Error complexity analysis of two algorithms for computing powers of arbitrary Hessenberg matrices (Q1090071)

The author shows that the standard matrix multiplication method for computing powers of a Hessenberg matrix has a better guaranteed numerical accuracy than a new ''efficient'' algorithm proposed by \textit{C. P. Huang} [Linear Algebra Appl. 21, 123-134 (1978; Zbl 0396.15002)]. Moreover, examples are given which demonstrate that even low order powers of a Hessenberg matrix with small superdiagonal elements computed by the new method are incorrect when floating point computations are used.