Multiparameter Machine Learning Quantifies Electronic Dominance in Pd-Catalyzed Formic Acid Dehydrogenation.

Area of Science:

• Catalysis
• Materials Science
• Computational Chemistry

Background:

• Geometric optimization shows limited efficiency for formic acid dehydrogenation (FAD).
• Advancing palladium-based catalysts necessitates a deeper understanding of electronic structural regulation.
• Current approaches often rely on single-factor analyses, limiting predictive power.

Purpose of the Study:

• To develop a machine learning framework for predicting catalytic activity in FAD.
• To establish a multiparameter correlation model integrating kinetic barriers and electronic descriptors.
• To elucidate the relative importance of electronic structure versus geometric factors in catalyst performance.

Main Methods:

• Development of a catalytic system using palladium confined within metal-nitrogen-doped carbon supports (Pd@MNC).
• Application of machine learning to build a multiparameter correlation model.
• Integration of intrinsic kinetic barriers (Eads) with diverse electronic descriptors.

Main Results:

• Electronic structure engineering (48% relative importance) is more critical than geometric tunability (12%) for catalytic kinetics.
• Key electronic descriptors include d-band center offset (εd, 30%) and Pd(II) proportion (ωPd(II), 18%).
• Experimental validation using Co and Cr doping confirmed the model's predictions.

Conclusions:

• The developed machine learning framework provides a predictive paradigm for rational catalyst design.
• Multidimensional electronic regulation significantly enhances FAD performance.
• The strategy shows broad applicability for other palladium-catalyzed reactions like Suzuki coupling and CO2 reduction.