English

Discrepancy principle for DSM

Functional Analysis 2007-05-23 v1 Numerical Analysis

Abstract

Let Ay=fAy=f, AA is a linear operator in a Hilbert space HH, yN(A):={u:Au=0}y\perp N(A):=\{u:Au=0\}, R(A):={h:h=Au,uD(A)}R(A):=\{h:h=Au,u\in D(A)\} is not closed, fδfδ\|f_\delta-f\|\leq\delta. Given fδf_\delta, one wants to construct uδu_\delta such that limδ0uδy=0\lim_{\delta\to 0}\|u_\delta-y\|=0. A version of the DSM (dynamical systems method) for finding uδu_\delta consists of solving the problem \bee \dotu_\delta(t)=-u_\delta(t)+T^{-1}_{a(t)} A^\ast f_\delta, \quad u(0)=u_0, \eqno{(\ast)}\eee where T:=AAT:=A^\ast A, Ta:=T+aIT_a:=T+aI, and a=a(t)>0a=a(t)>0, a(t)0a(t)\searrow 0 as tt\to\infty is suitably chosen. It is proved that uδ:=uδ(tδ)u_\delta:=u_\delta(t_\delta) has the property limδ0uδy=0\lim_{\delta\to 0}\|u_\delta-y\|=0. Here the stopping time tδt_\delta is defined by the discrepancy principle: \bee \eqno{(\ast\ast)}\eee c(1,2)c\in(1,2) is a constant. Equation ()(\ast) defines tδt_\delta uniquely and limδ0tδ=\lim_{\delta\to 0}t_\delta=\infty. Another version of the discrepancy principle is also proved in this paper.

Keywords

Cite

@article{arxiv.math/0603632,
  title  = {Discrepancy principle for DSM},
  author = {A. G. Ramm},
  journal= {arXiv preprint arXiv:math/0603632},
  year   = {2007}
}