Vertical ionization potential benchmark for unitary coupled-cluster and algebraic-diagrammatic construction methods.

The performance of several methods for the calculation of vertical ionization potentials (IPs) or, more generally, electron-detachment energies based on unitary coupled-cluster (UCC) theory and the algebraic-diagrammatic construction (ADC) scheme is evaluated with respect to benchmark data computed at the level of equation-of-motion coupled-cluster theory, including single, double, and triple excitations (IP-EOM-CCSDT). Based on a statistical evaluation of about 200 electron-detached states of 41 molecules, the second-order methods IP-ADC(2) and IP-UCC2 show modest accuracies with IP-EOM-CCSDT as reference, exposing a mean signed error and a standard deviation of the error of -0.54 ± 0.50 and -0.49 ± 0.54 eV, respectively, accompanied by a mean absolute error (MAE) of 0.61 and 0.58 eV, respectively. The strict third-order IP-ADC method demonstrates an accuracy of 0.26 ± 0.35 eV (MAE = 0.35 eV), while the IP-UCC3 method is slightly more accurate with 0.24 ± 0.26 eV (MAE = 0.29 eV). Employing the static self-energy computed using the Dyson expansion method (DEM) improves the IP-ADC(3) performance to 0.27 ± 0.28 eV, with the mean absolute error of this method being 0.32 eV. However, employing the simpler improved fourth-order scheme Σ(4+) for the static self-energy provides almost identical results as the DEM. Based on the quality of the present benchmark results, it therefore appears not necessary to use the computationally more demanding DEM.