Machine Learning-Accelerated Kinetic Simulations of Surface Reactions with Complex Coverage Effects.

Area of Science:

• Computational Chemistry
• Surface Science
• Catalysis

Background:

• Coverage effects and adsorbate interactions are critical for catalytic performance but computationally challenging to simulate.
• Accurate modeling of surface reactions requires understanding how multiple adsorbates influence each other.

Purpose of the Study:

• To develop an efficient machine learning-accelerated kinetic Monte Carlo (ML-kMC) framework for modeling coverage-dependent surface reactions.
• To provide a generalizable computational tool for insights into catalytic mechanisms.

Main Methods:

• Developed an automated site-encoding scheme for surface configurations.
• Utilized structure descriptors and machine learning models (Gaussian Process Classifier, Bayesian Ridge Regression) for stability and energy predictions.
• Applied ML-kMC to simulate CO oxidation on Pd(111).

Main Results:

• Achieved high accuracy in stability (AUC = 98.93%) and energy predictions (MAE = 0.04 eV).
• ML-kMC simulations successfully reproduced experimental observations for CO oxidation, including oxygen redistribution and volcano-shaped activity trends.
• Identified strong influence of spatial distribution and local coverage of surface O species on reaction rates.

Conclusions:

• The ML-kMC framework efficiently incorporates coverage effects into catalytic simulations.
• Adsorbate-adsorbate interactions play a crucial role in surface reaction mechanisms.
• This approach offers valuable insights into the microscopic details of catalytic processes.