Extrapolation to the Limit of a Complete Pair Natural Orbital Space in Local Coupled-Cluster Calculations.

Ahmet Altun,Frank Neese,Giovanni Bistoni

doi:10.1021/acs.jctc.0c00344

Ahmet Altun, Frank Neese + Show 1 more

Open Access

https://doi.org/10.1021/acs.jctc.0c00344

Copy DOI

Abstract

The domain-based local pair natural orbital (PNO) coupled-cluster DLPNO-CCSD(T) method allows one to perform single point energy calculations for systems with hundreds of atoms while retaining essentially the accuracy of its canonical counterpart, with errors that are typically smaller than 1 kcal/mol for relative energies. Crucial to the accuracy and efficiency of the method is a proper definition of the virtual space in which the coupled-cluster equations are solved, which is spanned by a highly compact set of pair natural orbitals (PNOs) that are specific for each electron pair. The dimension of the PNO space is controlled by the TCutPNO threshold: only PNOs with an occupation number greater than TCutPNO are included in the correlation space of a given electron pair, whilst the remaining PNOs are discarded. To keep the error of the method small, a conservative TCutPNO value is used in standard DLPNO-CCSD(T) calculations. This often leads to unnecessarily large PNO spaces, which limits the efficiency of the method. Herein, we introduce a new computational strategy to approach the complete PNO space limit (for a given basis set) that consists in extrapolating the results obtained with different TCutPNO values. The method is validated on the GMTKN55 set using canonical CCSD(T) data as the reference. Our results demonstrate that a simple two-point extrapolation scheme can be used to significantly increase the efficiency and accuracy of DLPNO-CCSD(T) calculations, thus extending the range of applicability of the technique.

Highlights

The coupled-cluster method with singles, doubles, and perturbatively included triples excitations, i.e., CCSD(T),[1] can be used to compute relative energies with errors compared to experimental reference data that often fall within the experimental uncertainty.[2]
To investigate the dependence of the DLPNO-CCSD(T) correlation energy on the dimension of the pair natural orbital (PNO) space, we initially considered the benzene dimer, in which the two monomers are arranged in a parallel shifted configuration
This scheme allows us to skip the DLPNO-CCSD part for TCutPNO = 10−X. ΔEwp (10−X) calculations, which accounts for ∼25% of the overall computational time, it introduces a non-negligible error in the TCutPNO = 10−X amplitudes that deteriorates the quality of the extrapolated results

Summary

INTRODUCTION

The coupled-cluster method with singles, doubles, and perturbatively included triples excitations, i.e., CCSD(T),[1] can be used to compute relative energies with errors compared to experimental reference data that often fall within the experimental uncertainty.[2]. This procedure was implemented in ORCA and tested on the S22 and WATER27 sets This scheme allows us to skip the DLPNO-CCSD part for TCutPNO = 10−X calculations, which accounts for ∼25% of the overall computational time, it introduces a non-negligible error in the TCutPNO = 10−X amplitudes that deteriorates the quality of the extrapolated results (see AutoExtr-S22 and AutoExtr-WATER27 sheets of the Supporting Information).

RESULTS AND DISCUSSION

CONCLUSIONS

■ REFERENCES