Two complex n x n matrices A and B are said to be partially commuting if A
and B have a common eigenvector. We propose a condensed form for such matri
ces that can be obtained from A and B by a finite rational computation. The
condensed form is a pair of block triangular matrices, with the sizes of t
he blocks being uniquely defined by the original matrices, We then show how
to obtain additional zeros inside the diagonal blocks of a condensed form
by using the generalized Lanczos procedure as given by Elsner and Ikramov.
This procedure can also be considered as a finite rational process. We poin
t out several applications of the constructions above. It turns out that fo
r Laffey pairs of matrices, i.e., for matrices (A, B) such that rank[A, B]
= 1, the condensed form is a pair of 2 x 2 block triangular matrices. Using
this fact, we show an economical way to find a spanning set for the matrix
algebra,generated by Laffey matrices A and B. Another application concerns
so-called k-self-adjoint matrices. We examine such matrices in the unitary
space as well as in a Krein space of defect 1. As an Appendix, we give a n
ew description of the Shemesh subspace of matrices A and B. This is the max
imal common invariant subspace of A and B, on which these matrices commute.
(C) 2000 Elsevier Science Inc. All rights reserved.