Thursday,
March 20, 12:15 p.m. PSA 206
Iveta Hnetynkova
Dept Math & Stats
Noise-Revealing Golub-Kahan Bidiagonalization with
Application in Hybrid Methods
Abstract
Regularization techniques based on Golub-Kahan bidiagonalization have been used
for the iterative solution of large ill-posed problems for years. First, the
original problem is projected onto a lower dimensional subspace using the
bidiagonalization algorithm and then some type of inner regularization and
parameter selection method, e.g. L-curve, the discrepancy principle, or
generalized cross validation, is applied to it. This also leads to a decision
when it is optimal to stop the bidiagonalization.
Recently, it has been proved that the Golub-Kahan bidiagonalization leads to a
fundamental decomposition of data, revealing the so-called core problem.
Applications to ill-posed problems have been studied by D. Sima, S. Van Huffel,
P.C. Hansen, etc.
In this contribution we consider an ill-posed problem with noisy right-hand
side and study how the noise in the data enters the projected problem obtained
by the bidiagonalization. We investigate a possibility of directly using this
information for constructing an effective stopping criterion in solving
ill-posed problems.
For further information please contact:
mittelmann@asu.edu