Date:

Monday, September 30, 2002

Speaker:

Junping Wang

Affiliation:

Department of Mathematics, Colorado School of Mines

e-mail:

jwang@ito.mines.edu

Title:

A Fast Multilevel Multiplication Method with Applications in Data Assimilation

Abstract:

The matrix-vector multiplication normally requires n^2 operations for a dense matrix A of size n-by-n. For large values of n (say n=10^8), the standard matrix-vector multiplication is known to be the bottleneck of many iterative solvers such as PCG for the matrix problem Ax=b. In this talk, we shall consider a class of matrices, for which a fast multilevel multiplication method will be designed by using a multilevel decomposition for the matrix A. The method is closely related to the fast multipole method that was introduced for N-body problems decades ago. Numerical experiments will be shown to demonstrate the efficiency of the method. Application to data assimilation will be discussed. The research was initiated during the 2002 summer visit to NASA's Goddard Space and Flight Center (GSFC) in Maryland. This is a joint work with Tom Clune of NCCS at GSFC.