Pointwise rational approximation and iterative methods for ill-posed problems (Q1115112): Difference between revisions

Let T be a selfadjoint positive semi-definite linear contraction on a Hilbert space, \(y\in R(T)\). The following class of iterative methods for generating approximate solutions \(x_ n\) to \(Tx=y\) is considered: Let \(r_ n\) be continuous, uniformly bounded rational functions on [0,1] with \(r_ n(0)=1\) and \(r_ n(\lambda)\to 0\) for all \(\lambda\in [0,1]\), \(s_ n\) be defined by \(r_ n(\lambda)=1+\lambda s_ n(\lambda).\) Then \(x_ n:=x_ 1+s_ n(T)(Tx_ 1-y).\) This class of methods includes explicit iterative method like Landweber iteration, implicit iterative methods like Lardy's method, and also Tikhonov regularization. The author proves results about convergence, convergence rates, and stability. The main part of the paper concerns converse and saturation results. It is shown that in general, Bernstein type theorems do not hold: An example shows that although for \(y\in R(T^{P+1})\), the optimal convergence rate is given by \(o(n^{-P})\), this rate need not imply that \(y\in R(T^{P+1})\). Such a converse result (due to C. Groetsch) holds for Tikhonov regularization with \(P=1\), however; this converse result is (slightly) extended in this paper.