The numerical stability of the generalised root iterations for polynomial zeros (Q796253)

The generalised root iterations for simultaneous finding polynomial complex zeros, with the convergence order \(k+2\) (\(k\in N)\), were proposed by the first author [Computing 27, 37-55 (1981; Zbl 0442.65030)] in terms of circular regions. The method produces the disks which contain the exact zeros, giving error bounds automatically for each approximation. This paper investigates the numerical stability of this algorithm in the presence of rounding errors. It is proved that the convergence order remains \(k+2\), the same as in the absence of round-off if: (I) the absolute value of the round-off with which the polynomial is evaluated in the proximity of a zero is small compared to the absolute value of the polynomial itself and (II) the rounding error is small as compared to the radius of the disk including the exact zero. If the condition (II) is weakened so that the round-off is independent of the disk size, although ''reasonably small'', then the convergence order reduces to \(k+1\).