen reduction reaction (NRR) using single-atom catalysts (SACs) shows promise for industrial ammonia synthesis due to high efficiency and selectivity.
• A significant challenge lies in rapidly screening SACs for optimal catalytic performance.

Purpose of the Study:

• To computationally screen 29 transition metal-doped C6N2 nanosheet single-atom catalysts (TM@C6N2) for their potential in electrocatalytic nitrogen reduction reaction (NRR).
• To develop a machine learning-guided strategy for efficient prediction of SAC catalytic activity and selectivity.
• To identify key intrinsic properties influencing NRR performance and establish a streamlined catalyst screening process.

Main Methods:

• First-principles calculations were employed to analyze the stability, adsorption, catalytic activity, and electronic properties of 29 TM@C6N2 SACs.
• Machine learning models were developed to predict reaction energetics using intrinsic material features, identifying the first ionization energy (IE1) as a key descriptor.
• A novel four-step screening strategy integrating machine learning and Integrated Crystal Orbital Hamilton Populations (ICOHP) was proposed and validated against traditional methods.

Main Results:

• Mo@C6N2 and Re@C6N2 demonstrated superior NRR catalytic performance with low limiting potentials of -0.29 V and -0.31 V, respectively.
• Machine learning analysis identified the first ionization energy (IE1) as the most critical descriptor for predicting catalytic activity.
• The proposed four-step screening strategy achieved high efficiency and yielded results consistent with traditional five-step approaches.

Conclusions:

• Mo@C6N2 and Re@C6N2 are highly promising SACs for efficient electrocatalytic ammonia production.
• Machine learning-guided descriptor identification, particularly IE1, significantly enhances the prediction of SAC catalytic performance.
• The developed streamlined screening strategy offers a faster and equally effective alternative for discovering high-performance SACs for NRR.