Post-CCSD(T) corrections in the S66 noncovalent interactions benchmark

TL;DR

This paper investigates the accuracy of CCSD(T) for noncovalent interactions in the S66 benchmark, showing how higher-order corrections improve results and proposing a simple model to predict these corrections efficiently.

Contribution

It introduces a two-parameter model to predict CCSDT(Q)–CCSD(T) differences, enhancing the accuracy of noncovalent interaction calculations without high computational cost.

Findings

CCSD(T) benefits from error cancellation in hydrogen bonds and London complexes.

Breakdown of error cancellation in π-stacking complexes leads to overbinding.

A simple model predicts higher-order correction differences with 0.01 kcal/mol RMS.

Abstract

For noncovalent interactions, it is generally assumed that CCSD(T) is nearly the exact solution within the 1-particle basis set. For the S66 noncovalent interactions benchmark, we present for the majority of species CCSDT and CCSDT(Q) corrections with a polarized double-zeta basis set. For hydrogen bonds, pure London complexes, and mixed-influence complexes, CCSD(T) benefits from error cancellation between (usually repulsive) higher-order triples, $T_3 - (T)$, and (almost universally attractive) connected quadruples, (Q). For $\pi$-stacking complexes, this cancellation starts breaking down and CCSD(T) overbinds; CCSD(T)$_\Lambda$ corrects the problem at the expense of London complexes. A fairly simple two-parameter model predicts CCSDT(Q)--CCSD(T) differences to 0.01 kcal/mol RMS, requiring no calculations that scale more steeply than $O(N^7)$.