Skip to:

CME 510: Linear Algebra Optimization Seminar

Speaker: James Lambers, Dept of Mathematics, University of Southern Mississippi

Title: A Crash Course on Matrices, Moments and Quadrature
 
Abstract: The aim of this talk is to give an overview of the beautiful mathematical relationships between matrices, moments, orthogonal polynomials, quadrature rules, and Krylov subspace methods.  The underlying goal is to obtain efficient numerical methods for estimating quantities of the form u^T f(A) v, where u and v are given vectors, A is a symmetric nonsingular
matrix, and f is a smooth function.
 
An obvious application is the computation of some elements of the matrix f(A) when all of f(A) is not required.  Computation of quadratic forms can yield error estimates in methods for solving systems of linear equations. Bilinear or quadratic forms also arise naturally for the computation of parameters in some numerical methods for solving least squares or total least squares problems, and also in Tikhonov regularization for solving ill-posed problems.  Furthermore, computation of bilinear forms is also useful in spectral methods for the numerical solution of PDE.
 
The main idea is to write I[f] as a Riemann-Stieltjes integral and then to apply Gaussian quadrature rules to approximate the integral.  The nodes and weights of these quadrature rules are given by the eigenvalues and eigenvectors of tridiagonal matrices whose nonzero coefficients describe the three-term recurrences satisfied by the orthogonal polynomials associated with the measure of the Riemann-Stieltjes integral.  Beautifully, these orthogonal polynomials can be generated by the Lanczos algorithm.
 
Results pertaining to orthogonal polynomials and quadrature rules may not be so well known in the matrix computation community, and conversely, researchers working with orthogonal polynomials and quadrature rules may not be very familiar with their applications to matrix computations.  We will see that it can be very fruitful to mix techniques coming from different areas. The resulting algorithms can also be of interest to scientists and engineers who are solving problems in which computation of bilinear forms arises naturally.
 
Bio: Jim Lambers graduated in 2003 from Gene Golub's SCCM program (the forerunner of ICME).  His current research is on spectral methods for time-dependent variable-coefficient PDE; simulation of gas injection processes for enhanced oil recovery (in collaboration with Margot Gerritsen), and denoising of images via nonlinear diffusion (in collaboration with Patrick Guidotti).
 
Location: 

Y2E2-101

Organizer: 
Michael Saunders
Date: 
Thursday, May 26, 2016 -
4:30pm to 5:45pm