Evaluating wavefunction methods, the counterpoise correction, and the frozen core approximation for the optimization...

A number of benchmarking studies have assessed the accuracy of various electronic structure methods for computing the interaction energies of van der Waals dimers, but fewer have systematically assessed the quality of dimer geometries obtained by these methods. We present optimized geometries of 21 van der Waals dimers using a highly accurate level of theory, namely coupled-cluster through perturbative triples at the complete basis set limit [CCSD(T)/CBS], and compare these results with optimizations performed at lower levels of theory. The lower levels of theory include variants of Møller-Plesset perturbation theory (MP2, MP2D, and MP2.5) and coupled-cluster theory [CCSD and CCSD(T)], with basis sets ranging from double- to quadruple-zeta. The accuracy of these methods is assessed by comparing errors in the least-root-mean-squared deviations (LRMSDs) of atomic coordinates, center-of-mass distances (ΔdCOM), interaction energies, and rotational constants. We also investigate the impact of the counterpoise correction and the frozen core approximation on the quality of the optimized geometries. Our findings show that increasing the basis set size beyond double-zeta significantly improves the accuracy of the geometries, while further improvements due to the basis set size depend on the method used. The frozen core approximation induces very small changes in geometries, while the counterpoise correction has a larger effect. For double-zeta basis sets, the counterpoise correction tends to degrade the quality of the optimized geometries, regardless of the method used. Several methods yield geometries with LRMSDs and ΔdCOM within 0.1 Å for all 21 dimers, and MP2D with the aug-cc-pVTZ basis set emerges as the most computationally efficient among these well-performing approaches with an average LRMSD and an absolute ΔdCOM of 0.02 Å.