Benchmarking the accuracy of seniority-zero wavefunction methods for non-covalent interactions

TL;DR

This study evaluates seniority-zero wavefunction methods, particularly pCCD with dynamic corrections, for modeling various non-covalent interactions, demonstrating their efficiency and reliability compared to traditional methods.

Contribution

It benchmarks pCCD-based methods against established data sets, showing their potential as computationally efficient alternatives for non-covalent interaction modeling.

Findings

pCCD is dispersion-free and effective for non-covalent interactions.

Post-pCCD methods offer accurate, efficient modeling with small errors.

Linearized coupled cluster correction on pCCD is most reliable across systems.

Abstract

In this paper, we scrutinize the ability of seniority-zero wavefunction-based methods to model different types of non-covalent interactions, such as hydrogen bonds, dispersion, and mixed non-covalent interactions as well as prototypical model systems with various contributions of dynamic and static electron correlation effects. Specifically, we focus on the pair Coupled-Cluster Doubles (pCCD) ansatz combined with two different flavours of dynamic energy corrections, (i) based on a perturbation theory correction and (ii) on a linearized coupled-cluster ansatz on top of pCCD. We benchmark these approaches against the A24 data set [\v{R}ez{\'{a}}{\v{c}} and Hobza, J.~Chem.~Theory~Comput., 9, 2151-2155 (2013)] extrapolated to the basis set limit and some model non-covalent complexes that feature covalent bond breaking. By dissecting different types of interactions in the A24 data set within the Symmetry-Adapted Perturbation Theory (SAPT) framework, we demonstrate that pCCD can be classified as a dispersion-free method. Furthermore, we found that both flavours of post-pCCD approaches represent encouraging and computationally more efficient alternatives to standard electronic structure methods to model weakly-bound systems, resulting in small statistical errors. Finally, a linearized coupled cluster correction on top of pCCD proved to be most reliable for the majority of investigated systems, featuring smaller non-parallelity errors compared to perturbation-theory-based approaches.