Why are GGAs so accurate for reaction kinetics on surfaces? Systematic comparison of hybrid vs. nonhybrid DFT for representative reactions.

Abstract

"Jacob's Ladder" of approximate exchange-correlation (XC) functionals in Kohn-Sham density functional theory are widely accepted to have systematic errors in reaction barriers. The first-rung local spin-density approximation (LDA) typically predicts barriers below generalized gradient approximations, which in turn predict barriers below experiment and below fourth-rung hybrid functionals incorporating a fraction of exact exchange. We show that several reactions from previous literature reports, as well as new simulations of carbon-carbon coupling in the Fischer-Tropsch process, do not follow this conventional picture. We introduce the AB9 test set of nine abnormal reaction barriers, in which density gradient corrections and exact exchange admixture tend to lower rather than to raise predicted barriers. Comparisons of normal and abnormal reactions rationalize this phenomenon in terms of how density gradient and exact-exchange corrections stabilize transition states relative to reaction intermediates. Multireference diagnostics confirm that this behavior is not merely a consequence of multireference character. Benchmarks of the AB9 set, using the best available ab initio reference values, highlight the role of symmetry breaking and show surprisingly good performance from both the LDA and "Rung 3.5" functionals. This motivates benchmarks of the AB9 set in future XC functional development.