Assessing the Transferability of General-Purpose Machine Learning Interatomic Potentials for Heterogeneous Catalysis with HetCat26

HetCat26 is a collection of benchmark tests designed to evaluate pre-trained machine learning interatomic potentials for heterogeneous catalysis. The resource assesses model transferability across key aspects of catalytic modeling, where the applicability of general-purpose potentials remains underexplored. It is presented as a benchmark dataset and research study within chemistry and materials science.