Machine learning interatomic potentials for transition metal complexes

Machine learning interatomic potentials (MLIPs) have emerged as powerful tools for accelerating atomistic simulations. We have developed data-driven quantum chemistry (DDQC) models that integrate high-level electronic structure information directly into machine learning frameworks for transition metal complexes, a field that traditional, general purpose MLIPs cannot properly describe. By learning from quantum chemical quantities, the DDQC approach enables the construction of interatomic potentials that accurately reproduce energies, forces, and equilibrium geometries across diverse coordination environments. By embedding physically meaningful electronic information into the learning process, DDQC models provide accurate and transferable MLIPs capable of predicting molecular properties and potential energy surfaces with high fidelity. These models further enable efficient simulations of molecular dynamics and structural transformations in transition metal systems, opening new opportunities to study catalytic mechanisms and reactive dynamics with near–ab initio accuracy.