On Levenberg-Marquardt-Kaczmarz iterative methods for solving systems of nonlinear ill-posed equations (Q601669)

The article deals with the following system of equations \[ F_i(x) = y_i, \quad i = 0,1,\dots,N - 1,\tag{1} \] where \(F_i: \;D_i \subset X \to Y\) are smooth operators between Hilbert spaces \(X\) and \(Y\). The following iterations \[ x_{k+1}^\delta = x_k^\delta - \omega_k(F_{[k]}'(x_k^\delta)^*F_{[k]}'(x_k^\delta) + \alpha I)^{-1}F_{[k]}'(x_k^\delta)^*(F_{[k]}(x_k^\delta) - y_{[k]}^\delta)\tag{2} \] with \(\omega_k = 1\) if \(\|F_{[k]}(x_k^\delta) - y_{[k]}^\delta\| \geq \tau \delta_{[k]}\) and \(\omega_k = 0\) otherwise are studied. Here \([k] = r \, (\text{mod} \, N) \in \{0,1,\dots,N - 1\}\), \(x_0^\delta = x_0\) is an initial approximation, \(\tau > 1\) a fixed constant, \(y_i^\delta\) are approximate measured data for \(y_i\) satisfying inequalities \(\|y_i^\delta - y_i\| \leq \delta_i\), \(i = 0,1,\dots,N - 1\); the stop-rule is defined by \[ k_*^\delta = \min \;\{lN \in {\mathbb N}: \;x_{lN}^\delta = x_{lN+1}^\delta = \dots = x_{lN+N}^\delta\}. \] These iterations are called loping Levenberg-Marquardtcon-Kaczmarz ones. In the article they are considered under the assumptions that the domains of \(F_i\) have common nonempty interior, \(F_i\) and \(F_i'\) are continuous and bounded and that the local tangential cone condition \[ \|F_i(\overline{x}) - F_i(x) - F_i(x)(\overline{x} - x)\| \leq \eta \|F_i(\overline{x}) - F_i(x)\| \] holds with some \(\eta < 1\). Under these assumptions the authors prove a semi-convergence for the approximations (2) provided that system (1) has an exact solution for exact data \(y_i\). In the end of the article a numerical experiment for an inverse doping problem is presented.