On pairs of almost diagonal matrices (Q2638723): Difference between revisions

This paper has two primary goals. One is to study the structure of a pair of almost diagonal matrices (A,B). The other is to apply this theory to investigate the asymptotic convergence of Jacobi methods for the generalized eigenvalue problem \(Ax=\lambda Bx.\) The first part of the paper presents bounds for \(\| A_{ii}- \lambda_ iB_{ii}\|\), where \(A_{ii}\) and \(B_{ii}\) are almost diagonal submatrices associated with the generalized eigenvalue \(\lambda_ i\). The theory covers the cases when A and B or some real combination \(\alpha A+\beta B\) is positive definite. Some of these results are refinements of earlier results by \textit{J. H. Wilkinson} [ibid. 1, 1-12 (1968; Zbl 0167.303)] and \textit{H. P. M. van Kempen} [Numer. Math. 9, 19-22 (1966; Zbl 0229.65038)]. The second part of the paper applies the theory from part one to the asymptotic convergence of Jacobi methods for the generalized eigenvalue problem. In particular, it is demonstrated that in the presence of eigenvalues with a multiplicity larger than two, the convergence of the Jacobi method may fail to be quadratic.