A Runge-Kutta type modified Landweber method for nonlinear ill-posed operator equations (Q2468149): Difference between revisions

Let \(X,Y\) be Hilbert spaces and \(F:D(F)\subseteq X\to Y\) be a nonlinear map. Assume that the operator equation \(F(x)=y\) has a solution \(x_*\) and let \(y^\delta\) be an available approximation data with \(| | y^\delta-y| | \leq \delta\) (=the noise level). The Runge-Kutta type modified Landweber method to be considered is: \[ \begin{aligned} x_{k+1}^\delta&=x_k^\delta-F'(z_k^\delta)^*(F(z_k^\delta)-y^\delta)- \alpha_k(z_k^\delta-\zeta),\\ z_k^\delta&=x_k^\delta-(1/2)F'(x_k^\delta)^*(F(x_k^\delta)-y^\delta)- (1/2)\alpha_k(x_k^\delta-\zeta),\end{aligned} \qquad k\geq 0. \] The following are proved by the authors: (i) this method is convergent and stable for an appropriate choice of \((\alpha_k)\), (ii) a convergence rate evaluation is available under the usual sourcewise representation \(x_*-\zeta=F'(x_*)^*\omega\).