Transferable Machine Learning Interatomic Potential for Pd-Catalyzed Cross-Coupling Reactions

Finding efficient substrate-catalyst combinations for palladium-catalyzed cross-coupling reactions remains a critical challenge in synthetic chemistry, with broad implications for pharmaceutical and materials manufacturing. We report AIMNet2-Pd, a machine learned interatomic potential that enables rapid, accurate computational studies of palladium-catalyzed cross-coupling reactions. AIMNet2-Pd replaces computationally expensive electronic structure calculations with a neural network-based model that performs geometry optimization, transition state searches, and energy calculations in seconds while maintaining accuracy within 1-2 kcal mol⁻¹ and ~0.1 Å compared to the reference QM calculations. AIMNet2-Pd makes computational high-throughput catalyst screening and mechanistic studies of realistic systems feasible by providing on-demand thermodynamic and kinetic predictions for each step of a catalytic cycle. Importantly, the applicability of the systems extends beyond the monophosphine ligands in Pd(0)/Pd(II) cycles for which it has been trained on to chemically diverse Pd complexes. This demonstrates AIMNet2-Pd's utility to serve as a general-purpose and high-throughput tool for studying catalytic reactions.