Stability and Dynamics of Subnanometer Clusters in a CO Atmosphere Via Machine Learning-Assisted Multiscale Modeling

Recently, single-atom catalysts (SACs) are being explored as effective catalysts as they offer high noble metal utilization and the potential to achieve superior activity and selectivity. However, the stability and dynamics of the catalysts, including single atoms and subnanometer clusters of a few atoms, remains elusive under reaction conditions due to experimental challenges. The cost to describe numerous cluster configurations and their reconstructions makes direct first-principles calculations impractical. Adsorbates, such as CO, further modify the catalyst structure. We have constructed a first-principles-based thermodynamic model to compute equilibrium structures and applied kinetic Monte Carlo (kMC) simulation to reveal the cluster dynamics. Machine learning surrogates trained on Density Functional Theory (DFT) data are used to parameterize the underlying Hamiltonian. We choose Pd clusters supported on CeO₂(111) under a CO pressure as a case study. The computational framework developed in this work allows the description of catalyst structures at the subnanometer scale under reaction conditions.