Carpe D.I.E.M.

Dynamics of Iterative Eigenvalue Methods

Last updated: August 4, 2003

This is the home of the research of Anthony Kellems and Dr. Mark Embree, Rice University, Summer 2003.

Our research consisted of looking at two main methods for solving eigenvalue problems. The first method, Rayleigh Quotient Iteration, was studied to discover how choices of initial starting vectors affected convergence. We developed some visual tools to let us analyze these problems in a very attractive manner for the 3-by-3 cases, and thus our attention was focused for problems of that order. Other visual tools let us examine certain aspects of RQI for matrices in higher dimensions, but the results, although somewhat visually interesting, were not very substantial.

The second method was Arnoldi's Method, an iterative scheme used to find a small number of eigenvalues for a very large matrix. Although it's main applications are for matrices of dimension 10,000 or greater, we found it beneficial to understanding the dynamics of this process by studying the 3-by-3 case and utilizing the visual tools we previously developed for RQI.

The following are links to some of the images, graphs, and results of our research.

Alpha Pics--A gallery of RQI convergence images organized by matrix class. Movie files can also be found here which show the evolution of the unit sphere and pseudospectra for each class.

A Visual Analysis of RQI--PDF file of the final report for our RQI research. It is entitled "A Visual Analysis of Convergence of Iterative Eigenvalue Methods". August 1, 2003.

Relevant Files--PDF file indicating and briefly annotating the important MATLAB scripts, graphs, and images produced during our research.

MATLAB Goodies--Critical MATLAB m-files written for this research. This code can be used to verify our findings or to explore RQI and Arnoldi's Method further.