Convergence results of a modified regularized gradient method for nonlinear ill-posed problems (Q711520)

The article deals with the following iterative method \[ x_{k+1} = x_k - \mu_k(\widetilde{F'}^*(x_k)\widetilde{F}(x_k) + \alpha_kL^*L(x_k - \xi)),\eqno(1) \] of solving a nonlinear operator equation \(F(x) = 0\) with a solution \(x^*\). Here \(F\) is a smooth operator between the Hilbert spaces \(X\) and \(X_1\), \(\widetilde{F}\) its smooth approximation, \(L\) a bounded linear operator between Hilbert spaces \(X\) and \(X_2\), \(\mu_k\) and \(\alpha_k\) a priori prescribed values of the stepsize and the regularization parameter, \(\xi \in X\) a parameter. It is assumed that the following conditions \[ \|F'(x) - F'(y)\|, \|\widetilde{F}(x) - \widetilde{F}(y)\| \leq l\|x - y\|, \quad \|x - x^*\|, \|y - x^*\| \leq R, \] \[ \|\widetilde{F}(x^*)\| \leq \delta, \quad \|\widetilde{F}'(x) - F'(x)\| \leq \delta, \quad \|x - x^*\| \leq R {a'}^* \] hold. The authors present two results for the upper bounds of \(\|x_k - x^*\|\), \(k = 0,1,\ldots,k(\delta,\Delta)\), where \(k(\delta,\Delta)\) is determined by the equation \(k(\delta,\Delta) = \max \;\{k:\;m(\delta + \Delta) \leq \alpha_k\}\) in the case of an a priori stopping rule and from the condition \[ \|\widetilde{F}(x_{k(\delta,\Delta)}^\delta)\| \leq \rho\delta < \|\widetilde{F}(x_k)\|, \quad k = 0,1,\dots,k(\delta,\Delta) - 1 \] in the case of an a posteriori stopping rule. Analogues of these results are also given for a continuous version of (1). In the end of the article, some numerical results are presented for the following boundary value problem \[ -(au_x)_x = f \;\;x \in (0,1), \quad u(0) - 2a(0)u_x(0) = 0, \;\;u(1) + 2a(1)u_x(1) = 0. \]