Accurate simple zeros of polynomials in floating point arithmetic

In the paper, we examine the local behavior of Newton’s method in floating point arithmetic for the computation of a simple zero of a polynomial assuming that an good initial approximation is available. We allow an extended precision (twice the working precision) in the computation of the residual. We prove that, for a sufficient number of iterations, the zero is as accurate as if computed in twice the working precision. We provide numerical experiments confirming this. c © 2008 Elsevier Ltd. All rights reserved.