Fast-Track to Catalyst Stability: Machine Learning Optimized Predictions for M1/M2-N₆-Gra Catalysts.

Area of Science:

• Materials Science
• Computational Chemistry
• Catalysis

Background:

• Graphene-based dual-atom catalysts (M1/M2-N₆-Gra) show promise but their stability is uncertain.
• Efficient methods are needed to identify thermodynamically stable M1/M2-N₆-Gra for practical applications.

Purpose of the Study:

• To develop a machine learning (ML) model for predicting the formation energies (E_f) of M1/M2-N₆-Gra.
• To screen a large number of potential M1/M2-N₆-Gra catalysts for thermodynamic stability.

Main Methods:

• Utilized 143 DFT-calculated samples to train a multilayer perceptron ML model.
• Employed feature engineering, data supplementation, and transfer learning to enhance model performance.
• Achieved a high prediction accuracy with a test set R² = 0.98.

Main Results:

• Successfully predicted the formation energies for 1134 possible M1/M2-N₆-Gra structures.
• Identified 604 thermodynamically stable M1/M2-N₆-Gra catalysts with E_f < 0 eV.
• Determined that coordination number significantly impacts catalyst stability, with different energy contributions at low and high coordination.

Conclusions:

• Machine learning provides an efficient and accurate approach for screening stable dual-atom catalysts.
• The study reveals fundamental principles governing the stability of M1/M2-N₆-Gra catalysts.
• This work paves the way for accelerated discovery of advanced catalytic materials.