Efficient Hartree-Fock Exchange Algorithm with Coulomb Range Separation and Long-Range Density Fitting

TL;DR

This paper introduces a novel Hartree-Fock exchange algorithm that combines Coulomb range separation with long-range density fitting, significantly improving efficiency and accuracy for large molecules.

Contribution

The paper presents a combined Coulomb range separation and long-range density fitting approach that reduces computational cost and enhances accuracy in Hartree-Fock calculations.

Findings

Outperforms conventional methods in large molecules

Achieves over twice the performance of traditional algorithms

Reduces energy error to 0.1 μE_h per atom

Abstract

Separating the Coulomb potential into short-range and long-range components enables the use of different electron repulsion integral algorithms for each component. The short-range part can be efficiently computed using the analytical algorithm due to the locality in both Gaussian-type orbital basis and the short-range Coulomb potentials. The integrals for the long-range Coulomb potential can be approximated with the density fitting method. A very small auxiliary basis is sufficient for the density fitting method to accurately approximate the long-range integrals. This feature significantly reduces the computational efforts associated with the $N^4$ scaling in density fitting algorithms. For large molecules, the range separation and long-range density fitting method outperforms the conventional analytical integral evaluation scheme employed in Hartree-Fock calculations and provides more than twice the overall performance. Additionally, this method yields higher accuracy compared to regular density fitting methods. The error in the Hartree-Fock energy can be easily reduced to 0.1 $\mu E_h$ per atom, which is significantly more accurate than the typical error of 10 $\mu E_h$ per atom observed in regular density fitting methods.