First-principles calculations steered multi-task transformer model to screen dual-atom catalysts for C-H activation

Area of Science:

• Heterogeneous catalysis
• Materials science
• Computational chemistry

Background:

• Activating stable C-H bonds in light alkanes (methane, ethane, propane) is a significant challenge in catalysis.
• Dual-atom catalysts (DACs) present a promising approach for balancing catalytic activity and durability.

Purpose of the Study:

• To accelerate the design of active dual-atom catalysts (DACs) for light alkane conversion.
• To develop and optimize a machine learning model for predicting DAC performance.

Main Methods:

• Integration of first-principles calculations with a modified multi-task transformer machine learning model.
• Screening of over 200 dual-atom catalysts (DACs) comprising fourth/fifth period transition metals on N-doped graphene.
• Comparative analysis against conventional machine learning methods (KNN, RDF, GBRT).

Main Results:

• The optimized transformer model achieved R² > 0.85 in predicting adsorption energies and C-H activation barriers.
• The transformer model outperformed traditional methods on small datasets.
• Gradient boosting regression tree analysis identified metal-N coordination distance as a critical factor.

Conclusions:

• An accurate predictive model for efficient DAC screening in light alkane conversion was developed.
• The study advances the design principles for dual-atom catalysts.
• Machine learning accelerates the discovery of novel catalysts for challenging chemical transformations.