Multiple eigenvectors around the homo-lumo gap as a cheap by-product in linear scaling electronic structure calculations

TL;DR

This paper introduces an efficient method for computing multiple eigenvectors near the homo-lumo gap in large-scale electronic structure calculations, leveraging polynomial filtering and the purify-shift-and-project technique to improve convergence.

Contribution

The paper presents a novel implementation of the purify-shift-and-project method for linear scaling eigenvector computation around the homo-lumo gap, integrated into the Ergo software.

Findings

Efficient computation of 30 eigenvectors around the homo-lumo gap in large systems.

Linear scaling density matrix construction with recursive polynomial expansions.

Enhanced eigenvalue separation using polynomial filtering improves convergence.

Abstract

In this work we present and evaluate an implementation of the purify-shift-and-project method [J. Chem. Phys. 128, 176101 (2008)] for linear scaling computation of multiple eigenvectors around the homo-lumo gap of the Fock/Kohn-Sham matrix. Recursive polynomial expansions allow for linear scaling density matrix construction if matrices are sufficiently sparse. However, a drawback is that, compared to the traditional diagonalization approach, eigenvectors of the Fock/Kohn-Sham matrix are not readily available. The sharp polynomial filter, constructed in intermediate iterations of the recursive polynomial expansion, increases the relative separation of eigenvalues near the homo-lumo gap. The computed density matrix approximation is used to project away the uninteresting part of the spectrum, so that the eigenvalues of interest become the extreme eigenvalues, enabling fast convergence of a Lanczos eigensolver. We implement the purify-shift-and-project algorithm in the quantum chemistry program Ergo [SoftwareX 7, 107 (2018)]. We illustrate the performance of the method by computing 30 eigenvectors around the homo-lumo gap for large scale systems.