Convergence Analysis of a Locally Accelerated Preconditioned Steepest Descent Method for Hermitian-Definite Generalized Eigenvalue Problems

Cai, Yunfeng; Bai, Zhaojun; Pask, John E.; Sukumar, N.

doi:10.4208/jcm.1703-m2016-0580

Title: Convergence Analysis of a Locally Accelerated Preconditioned Steepest Descent Method for Hermitian-Definite Generalized Eigenvalue Problems

Abstract

By extending the classical analysis techniques due to Samokish, Faddeev and Faddeeva, and Longsine and McCormick among others, we prove the convergence of preconditioned steepest descent with implicit deflation (PSD-id) method for solving Hermitian-definite generalized eigenvalue problems. Moreover, we derive a nonasymptotic estimate of the rate of convergence of the PSD-id method. We show that with a proper choice of the shift, the indefinite shift-and-invert preconditioner is a locally accelerated preconditioner, and is asymptotically optimal that leads to superlinear convergence. Numerical examples are presented to verify the theoretical results on the convergence behavior of the PSDid method for solving ill-conditioned Hermitian-definite generalized eigenvalue problems arising from electronic structure calculations. While rigorous and full-scale convergence proofs of preconditioned block steepest descent methods in practical use still largely eludes us, we postulate that the theoretical results introduced in this paper sheds light on an improved understanding of the convergence behavior of these block methods.

Authors:

Cai, Yunfeng ^[1]; Bai, Zhaojun ^[2]; Pask, John E. ^[3]; Sukumar, N. ^[2]

Peking Univ., Beijing (China)
Univ. of California, Davis, CA (United States)
Lawrence Livermore National Lab. (LLNL), Livermore, CA (United States)

Publication Date:: Fri Jun 01 00:00:00 EDT 2018

Research Org.:: Lawrence Livermore National Lab. (LLNL), Livermore, CA (United States)

Sponsoring Org.:: USDOE National Nuclear Security Administration (NNSA)

OSTI Identifier:: 1526870

Report Number(s):: LLNL-JRNL-776239
Journal ID: ISSN 0254-9409; 968787

Grant/Contract Number:: AC52-07NA27344; DMS-1522697; CCF-1527091

Resource Type:: Accepted Manuscript

Journal Name:: Journal of Computational Mathematics

Additional Journal Information:: Journal Volume: 36; Journal Issue: 5; Journal ID: ISSN 0254-9409

Country of Publication:: United States

Language:: English

Subject:: 97 MATHEMATICS AND COMPUTING; Eigenvalue problem; Steepest descent method; Preconditioning; Superlinear convergence

Citation Formats


                    Cai, Yunfeng, Bai, Zhaojun, Pask, John E., and Sukumar, N. Convergence Analysis of a Locally Accelerated Preconditioned Steepest Descent Method for Hermitian-Definite Generalized Eigenvalue Problems.  United States: N. p., 2018. 
Web.  doi:10.4208/jcm.1703-m2016-0580.

Copy to clipboard


                    Cai, Yunfeng, Bai, Zhaojun, Pask, John E., & Sukumar, N. Convergence Analysis of a Locally Accelerated Preconditioned Steepest Descent Method for Hermitian-Definite Generalized Eigenvalue Problems.  United States.  https://doi.org/10.4208/jcm.1703-m2016-0580

Copy to clipboard


                    Cai, Yunfeng, Bai, Zhaojun, Pask, John E., and Sukumar, N. Fri .  
"Convergence Analysis of a Locally Accelerated Preconditioned Steepest Descent Method for Hermitian-Definite Generalized Eigenvalue Problems".  United States.  https://doi.org/10.4208/jcm.1703-m2016-0580.  https://www.osti.gov/servlets/purl/1526870.

Copy to clipboard


                    
@article{osti_1526870,

  title        = {Convergence Analysis of a Locally Accelerated Preconditioned Steepest Descent Method for Hermitian-Definite Generalized Eigenvalue Problems},

  author       = {Cai, Yunfeng and Bai, Zhaojun and Pask, John E. and Sukumar, N.},

  abstractNote = {By extending the classical analysis techniques due to Samokish, Faddeev and Faddeeva, and Longsine and McCormick among others, we prove the convergence of preconditioned steepest descent with implicit deflation (PSD-id) method for solving Hermitian-definite generalized eigenvalue problems. Moreover, we derive a nonasymptotic estimate of the rate of convergence of the PSD-id method. We show that with a proper choice of the shift, the indefinite shift-and-invert preconditioner is a locally accelerated preconditioner, and is asymptotically optimal that leads to superlinear convergence. Numerical examples are presented to verify the theoretical results on the convergence behavior of the PSDid method for solving ill-conditioned Hermitian-definite generalized eigenvalue problems arising from electronic structure calculations. While rigorous and full-scale convergence proofs of preconditioned block steepest descent methods in practical use still largely eludes us, we postulate that the theoretical results introduced in this paper sheds light on an improved understanding of the convergence behavior of these block methods.},

  doi          = {10.4208/jcm.1703-m2016-0580},

  journal      = {Journal of Computational Mathematics},

  number       = 5,

  volume       = 36,

  place        = {United States},

  year         = {Fri Jun 01 00:00:00 EDT 2018},

  month        = {Fri Jun 01 00:00:00 EDT 2018}

}

Copy to clipboard

Journal Article:

