Friday, January 18, at 327 Yost
Refreshments: 3:00 - 3:30 p.m, Talk: 3:30
- 4:30 p.m.
The aim of most research in Scientific Computing is to develop new efficient
numerical methods for the solution of large problems in Science and
Engineering. Due to measurement errors in the data, the need to terminate
the computations in reasonable time, and the propagation of round-off
errors, one generally only can compute approximate solutions to large
problems. This talk proposes some new techniques to assess the accuracy of
the computed approximate solutions of large linear systems of equations,
using orthogonal polynomials and Gauss quadrature rules to inexpensively
compute upper and lower bounds of certain matrix functionals. The
successful application of these techniques to the determination of the value
of the regularization parameter for large ill-posed problems also will be
described.