SORCI for photochemical and thermal reaction paths: A benchmark study

Dynamic electron correlation is very important for the correct description of potential energy surfaces, especially surface with differential dynamic electron correlation. Moreover, an accurate description of both the ground and excited state potential energy surfaces is essential to understand thermal and photochemical reactions of molecular systems. In the present work we test the performance of the Spectroscopy ORiented Configuration Interaction (SORCI) method that was originally designed to account for the dynamic electron correlation in an efficient manner at fixed geometry. Here, we assess the performance of SORCI along several paths on the ground and excited state potential energy surfaces that are related to the thermal and photochemical isomerization of a reduced retinal chromophore model. The ground state mapping involves three paths. Two of them are minimum energy paths that lead through different transitions states from one cis/trans stereoisomer to the other. One transition state has a diradical (open-shell) character while the other one has a charge-transfer character. The third ground state path connects these two transition states via a conical intersection. In addition, we assess the performance of SORCI for three excited state paths that comprise a minimum energy path starting from each of the cis and from the trans chromophore. These two pathways arrive at different conical intersections that are part of one seam of conical intersections. The third excited state path presents a profile along this seam. Despite the high complexity and varying mixture of different electronic structures involved in this benchmark, SORCI produces energy differences and energy profiles in good agreement with the multireference configuration interaction plus quadruples correction (MRCISD+Q) reference along all six paths. We find that SORCI is capable of producing smooth energy profiles, which require tighter thresholds than the present default. A further tightening of these thresholds by one order of magnitude yields converged SORCI values as compared to the limit of the method obtained by setting the three cutoffs to zero. However, we note also some deficiencies relative to MRCISD+Q in the description of relative energy differences. When comparing energies of structures that have very different geometries on the same potential energy surface we find that SORCI underestimates the reference values. Here SORCI underestimates the corresponding MRCISD+Q reference values thus indicating a non-negligible effect of the inactive double excitations. In addition we observe some artifacts at conical intersections due to the Davidson correction for higher excitations. Ways to overcome these drawbacks are discussed.