Padé iteration method for regularization (Q849832)

For the linear operator equation \(Kx=y\) with a compact operator \(K\) acting on an infinite dimensional Hilbert space iterative regularization methods of the form \[ x_{i+1}=(I-\omega T^{-1}K^*K)x_i+\omega K^*y, \quad i=0,1,2,..., \] are considered. This family of methods contains the Landweber method with \(T=I\) as well as the Padé methods when regular preconditioner \(T\) is of \((k,j)\)-Padé type. The authors prove that the Padé method is convergent and of optimal order and has infinite qualification in the case when the discrepancy principle is used as the a posteriori stopping criterion. Moreover, considering the relation between the number of steps and the optimal relaxation parameter \(\omega\) the authors propose a new algorithm of the hybrid Padé iteration method consisting of the Landweber- and the (2,1)-Padé methods. Two numerical examples show effectiveness of the hybrid Padé method comparing with the Landweber, (1,1)-Padé, (2,1)-Padé and the conjugate gradient methods. The paper is a continuation of the authors' previous paper in [ibid. 171, No. 1, 385--397 (2005; Zbl 1087.65029)].