New improved convergence analysis for Newton-like methods with applications (Q683881): Difference between revisions

The article deals with the Newton-Kantorovich approximations \[ x_{n+1} = x_n - T_nF(x_n), \quad n = 0,1,2,\ldots, \] for approximatively solving the nonlinear operator equation \(F(x) = 0\) with a nonlinear operator \(F\) between Banach spaces \(X\) and \(Y\). The authors prove two theorems. In the first of them it is assumed that \(F\) is defined and Fréchet differentiable on a convex open domain \({\mathcal D}\), \(x_0, x_{-1} \in {\mathcal D}\) and \(\| x_0 - x_{-1}\| \leq b\), that there exists a mapping \(\delta F:\;{\mathcal D} \times {\mathcal D} \to L(X,Y)\) such that \(T_0\) (either \(T_0 = \delta F(x_0,x_{-1})\) or \(T_0 = \delta F(x_{-1},x_0)\)) and that the following conditions hold: \[ \begin{aligned} \| T_0F(x_0)\| &\leq c\|,\\ \| T_0(\delta F(x_1,x_2) - F'(x_0))\| &\leq k_0\| x_1 - x_0\| + k\| x_2 - x_0\|,\\ \| T_0(\delta F(x_1,x_2) - F'(z))\| &\leq k_1\| x_1 - z\| + k_2\| x_2 - z\| \end{aligned} \] with some constants \(b\), \(c\), \(k_0\), \(k\), \(k_1\), \(k_2\) satisfying some inequalities. Under these conditions and \(T_n = \delta F(x_{p_n},x_{q_n})^{-1}\) (\((p_n)\) and \((q_n)\) are two nondecreasing sequences of integers satisfying \(q_0 = -1\), \(p_0 = 0\), \(q_n \leq p_n \leq n\) for \(n = 0,1,2,\ldots\)) the approximations are well defined, remain in a ball with the center \(x_0\) and converge to a solution \(x_*\) of the equation \(F(x) = 0\). The second theorem is similar, but it is stated under the following conditions: \(F(x_*) = 0\), \(F'(x_*)^{-1} \in L(Y,X)\), \[ \begin{aligned} \|[F'(x_*)^{-1}](\delta F(x_1,x_2) - F'(x_*)) \| &\leq l_0\| x_1 - x_*\| + l\| x_2 - x_*\|,\\ \|[F'(x_*)^{-1}](\delta F(x_1,x_2) - F'(z)) \| &\leq l_1\| x_1 - z\| + l_2\| x_2 - z\|. \end{aligned} \] Two numerical examples are presented.