Computation of generalized eigenvectors (Q1824352)

Let us assume that the \(n\times n\) matrix A and its eigenvalues \(\nu_ 1,\nu_ 2,...,\nu_ n\) are given together with some criterion to decide which of these eigenvalues are multiple and that the result of this grouping and approximation is the list of eigenvalues \(\lambda_ 1,\lambda_ 2,...\lambda_ p\), where \(p\leq n\). The \(\lambda_ j's\) are distinct in that \(j\neq k\) implies \(\lambda_ j\neq \lambda_ k.\) The algorithm described in the paper generates the Jordan form of a matrix by simultaneously computing the eigenstructure and associated eigenvector-generalized eigenvector chains. It does this by systematically finding vectors which can serve as ends of chains, starting with the shortest chains and increasing the length until the invariant subspaces are exhausted. The algorithm is straightforward, and can be implemented in code by a computer programmer of average ability.