Inexact Newton regularization methods in Hilbert scales (Q629902)

The article deals with a new iterative scheme for approximate solving a nonlinear operator equation \[ F(x) = y\tag{1} \] where \(F:\;D(F) \subset X \to Y\) is a nonlinear Fréchet differentiable operator between two Hilbert spaces \(X\) and \(Y\). It is assumed that (1) has a solution \(x^*\), however the invertibility of \(F'(x)\) is not assumed. The following inexact Newton method \[ x_{n+1} = x_n - g_{\alpha_n}(L^{-2s}F'(x_n)^*F'(x_n))L^{-2s}F'(x_n)^*(F(x_n) - y^\delta),\tag{2} \] is considered, where \(L\) is a densely defined self-adjoint strictly positive definite linear operator in \(X\), \(s \in {\mathbb R}\), \(\alpha_n\) is an a priori given sequence of positive numbers with suitable properties, \(g\) is a given spectral filter function. Moreover, the iterations (2) are terminated by the discrepancy principle \[ \|F(x_{n_\delta}) - y^\delta\| \leq \tau\delta < \|F(x_n) - y^\delta\|, \quad 0 \leq n < n_\delta \] with a given number \(\tau > 1\). The main result is concerned with the behavior of \(x_{n_\delta} \to x^*\) as \(\delta \to 0\). In particular, the authors present conditions under that the estimate \(\|x_{n_\delta} - x^*\| = O(\delta^\frac{\mu-r}{a+\mu})\) holds; here \(a\) and \(\mu\) are some numerical characteristics of operators \(F'(x)\). In the end of the article there are presented 4 numerical examples with different spectral filter functions \(g\)