Implementation of an ellipsoidal-cavity field correction for computed molecular oscillator strengths in solution: A(nother) benchmark study

We recently compared oscillator strengths (OS) obtained from electronic structure calculations f_comp to OSs derived from experimental spectra f_exp multiplied by the refractive index n of the solution in which the spectra were measured. The choice of nf_exp instead of f_exp as reference accounts for the macroscopic flux of energy in a dielectric (the experimental solvent). Here, we apply a correction to f_comp values that accounts for the local electromagnetic field driving the absorption transition (which is generally different from the macroscopic field). We refer to these modified OSs as f_comp,S. The correction is obtained by assuming that each molecule occupies an ellipsoidal cavity, fitted to its van der Waals surface, surrounded by a continuum dielectric model representing the solvent. Sets ranging from 33 - 85 experimental transitions are used for the benchmarking. For LR-CCSD and EOM-CCSD, we find that f_comp,S generally gives a better agreement with experimental strengths than f_comp. For LR-CCSD in the length gauge, for instance, there is a 1 to 1 scaling of the (nf_exp, f_comp,S) pairs. Instead, the results for TD-DFT depend on the amount of HF exchange used in the functional: pure functionals typically also have a 1 to 1 scaling of the (nf_exp, f_comp,S) pairs, while for hybrid functionals f_comp,S overestimates nf_exp to a degree that appears proportional to the amount of HF exchange present in the functional.