Accuracy of Coupled Cluster Excited State Potential Energy Surfaces

The validation of the quality of the description of excited electronic states is of special importance in quantum chemistry as the general reliability of ab initio methods shows a much larger variation for these states than for the ground state. In this study, we investigate the quality of excited state energy gradients and potential energy surfaces on selected systems, as provided by the single reference coupled cluster variants CC2, CCSD, CCSD(T)(a)*, and CC3. Gradients and surface plots that follow the Franck-Condon forces are compared to the respective CCSDT reference values, thereby establishing a useful strategy for judging each variant's accuracy. The results reveal serious flaws of lower order methods - in particular, CC2 - in several situations where they otherwise give accurate vertical excitation energies, as well as excellent accuracy and consistency of the recently proposed CCSD(T)(a)* method.