by the conjugate residual method R. Plato.pdf

Fields Institute CommunicationsVolume 00, 0000 The solution of linear semidenite ill-posed problemsby the conjugate residual methodR. PlatoFachbereich Mathematik, Technische Universitat Berlin,Strae des 17. Juni 135, D - 10623 Berlin, GermanyAbstract. In Hilbert spaces we consider linear ill-posed problems withcorresponding operators that are symmetric, positive semidenite andcompact, and we review some results on the conjugate residual methodwhich by denition minimizes the residual over Krylov subspaces andwhich can be realized by a conjugate gradient-type algorithm. In parti-cular, for a nite-dimensional setting of the conjugate residual methodan a posteriori stopping criterion is considered and estimates for theassociated approximations are provided which are order-optimal withrespect to perturbations of the right-hand side and with respect to thediscretization error. 1 IntroductionIn the sequel we consider equations of the formAu = f; (1.1)where A 2 L(H) is compact; A = A  0; f 2 R(A); (1.2)here H denotes a real Hilbert space with inner product h
;
i : H  H ! R andcorresponding norm jj
jj : H ! R, and L(H) denotes the space of operators T : H !H that are bounded and linear. Moreover, A 2 L(H) in (1.2) denotes the adjointoperator of A, and A  0 by denition means that A is positive semidenite, i.e.,hAu; u i  0 holds for each u 2 H. Finally, R(A) = Av : v 2 H denotes therange of A that in general may be innite-dimensional which means that equation(1.1) is ill-posed. In the sequel we assume that only an approximation f  for f isavailable, that is, f 2 H; f f   ; (1.3)where   0 is a small and known error bound.For solving those problems, in this review we consider the conjugate residualmethod which by denition minimizes the residual over Krylov subspaces. However,1991 Mathematics Subject Classication. Primary 65J20, 65R30; Secondary 45E99, 65N38.The author gratefully acknowledges travel support by the Deutsche Forschungsgemein