CatFlow: An Automated\nWorkflow for Training Machine\nLearning Potentials to Compute Free Energies in Dynamic Catalysis

Dynamic effects of catalysts play\na crucial role in catalytic reactions,\nnecessitating the incorporation of statistical sampling and understanding\nof the impact of dynamic structures in free energy calculations. However,\nthe complexity of catalytic systems poses challenges in effectively\nexploring the vast configurational space effectively. In this work,\nwe propose CatFlow, an automated workflow for training machine learning\npotentials (MLPs) to compute free energies of catalytic reactions.\nCatFlow combines constrained molecular dynamics (MD) simulation with\nconcurrent training of MLPs and sequential calculation of free energies\nwith well trained MLPs. By rapidly generating reliable MLPs, CatFlow\nfacilitates rigorous free energy calculations, enabling the determination\nof the reaction profiles in an end-to-end manner. We showcased the\ncapabilities of CatFlow by investigating the activation of O<sub>2</sub> catalyzed by Pt clusters and demonstrated the effects of phase transition\non the activities of the catalytic reaction. CatFlow offers an efficient\nand automated solution for studying the catalytic elementary reaction\nprocesses. It reduces the need for human intervention and provides\nresearchers with a powerful tool to investigate free energies of dynamic\ncatalysis.