Fig. 3: Correct identification of the NO on MgO(001) adsorption configuration.

For NO on MgO(001), six adsorption configurations have been proposed: ‘dimer Mg’, ‘bent Mg’, ‘upright Mg’, ‘bent bridge’, ‘bent O’ and ‘upright hollow’. These names reflect the orientation and binding site on the surface. The adsorption enthalpy H_ads is calculated for each configuration with the autoSKZCAM framework and a set of six DFAs in DFT. The estimates from the six DFAs are plotted as grey-filled circles, with a light grey bar to highlight the range between the smallest and largest values. Experiments^10,11 indicate that the dimer Mg configuration is the most stable. The autoSKZCAM framework, plotted as blue lines with a corresponding error bar, is the only method that correctly identifies this configuration while reproducing the experimental H_ads measurement by Wichtendahl et al.⁴³. The experimental error on H_ads is based on temperature-programmed desorption (TPD) analysis by Campbell and Sellers²³, taken as twice the standard deviation in the predicted pre-exponential factor against a test set of ~20 adsorbed molecules (Supplementary Section 11). The experimental error range is indicated with a light pink bar, which is highlighted with a dashed pink line for the lowest energy ‘dimer Mg’ configuration. The autoSKZCAM simulation errors are the root squared sum of several systematic contributions described in Supplementary Section 9, with the majority arising from errors using a geometry optimized by DFT, which we estimate as twice the root mean square error from an ensemble of six DFAs. The DFAs used are PBE-D2[Ne], revPBE-D4, vdW-DF2, rev-vdW-DF2, PBE0-D4 and B3LYP-D2[Ne], with B3LYP-D2[Ne] and rev-vdW-DF2 explicitly indicated with an open circle and square, respectively, as these sit at either end of the DFT range. The autoSKZCAM and DFT estimates are tabulated in Supplementary Section 1c.