No label defined (Q4528683)

Let \(X, Y\) be Hilbert spaces and \(A: X\to Y\) be a linear bounded operator with a nonclosed range. The implicit iterative methods \[ (A^*A+\alpha_kI)x_k=A^*y+\alpha_kx_{k-1}, \qquad k=1,2,\dots,x_0=0 \] and its noisy counterpart \[ (A^*A+\alpha_kI)x_k^\delta=A^*y_\delta+\alpha_kx_{k-1}^\delta, \qquad k=1,2,\dots, x_0=0 \] are considered as methods of regularization of the ill-posed problem \(Ax=y\) with the exact and the perturbed right hand sides, respectively. In contradiction to the previously investigated implicit iterative schemes by \textit{G. M. Vainikko} [Zh. Vychisl. Mat. Mat. Fiz. 22, 499-515 (1982; Zbl 0528.65039)] and by \textit{G. He} and \textit{L. Liu} [J. Comput. Math. 17, No. 3, 275-284 (1999; Zbl 0943.65067)] the parameters \(\alpha_k\) are not assumed to be one fixed constant. Under the assumption \(0<\underline\alpha\leq\alpha_k \leq\overline\alpha<\infty\) the authors obtain the same convergence results and convergence rate as in the case of a constant \(\alpha\). A priori and a posteriori (Morozov discrepancy principle) strategies of choosing \(k\) (depending on the data error) are considered. Moreover, it is shown that in the case \(\alpha_i\not=\alpha_j\) for \(i\not=j\) the parallel computation can be implemented which explains what is the reason of investigating the method for different \(\alpha_k\).