Computational study of molecular properties with dual basis sets

We have studied the performance of dual basis (DB) sets for the evaluation of molecular properties via second order Møller-Plesset perturbation theory (MP2). In addition to savings derived from using a trimmed basis set for the underlying Hartree-Fock (HF) calculation, we pursued a systematic truncation of the virtual subspace for further reductions in computational overhead during the post-HF step. Calculated total energies and molecular properties within the DB framework without virtual space truncation are generally in excellent agreement with full basis calculations. When aug-cc-pV5Z is used as the parent basis, mean absolute error for DB-HF (DB-MP2) total energies of molecules within our test set is 9.7 × 10(-5) au (8.0 × 10(-5) au) while mean absolute relative errors for static electrical response properties like dipole moments, isotropic dipole polarizabilities and polarizability anisotropies are 0.15% (0.14%), 0.56% (0.72%), and 0.76% (0.83%) respectively. When DB is coupled with virtual space truncation at the MP2 level, the corresponding errors are larger but still within 2% of full basis values.