On the use of incomplete LU decomposition as a preconditioning technique for density fitting in electronic structure computations

Authors:
Rui Yang;Alistair P. Rendell;Michael J. Frisch
Affiliations:
Department of Computer Science, College of Engineering and Computer Science, The Australian National University, Canberra, ACT, Australia;Department of Computer Science, College of Engineering and Computer Science, The Australian National University, Canberra, ACT, Australia;Gaussian Inc., Wallingford, CT
Venue:
ICCSA'07 Proceedings of the 2007 international conference on Computational science and its applications - Volume Part I
Year:
2007

Citing 4
Cited 0

GMRES: a generalized minimal residual algorithm for solving nonsymmetric linear systems

SIAM Journal on Scientific and Statistical Computing
BILUM: Block Versions of Multielimination and Multilevel ILU Preconditioner for General Sparse Linear Systems

SIAM Journal on Scientific Computing
Preconditioning techniques for large linear systems: a survey

Journal of Computational Physics
Iterative Methods for Sparse Linear Systems

Iterative Methods for Sparse Linear Systems

Quantified Score

Hi-index	0.00

Visualization

Abstract

Incomplete factorization preconditioners combined with Krylov subspace accelerators are currently among the most effective methods for iteratively solving large systems of linear equations. In this paper we consider the use of a dual threshold incomplete LU factorization (ILUT) preconditioner for the iterative solution of the linear equation systems encountered when performing electronic structure calculations that involve density fitting. Two questions are addressed, how the overall performance of the ILUT method varies as a function of the accuracy of the preconditioning matrix, and whether it is possible to make approximations to the original matrix on which the LU decomposition is based and still obtain a good preconditioner. With respect to overall performance both computational and memory storage requirements are considered, while in terms of approximations both those based on numerical and physical arguments are considered. The results indicate that under the right circumstances the ILUT method is superior to fully direct approaches such as singular value decomposition.