Chemically Accurate 0-0 Energies with Not-so-Accurate Excited State Geometries.

Using a series of increasingly refined wave function methods able to tackle electronic excited states, namely ADC(2), CC2, CCSD, CCSDR(3), and CC3, we investigate the interplay between geometries and 0-0 energies. We show that, due to a strong and nearly systematic error cancelation between the vertical transition and geometrical reorganization energies, CC2 and CCSD structures can be used to obtain chemically accurate 0-0 energies, though the underlying geometries are rather far from the reference ones and would deliver significant errors for several chemical and physical properties. This indicates that obtaining 0-0 energies matching experiment does not demonstrate the quality of the underlying geometrical parameters. By computing CC3 total energies on CCSD structures, we model a large set of compounds (including radicals) and electronic transitions (including singlet-triplet excitations) and successfully reach chemical accuracy in a near systematic way. Indeed, for this particular set, we obtain a mean absolute error as small as 0.032 eV, chemical accuracy (error smaller than 1 kcal·mol or 0.043 eV) being obtained in 80% of the cases. In only three cases out of more than 100 examples, the error exceeds 0.15 eV which is of the order of the typical error provided by TD-DFT or second-order wave function methods for 0-0 energies. The present composite approach seems therefore effective, at least for low-lying states, despite the fact that the geometries may not be considered as very accurate.