Numerical solution of matrix eigenvalue problems

This page contains material from a minicourse at the Zurich summer school 2008 on eigenvalue problems.

Part 0: Introduction

lecture0.pdf - slides from the presentation [more printable 2p1 version]

Part 1: Power method and friends

lecture1.pdf - Slides from the presentation [more printable 2p1 version]

transient.m - Matlab script demonstrating that the initial convergence of the power method may be erratic for nonsymmetric matrices. [Inspired by the Trefethen/Embree book on Pseudospectra.]
ssiter.m - Matlab script demonstrating the convergence of Ritz values during the subspace iteration.
rqidomain.m - Global convergence of the Rayleigh quotient iteration for a 3-by-3 matrix. [Inspired by similar examples from Pierre-Antoine Absil's PhD thesis.]

Part 2: Krylov subspace methods

lecture2.pdf - slides from the presentation [more printable 2p1 version]

loss.m - Loss of orthogonality in the Arnoldi method.
noloss.m - No loss of orthogonality in the Arnoldi method.
arngo.m - Convergence of Ritz values produced by the Arnoldi method. [Written by David Watkins.]
arngo2.m - Convergence of Ritz values produced by the shift-and-invert Arnoldi method. [Written by David Watkins.]
losslan.m - Loss of orthogonality in the Lanczos method.
nolosslan.m - No loss of orthogonality in the Lanczos method.

Part 3: Preconditioned Eigensolvers

lecture3.pdf - slides from the presentation [more printable 2p1 version]

ppd.m - Convergence of the preconditioned inverse iteration for three different preconditioners [Based on an implementation from the Numerics for CSE lecture notes.]
runlopgc.m - Convergence of the locally optimal preconditioned conjugate gradient method for three different preconditioners [Implementation of LOPCG based on Peter Arbenz' course notes.]
runjdqr.m - Convergence of Jacobi-Davidson for three different preconditioners [Requires the file jdqr.m.gz from G.L.G. Sleijpen.]

Additional Resources

Eigenvalue Templates [For C++ implementations of some of the algorithms, see the The Iterative Eigensolver Template Library.]
Jack Dongarra's list on freely available software for linear algebra
Matlab programs for David Watkins' book The Matrix Eigenvalue Problem: GR and Krylov Subspace Methods.
Matlab programs for Jim Demmel's book Applied Numerical Linear Algebra.
ARPACK
SLEPc
Trilinos
Jacobi-Davidson gateway

For the class room presentation of Matlab snippets it can be helpful to increase the font size and line widths of a figure, see makelarge.m.

Contact

Daniel Kressner