Adsorption on transition metal surfaces: Transferability and accuracy of DFT using the ADS41 dataset

Abstract

We present a benchmarking study of adsorption energies on transition metal surfaces computed with select functionals across different density functional theory codes. In addition to gradient corrected functionals, we evaluate the accuracies of representative metaGGAs including MS2, SCAN, and SCAN+rVV10 as well as a short-range screened hybrid functional, HSE06. The study shows that the challenge of finding a functional that can simultaneously capture both covalent and noncovalent molecule-surface interactions persists, with no single functional in the benchmarking study with average errors < 0.2 eV. We find that HSE06 on average does not improve the accuracy compared to PBE for the surface chemistry of transition metals. The BEEF-vdW dispersion-corrected GGA and the MS2 metaGGA yield the lowest errors in both chemisorption and dispersion energies, demonstrating that moving up the Jacob's ladder of functionals to screened hybrids does not necessarily improve the description of transition metal surface chemistry.I.

MCML: Combining physical constraints with experimental data for a multi‐purpose meta‐generalized gradient approximation

“…Much larger gradients occurring in the vicinity of nuclei do not explicitly appear in our simulations, as these are frozen in the employed pseudopotentials (see Section 2.2), which are generated from PBE all‐electron atomic orbitals. The usage of such PBE pseudopotentials is common practice for pseudopotential meta‐GGA calculations due to the (current) lack of meta‐GGA pseudopotentials 29,32 …”

“…This wide variety ensures a range of applicability for material property predictions of the optimized meta‐GGA. The systems contained in each data set with MCML predictions and comparisons to MS2, SCAN, mBEEF, and PBE can be found in the Supporting Information, Appendices S1 (overview of reactions) and S6 (DFT predictions). Surface‐phase data ADS41 (reactions listed in Table S5): The ADS41 data set 32 originates from high‐precision experimental data (equilibrium adsorption studies, temperature programmed desorption, and single crystal adsorption calorimetry), which were zero‐point energy corrected 46 . We split this data set into chemisorption‐dominated systems (called CHEMI26) and dispersion‐dominated systems (called DISP15) so that these properties can be separately evaluated and optimized.…”

“…Surface‐phase data ADS41 (reactions listed in Table S5): The ADS41 data set 32 originates from high‐precision experimental data (equilibrium adsorption studies, temperature programmed desorption, and single crystal adsorption calorimetry), which were zero‐point energy corrected 46 . We split this data set into chemisorption‐dominated systems (called CHEMI26) and dispersion‐dominated systems (called DISP15) so that these properties can be separately evaluated and optimized.…”

“…For the development of one of the most recent meta‐GGAs, SCAN, 24 all 17 known exact constraints and norms of a semi‐local functional for model systems with analytical solutions and limiting cases 28 were used, leading to improved thermodynamic stability predictions 29 and to excellent bulk structures and elastic properties for several material classes 30,31 . Chemisorption energetics on transition metal surfaces are, however, predicted less accurately than with several other GGA and meta‐GGA approaches, with SCAN generally predicting too strong chemisorption 27,32,33 …”

Dioxygen Binding to all 3d, 4d, and 5d Transition Metals from Coupled‐Cluster Theory

“…Experimental errors in both molecular and surface M−O bond energies are perhaps ∼20 kJ/mol . In comparison, typical oxygen‐bonding errors from density functional theory (DFT) are 50 kJ/mol both for surfaces and molecular systems . Best performances do not exceed 30 kJ/mol, even for the cleanest and simplest systems where chemical composition and thermodynamic effects are fully controlled .…”