On the local and semilocal convergence of a parameterized multi-step Newton method (Q1987448)

This paper is devoted to a family of Newton-like methods with frozen derivatives used to approximate a locally unique solution of the equation \(F(x)=0\). The authors study the method defined for each \(n=0,1,2,\ldots\) (base part) by: \[ \begin{array}{l} F'(y_0^{(n)})\phi_1=F(y_0^{(n)}), \\[6pt] y_1^{(n)}=y_0^{(n)}-(1+\theta-\theta^2)\phi_1, \\[4pt] F'(y_0^{(n)})\phi_2=F(y_0^{(n)}-\frac{1}{\theta}\phi_1), \\[4pt] y_2^{(n)}=y_1^{(n)}-\theta^2\phi_2, \end{array} \] and for each \(i=1,\ldots ,n-2\) (multi-step part) by: \[ \begin{array}{l} F'(y_0^{(n)})\phi_{i+2}=F(y_{i+1}^{(n)}), \\[4pt] y_{i+2}^{(n)}=y_{i+1}^{(n)}-\phi_{i+2}. \end{array} \] A study of the convergence study and an analysis of the efficiency are given. By an appropriate selection of the parameter \(\theta\), the new generalized multi-step Newton method can have both a faster convergence and a larger radius of convergence. The method can be applied to many type of problems, including the discretization of ordinary differential equations, integral equations, integro-differential equations or partial differential equations.