Robust approximate zeros in Banach space (Q2465576)

This paper is about the Newton method for analytic functions in Banach spaces, and verification of its convergence based on initial points. Three computational models are discussed: 1) The exact model where the Newton iteration is computed without error; 2) the weak model where all operations in the Newton iteration are computed to a fixed precision; 3) the strong model where all operations can be done with varying precision. The concept of approximate zeros, i.e., the initial points from which the Newton method converges, is extended to both weak and strong models and error estimates are refined. Furthermore, for systems of integer polynomials in Banach spaces the bound of complexity of the strong model is estimated with the condition number of the system.