Mechanism-Guided Catalyst Discovery for Methane C-H Activation via Structure-Aware Multisource Transfer Learning

Area of Science:

• Catalysis
• Materials Science
• Computational Chemistry

Background:

• Catalyst discovery in complex systems is limited by scarce kinetic data and unreliable linear scaling relationships.
• Diverse local coordination environments in heterogeneous catalysts break traditional predictive models.

Purpose of the Study:

• To develop a mechanism-driven, data-efficient approach for catalyst discovery.
• To create a multisource transfer learning (MS-TL) framework for accurate prediction of catalytic activity.

Main Methods:

• Formulated a mechanism-driven approach using multisource transfer learning (MS-TL).
• Extracted thermodynamic descriptors for initial, transition, and final states from adsorption data.
• Applied the framework to methane C-H activation on complex alloys.

Main Results:

• Achieved universal predictions of reaction barriers across various alloy facets and compositions.
• Recovered the Sabatier principle beyond linear scaling, revealing a multidimensional volcano trend.
• Developed a composite descriptor linking theoretical calculations to experimental methane oxidation rates.

Conclusions:

• The MS-TL framework effectively leverages abundant adsorption data for catalyst design.
• Established a new paradigm for rational catalyst design under realistic conditions.
• Bridged microscopic theory and macroscopic experimental observations for catalyst performance.