Third order M{\o}ller-Plesset theory made more useful? The role of density functional theory orbitals

TL;DR

This paper demonstrates that using density functional theory orbitals in Møller-Plesset perturbation theory significantly enhances its accuracy for various molecular properties, surpassing some coupled cluster methods.

Contribution

It introduces the use of DFT orbitals in MP2 and MP3 calculations, improving their performance and efficiency in electronic structure predictions.

Findings

DFT orbitals improve MP2 and MP3 accuracy across multiple properties.

MP3 with DFT orbitals can outperform CCSD in certain datasets.

Range-separated hybrid functionals yield the best results.

Abstract

The practical utility of M{\o}ller-Plesset (MP) perturbation theory is severely constrained by the use of Hartree-Fock (HF) orbitals. It has recently been shown that use of regularized orbital-optimized MP2 orbitals and scaling of MP3 energy could lead to a significant reduction in MP3 error (J. Phys. Chem. Lett. 10, 4170, 2019). In this work we examine whether density functional theory (DFT) optimized orbitals can be similarly employed to improve the performance of MP theory at both the MP2 and MP3 levels. We find that use of DFT orbitals leads to significantly improved performance for prediction of thermochemistry, barrier heights, non-covalent interactions, and dipole moments relative to standard HF based MP theory. Indeed MP3 (with or without scaling) with DFT orbitals is found to surpass the accuracy of coupled cluster singles and doubles (CCSD) for several datasets. We also found that the results are not particularly functional sensitive in most cases, (although range-separated hybrid functionals with low delocalization error perform the best). MP3 based on DFT orbitals thus appears to be an efficient, non-iterative $O(N^6)$ scaling wave function approach for single-reference electronic structure computations. Scaled MP2 with DFT orbitals is also found to be quite accurate in many cases, although modern double hybrid functionals are likely to be considerably more accurate.