Efficient evaluation of triple excitations in symmetry-adapted perturbation theory via second-order Moller-Plesset perturbation theory natural orbitals

An accurate description of dispersion interactions is required for reliable theoretical studies of many noncovalent complexes. This can be obtained with the wave function-based formulation of symmetry-adapted perturbation theory (SAPT) provided that the contribution of triple excitations to dispersion is included. Unfortunately, this triples dispersion correction limits the applicability of SAPT due to its O(N-7) scaling. The efficiency of the evaluation of this correction can be greatly improved by removing virtual orbitals from the computation. The error incurred from truncating the virtual space is reduced if second-order Moller-Plesset perturbation theory (MP2) natural orbitals are used in place of the canonical Hartree-Fock molecular orbitals that are typically used. This approximation is further improved if the triples correction to dispersion is scaled to account for the smaller virtual space. If virtual MP2 natural orbitals are removed according to their occupation numbers, in practice, roughly half of the virtual orbitals can be removed (with the aug-cc-pVDZ basis set) with negligible errors if the remaining triples dispersion contribution is scaled. This typically leads to speedups of 15-20 times for the cases considered here. By combining the truncated virtual space with the frozen core approximation, the triples correction can be evaluated approximately 50 times faster than the canonical computation. These approximations cause less than 1% error (or at most 0.02 kcal mo1(-1)) for the cases considered. Truncation of greater fractions of the virtual space is possible for larger basis sets (leading to speedups of over 40 times before additional speedups from the frozen core approximation). (c) 2010 American Institute of Physics. [doi:10.1063/1.3479400]