Error bounds in the isometric Arnoldi process (Q1378993): Difference between revisions

An Arnoldi algorithm is applied to a large and sparse unitary matrix. The last column of the Hessenberg matrix is modified in each step, to give a unitary Hessenberg matrix whose eigenvalues interlace those of the original matrix on the unit circle in the complex plane. Error bounds are studied, and it is shown how to do the column modification to get approximations as close to those of the original matrix as possible. Several numerical examples, coming from the analysis of perturbed periodic signals, are included.