The $S$-diagnostic -- an a posteriori error assessment for single-reference coupled-cluster methods

TL;DR

The paper introduces the $S$-diagnostic, a new a posteriori error assessment tool for single-reference coupled-cluster methods, demonstrating its superior performance over existing diagnostics in various chemical systems.

Contribution

It presents the derivation and numerical validation of the $S$-diagnostic, showing it outperforms traditional diagnostics and effectively detects multi-reference character in coupled-cluster calculations.

Findings

$S$-diagnostic outperforms $T_1$, $D_1$, and $D_2$ diagnostics.

$S$-diagnostic correlates well with error measures in geometry optimizations.

$S$-diagnostic correctly identifies multi-reference regimes and successful computations.

Abstract

We propose a novel a posteriori error assessment for the single-reference coupled-cluster (SRCC) method called the $S$-diagnostic. We provide a derivation of the $S$-diagnostic that is rooted in the mathematical analysis of different SRCC variants. We numerically scrutinized the $S$-diagnostic, testing its performance for (1) geometry optimizations, (2) electronic correlation simulations of systems with varying numerical difficulty, and (3) the square-planar copper complexes [CuCl$_4$]$^{2-}$, [Cu(NH$_3$)$_4$]$^{2+}$, and [Cu(H$_2$O)$_4$]$^{2+}$. Throughout the numerical investigations, the $S$-diagnostic is compared to other SRCC diagnostic procedures, that is, the $T_1$, $D_1$, and $D_2$ diagnostics as well as different indices of multi-determinantal and multi-reference character in coupled-cluster theory. Our numerical investigations show that the $S$-diagnostic outperforms the $T_1$, $D_1$, and $D_2$ diagnostics and is comparable to the indices of multi-determinantal and multi-reference character in coupled-cluster theory in their individual fields of applicability. The experiments investigating the performance of the $S$-diagnostic for geometry optimizations using SRCC reveal that the $S$-diagnostic correlates well with different error measures at a high level of statistical relevance. The experiments investigating the performance of the $S$-diagnostic for electronic correlation simulations show that the $S$-diagnostic correctly predicts strong multi-reference regimes. The $S$-diagnostic moreover correctly detects the successful SRCC computations for [CuCl$_4$]$^{2-}$, [Cu(NH$_3$)$_4$]$^{2+}$, and [Cu(H$_2$O)$_4$]$^{2+}$, which have been known to be misdiagnosed by $T_1$ and $D_1$ diagnostics in the past. This shows that the $S$-diagnostic is a promising candidate for an a posteriori diagnostic for SRCC calculations.