A lower scaling four-component relativistic coupled cluster method based on natural spinors

Chamoli, Somesh; Surjuse, Kshitijkumar; Nayak, Malaya K.; Dutta, Achintya Kumar

doi:10.48550/arxiv.2201.07752

Cited by 1 publication

(1 citation statement)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We note that upon completing this manuscript, we have become aware of another implementation of the MP2FNOs approach for relativistic correlated methods, in the BAGH code 49 . While the main features of the MP2FNOs method are the same in both implementations we first note that our implementation fully exploits ExaTENSOR's single-node or distributed memory (multi-node) and GPU acceleration capabilities, and as such can be efficiently employed in systems ranging from local clusters to latest-generation supercomputers.…”

Section: Introductionmentioning

confidence: 99%

Assessing MP2 frozen natural orbitals in relativistic correlated electronic structure calculations

Yuan,

Visscher,

Gomes

2022

Preprint

View full text Add to dashboard Cite

The high computational scaling with the number of correlated electrons and the size of the basis set is a bottleneck which limits applications of coupled cluster (CC) algorithms. This is particularly so for calculations based on 4-component relativistic Hamiltonians, which generally employ uncontracted basis sets and lead to large virtual molecular orbital (VMO) spaces. This problem may be alleviated by employing a more compact set of virtual spinors than those provided by the canonical Hartree-Fock (HF) set, such as those based on natural orbitals (NOs). In this paper we describe the implementation of a module for generating NOs for correlated wavefunctions, and in particular MP2 frozen natural orbitals (MP2FNOs), as a component of our novel implementation of relativistic coupled cluster theory for massively parallel architectures [J. Pototschnig et. al. J. Chem. Theory Comput. 17, 5509, 2021]. Our implementation is capable of manipulating both complex and quaternion density matrices, thus allowing for the generation of both Kramersrestricted and Kramers-unrestricted MP2FNOs. Furthermore, NOs are re-expressed in the parent atomic orbital (AO) basis, so that the code also makes it possible to generate CCSD natural orbitals in AO basis for further analysis. By investigating the truncation errors of MP2FNOs for both the correlation energy and molecular properties at CCSD level such as the electric field gradients at the nuclei (EFGs), electric dipole and quadrupole moments for hydrogen halides HX (X=F, Cl, Br, I, At, Ts), and parity-violating energy differences (PV) for H 2 Y 2 (Y=O, S, Se), we find that MP2FNOs accelerate the convergence of the correlation energy in a roughly uniform manner across the periodic table and that, with VMO spaces truncated to around half the size of the full spaces ones, it is possible to obtain reliable estimates for both energies and all molecular properties considered.Recently, Pototschnig et al. 36 described a new, efficient relativistic coupled cluster implementation based on ExaTEN-SOR 37 , a distributed numerical tensor algebra library for GPU-accelerated HPC platforms. This code enabled the calculation of molecular properties with CCSD wave functions for systems such as [(UO 2 )(NO 3 ) 3 ] − , for which 200 elec-

show abstract

Section: Introductionmentioning

confidence: 99%