The relative efficacy of using low order trucations with large reference space vs high order methods with small reference space is tested by comparing quasidegenerate many-body perturbation theory (QDMBPT) calculations of potential curves for the five lowest electronic states of LiH with the multireference coupled cluster calculations of Ben-Shlomo and Kaldor [J. Chem. Phys. 89, 956 (1988)]. The infinite order coupled cluster calculations use two configurational reference spaces involving the 2σ, 3σ, and 1π orbitals, while the QDMBPT computations are truncated at either second or third orders and employ the full active reference space formed either from the 2σ, 3σ, and 1π or from the 2σ, 3σ, 4σ, and 1π orbitals. This gives us the opportunity of testing the dependence of QDMBPT computations on the size of reference space, the available freedom in choosing valence orbitals and orbital energies, and the order of truncation. Second order, four valence orbital space QDMBPT calculations provide good agreement with the repulsive portion of the coupled cluster potentials, but yield a separated atom limit that is too high and that therefore distorts the remainder of the potential. Third order improves the separated atom limit considerably, providing good agreement with the coupled cluster calculations. The ‘‘full chemical’’ five orbital reference space, on the other hand, yields very good agreement with coupled cluster potentials when using only the simpler second order QDMBPT calculations, and third order corrections in this case are very small but generally improve agreement with coupled cluster potentials. The five orbital reference space calculations are quite insensitive to a wide range of different choices of valence orbitals and orbital energies, demonstrating a robustness to the QDMBPT formalism used.
Read full abstract