(362a) Modeling Supported Sub-Nanometer Clusters Via Actively Trained Machine Learning Potentials and Global Optimization

Supported sub-nanometer metal clusters are active and selective for several industrial reactions.¹ However, unlike ordered catalysts, like zeolites and metal crystals, supported clusters cannot be precisely characterized by spatiotemporally averaged spectroscopic methods. Furthermore, supported clusters have large configurational spaces, which cannot be reliably sampled using expensive ab initio computational methods. Consequently, their structure and structure-property relationships remain poorly understood. Prior computational work using on-lattice models has been applied to discover stable supported metal clusters.² Here we extend the procedure to include off-lattice configurations. We develop actively trained machine learning potentials to discover stable clusters on oxide supports with global optimization methods. We show that off-lattice models can discover stable configurations which are not accessible to lattice models. Our work also suggests that some support reconstructions and cluster conformations can be detected via spectroscopy. Our framework can be easily generalized to any metal/support combination.

References

1. Liu, L.; Corma, A., Metal Catalysts for Heterogeneous Catalysis: From Single Atoms to Nanoclusters and Nanoparticles. Chemical Reviews 2018, 118 (10), 4981-5079.
2. Wang, Y.; Su, Y.-Q.; Hensen, E. J. M.; Vlachos, D. G., Finite-Temperature Structures of Supported Subnanometer Catalysts Inferred via Statistical Learning and Genetic Algorithm-Based Optimization. ACS Nano 2020, 14 (10), 13995-14007.