Key points are not available for this paper at this time.
High Resolution Image Download MS PowerPoint Slide Self-interaction error (SIE) in density functional theory (DFT) calculations can lead to inaccurate descriptions of catalytic processes. In this work, we investigate the effects of SIE on a catalytic NO activation cycle on a cluster model of the Cu-SSZ-13 zeolite, systematically identifying transition states and reaction pathways for both single- and multistep reactions. We assess SIE using the Perdew–Zunger self-interaction correction (PZSIC) method, implemented with the Fermi-Löwdin orbital SIC (FLOSIC) approach. We benchmark the performance of standard DFT functionals (LDA, PBE, and r 2 SCAN) and SI-corrected functionals (PZSIC-LDA and the locally scaled LSIC( z σ )-LDA) against CCSD(T) reference energies for both adsorption energies and reaction barriers. The standard functionals increase in accuracy in the order LDA < PBE < r 2 SCAN, but all three tend to overestimate adsorption energies and underestimate reaction barriers. While PZSIC-LDA improves the description of reaction energetics over LDA for several reaction steps, it introduces large errors for processes that involve a change in the Cu-oxidation state. These errors are linked to the spurious destabilization of a filled 3d 10 electronic shell relative to 3d 9 in PZSIC calculations as described recently by Maniar et al. (Proc. Nat. Acad. Sci. 122, e2418305122 (2025)). This causes 3d 9 -like electronic configurations for the Cu active site to be favored over reference 3d 10 configurations at several points in the NO reaction cycle, giving rise to self-consistent PZSIC densities that are qualitatively incorrect. By locally scaling the SIC, LSIC mitigates some of the PZSIC errors and gives more consistent results than PZSIC-LDA. We conclude that to obtain accurate descriptions of catalytic reactions with SIC-based methods, the spurious energetics of transition metal complexes in competing oxidation states must be eliminated.
Shukla et al. (Sun,) studied this question.