A convergence analysis for directional two-step Newton methods

From MaRDI portal





The authors are concerned with the problem of approximating a zero \(x^{*}\) of a differentiable function \(F\) defined on a convex subset \(\mathcal{D}\subset \mathcal{H}\) with values in \(\mathbb{R}\), \(\mathcal{H}\) being Hilbert. They provide a semilocal convergence analysis of the two step directional Newton method (TSDNM) starting from \(x_{0}\in\mathcal{D}:\) \[ \begin{aligned} &y_{n}=x_{n}-\frac{F(x_{n})}{\nabla F(x_{n})\cdot d_{n}}d_{n} \\ &x_{n+1}=y_{n}-\frac{F(y_{n})}{\nabla F(x_{n})\cdot d_{n}}d_{n}. \end{aligned} \] Here \(d_{n}\) are directions sufficiently close to the gradient, satisfying e.g., a condition: \[ \text{For } \| d_{n}\|=1, \quad \exists\gamma\in[0,1] \quad \text{such that } |\nabla F(x_{n})\cdot d_{n}|\geq\gamma\|\nabla F(x_{n})\|. \] Two different techniques to generate sufficient semilocal convergence conditions are provided together withe corresponding error bounds for the cubically convergent (TSDNM). The first technique uses the concept of recurrent sequences, the second uses a new idea of the authors' on recurrent functions. It is also shown that the order of convergence is three.




Cited in
(21)








This page was built for publication: A convergence analysis for directional two-step Newton methods

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q607524)