Interpreting Functional-Sensitivity in Catalysis: Exchange-Tuning Vs. DFT+U

Density functional theory is the preferred tool for the mechanistic study of heterogeneous catalysis. First-principles simulations provide valuable insight into barriers and structures associated with short-lived intermediates and transition states. However, DFT predictions of binding energetics and activation energies are highly sensitive to the form of approximate exchange-correlation functional employed. We present recent work in examining relationships between chemical composition of a catalyst (i.e., metal, ligand environment) with corresponding sensitivity of key descriptors of reactivity (e.g., small molecule binding energetics and structure) with respect to xc functional variation. These studies have enabled our development of a descriptor-based approach to identifying the reliability of a DFT prediction obtained, e.g. at reduced cost with a generalized gradient approximation (GGA) versus one obtained with an admixture of Hartree-Fock exchange. It has also allowed us to idenify how to optimally employ the affordable DFT+U approach to correct for underlying GGA shortcomings. We demonstrate these ideas on a number of paradigmatic systems including CO binding at transition metal catalysts for CO₂ reduction and in the properties of spin-crossover solid surfaces and complexes.