Free Publicly Available Full Text

Accepted Manuscript (DOE)

Publisher's Version of Record

https://doi.org/10.4208/jcm.1703-m2016-0580

Other availability

Search WorldCat to find libraries that may hold this journal

Citation Metrics:

Cited by: 1 work

Citation information provided by
Web of Science

Figures / Tables:

Algorithm 2.1: PSD-id

All figures and tables (3 total)

Save / Share:

Export Metadata

Save to My Library

Figures / Tables found in this record:

Algorithm 2.1 (p. 6)

Fig. 5.1 (p. 22)

Fig. 5.2 (p. 23)

Figures/Tables have been extracted from DOE-funded journal article accepted manuscripts.

Similar Records in DOE PAGES and OSTI.GOV collections:

Preconditioned steepest descent methods for some nonlinear elliptic equations involving p-Laplacian terms

Journal Article Feng, Wenqiang ; Salgado, Abner J., E-mail: asalgad1@utk.edu ; Wang, Cheng ; ... - Journal of Computational Physics

We describe and analyze preconditioned steepest descent (PSD) solvers for fourth and sixth-order nonlinear elliptic equations that include p-Laplacian terms on periodic domains in 2 and 3 dimensions. The highest and lowest order terms of the equations are constant-coefficient, positive linear operators, which suggests a natural preconditioning strategy. Such nonlinear elliptic equations often arise from time discretization of parabolic equations that model various biological and physical phenomena, in particular, liquid crystals, thin film epitaxial growth and phase transformations. The analyses of the schemes involve the characterization of the strictly convex energies associated with the equations. We first give a generalmore »« less
https://doi.org/10.1016/J.JCP.2016.12.046
Hybrid preconditioning for iterative diagonalization of ill-conditioned generalized eigenvalue problems in electronic structure calculations

Journal Article Cai, Yunfeng ; Department of Computer Science, University of California, Davis 95616 ; Bai, Zhaojun ; ... - Journal of Computational Physics

The iterative diagonalization of a sequence of large ill-conditioned generalized eigenvalue problems is a computational bottleneck in quantum mechanical methods employing a nonorthogonal basis for ab initio electronic structure calculations. We propose a hybrid preconditioning scheme to effectively combine global and locally accelerated preconditioners for rapid iterative diagonalization of such eigenvalue problems. In partition-of-unity finite-element (PUFE) pseudopotential density-functional calculations, employing a nonorthogonal basis, we show that the hybrid preconditioned block steepest descent method is a cost-effective eigensolver, outperforming current state-of-the-art global preconditioning schemes, and comparably efficient for the ill-conditioned generalized eigenvalue problems produced by PUFE as the locally optimal blockmore »« less
https://doi.org/10.1016/J.JCP.2013.07.020
Polynomial preconditioning for conjugate gradient methods

Technical Report Ashby, S F

The solution of a linear system of equations, Ax = b, arises in many scientific applications. If A is large and sparse, an iterative method is required. When A is hermitian positive definite (hpd), the conjugate gradient method of Hestenes and Stiefel is popular. When A is hermitian indefinite (hid), the conjugate residual method may be used. If A is ill-conditioned, these methods may converge slowly, in which case a preconditioner is needed. In this thesis we examine the use of polynomial preconditioning in CG methods for both hermitian positive definite and indefinite matrices. Such preconditioners are easy to employmore »« less
Accelerated convergence of the steepest-descent method for magnetohydrodynamic equilibria

Technical Report Handy, C. R. ; Hirshman, S. P.

Iterative schemes based on the method of steepest descent have recently been used to obtain magnetohydrodynamic (MHD) equilibria. Such schemes generate asymptotic geometric vector sequences whose convergence rate can be improved through the use of the epsilon-algorithm. The application of this nonlinear recursive technique to stiff systems is discussed. In principle, the epsilon-algorithm is capable of yielding quadratic convergence and therefore represents an attractive alternative to other quadratic convergence schemes requiring Jacobian matrix inversion. Because the damped MHD equations have eigenvalues with negative real parts (in the neighborhood of a stable equilibrium), the epsilon-algorithm will generally be stable. Concern formore »« less
https://doi.org/10.2172/7045611

Full Text Available
Deflation as a method of variance reduction for estimating the trace of a matrix inverse

Journal Article Gambhir, Arjun Singh ; Stathopoulos, Andreas ; Orginos, Kostas - SIAM Journal on Scientific Computing

Many fields require computing the trace of the inverse of a large, sparse matrix. The typical method used for such computations is the Hutchinson method which is a Monte Carlo (MC) averaging over matrix quadratures. To improve its convergence, several variance reductions techniques have been proposed. In this paper, we study the effects of deflating the near null singular value space. We make two main contributions. First, we analyze the variance of the Hutchinson method as a function of the deflated singular values and vectors. Although this provides good intuition in general, by assuming additionally that the singular vectors aremore »« less
Cited by 17
https://doi.org/10.1137/16M1066361

Full Text Available

Similar Records