A unified approach for the convergence of certain numerical algorithms, using recurrent functions (Q605045)

The Kantorovich-analysis deals with an approximating solution of an abstract equation of the type \(F(x)+ G(x)= 0\), where \(F\) is a Fréchet-differentiable operator on an open subset of a Banach space \(X\) and \(G\) is a continuous operator. The first Newton-like iterative scheme for calculating an approximating sequence \(\{x_n\}\subseteq X\) was presented by \textit{L. V. Kantorovich} and \textit{G. P. Akilov} in their famous Russian monograph [Functional analysis in normed spaces. Moskva: Staatlicher Verlag für Physikalisch-mathematische Literatur. 684 pp. (1959; Zbl 0127.06102)]. The authors refer to an edition of 1982. There are a lot of papers on this topic. In the last time Argyros and Hilout published some generalized results. The sequence is constructed as \[ x_{n+1}= x_n- A_n^{-1}[F(x_n)- G(x_n)], \] where \(A_n\) can be considered as an approximation of \(F'(x_N)\). Some authors used a recurrent relation's approach under Lipschitz-Hölder-type conditions to provide a semilocal convergence. In the present paper the same approach is used to provide a finer convergence analysis. The advantages are a larger convergence domain, finer error bounds on the distance \(\| x_n- x^*\|\) and a precise information on the location of the solution \(x^*\). The method is applicable to locate the roots of polynomials of high order, too. The fundamental idea is the well-known construction of majorizing real sequences in order to formulate sufficient conditions under which the convergence theorem is valid. Numerical examples are considered, in which the convergence is realized by the theorem of this paper, whereas other results are not applicable. A boundary value problem is considered and reformulated as an integral equation by means of a corresponding Green function. In this example the sufficient conditions of the main theorem are fulfilled.