Density Functional Theory Surrogate Enables Fast and Broad Computational Evaluation of Homogeneous Transition Metal Catalytic Energy Landscapes

The recently disclosed machine learning interatomic potential (MLIP), universal models for atoms (UMA), due to extensive training, has the potential to act as an extremely fast density functional theory (DFT) surrogate to evaluate the energies and structures of transition metal catalyzed reactions. Here, we report the evaluation and general success of UMA to calculate intermediates and transition structures and energies for catalytic cycles with a variety of metal centers, oxidation states, and ligand frameworks. Examples of catalytic cycles evaluated include alkane dehydrogenation, hydroformylation, olefin metathesis, cross-coupling, and asymmetric reactions. Overall, this evaluation indicates that UMA can be a reasonable surrogate for DFT and UMA energies and structures mimic what can be obtained with the ωB97M-V density functional. However, similar to using the ωB97M-V functional, caution should be taken when applying this MLIP to open-shell and multireference systems. Regardless, it is very likely that UMA and related MLIPs will rapidly be integrated into the computational assessment of homogeneous catalytic cycles.