Parallel, linear-scaling building-block and embedding method based on localized orbitals and orbital-specific basis sets

TL;DR

This paper introduces a parallel, linear-scaling method for energy minimization in quantum chemistry calculations, utilizing localized orbitals and orbital-specific basis sets to improve efficiency and scalability.

Contribution

It presents a novel parallel approach based on localized orbitals and embedded cluster equations, enabling efficient large-scale quantum chemical computations.

Findings

Achieved linear scaling in energy minimization for large molecules.

Demonstrated high parallel efficiency in solving embedded cluster equations.

Validated method on poly(ethylene oxide) and carbon monoxide clusters.

Abstract

We present a new linear scaling method for the energy minimization step of semiempirical and first-principles Hartree-Fock and Kohn-Sham calculations. It is based on the self-consistent calculation of the optimum localized orbitals of any localization method of choice and on the use of orbital-specific basis sets. The full set of localized orbitals of a large molecule is seen as an orbital mosaic where each tessera is made of only a few of them. The orbital tesserae are computed out of a set of embedded cluster pseudoeigenvalue coupled equations which are solved in a building-block self-consistent fashion. In each iteration, the embedded cluster equations are solved independently of each other and, as a result, the method is parallel at a high level of the calculation. In addition to full system calculations, the method enables to perform simpler, much less demanding embedded cluster calculations, where only a fraction of the localized molecular orbitals are variational while the rest are frozen, taking advantage of the transferability of the localized orbitals of a given localization method between similar molecules. Monitoring single point energy calculations of large poly(ethylene oxide) molecules and three dimensional carbon monoxide clusters using an extended Huckel Hamiltonian are presented.