Machine-Learning-Accelerated DFT Conformal Sampling of Catalytic Processes.

Area of Science:

• Computational chemistry
• Materials science
• Catalysis

Background:

• Computational modeling of gas/solid catalytic interfaces is crucial for materials and process optimization.
• Current methods require enhancements in efficiency, accuracy, and throughput for broader practical impact.

Purpose of the Study:

• To develop an original approach, Conformal Sampling of Catalytic Processes (CSCP), for accelerating accurate and thorough sampling of novel catalytic systems.
• To leverage existing computational data for new systems, enhancing efficiency and predictive power.

Main Methods:

• Combining interpolation of Density Functional Theory (DFT) energetics using Machine-Learning Potentials.
• Employing conformal techniques for building training databases.
• Utilizing an active-learning strategy within the CSCP framework.

Main Results:

• CSCP achieved DFT-accuracy-level predictions for reaction energy diagrams after only two active-learning iterations.
• Successfully modeled methanol decomposition across seven diverse metal systems (Pt, Pd, Ni, Au, Ag, Cu, Co, Fe).
• Accurately reproduced changes in adsorption sites and reaction mechanisms, demonstrating robustness.

Conclusions:

• CSCP offers an operative tool to accelerate high-throughput sampling of catalytic processes.
• The approach effectively transfers knowledge from worked-out cases to novel systems.
• Enables efficient and accurate computational exploration of catalytic materials and reactions.