Understanding DFT uncertainties for more reliable reactivity predictions by advancing the analysis of error sources

Decades of advancements and thousands of successful applications contributed to the reliability of density functional theory (DFT) methods. Especially in main group chemistry, DFT predictions tend to be increasingly more predictive. In this study we deeply analyze unexpected (ca. 8–13 kcal/mol) DFT disagreements obtained for a few organic reactions using only widely adopted, modern, hybrid and higher rung DFT methods. To understand the underlying causes, here, we move beyond conventional statistics-based benchmarks by combining recent advances in DFT error decomposition with affordable gold-standard references. This approach helps to characterize and disentangle multiple functional and density-based error types and enables us to find functional(s) suitable for broad mechanistic studies in all studied examples. The proposed tools are cost-efficient, readily accessible, and easy to integrate into routine thermochemistry workflows. While the focus is on main group reactions, the approach is applicable also for transition metal, bio-, and surface chemistry to assist more predictive reactivity modeling.