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

The presence of multiple adsorbates and their lateral interactions significantly influence catalytic performance, yet accurately simulating these coverage effects remains computationally challenging. We present a machine learning-accelerated kinetic Monte Carlo (ML-kMC) framework to efficiently model coverage-dependent surface reactions. Our approach integrates three key functions: an automated site-encoding scheme for generating representative surface configurations, structure descriptors for structural characterization, and specialized machine learning models for stability and energy predictions. The Gaussian Process Classifier achieves outstanding stability prediction performance (AUC = 98.93%), and Bayesian Ridge Regression delivers accurate energy predictions with a mean absolute error of 0.04 eV. Taking CO oxidation on Pd(111) as a model system, our ML-kMC simulations successfully reproduce experimental observations including oxygen redistribution under CO coadsorption and the characteristic volcano-shaped activity trend. Further analysis reveals that the reaction rate is strongly influenced by the spatial distribution and local coverage of surface O species, highlighting the essential role of adsorbate-adsorbate interactions. This work provides an efficient and generalizable computational framework for incorporating coverage effects into catalytic simulations, offering valuable insights into the microscopic mechanisms of surface reactions.