English

On the discrepancy principle for the dynamical systems method

Dynamical Systems 2007-05-23 v1

Abstract

Assume that Au=f,(1) Au=f,\quad (1) is a solvable linear equation in a Hilbert space, A<||A||<\infty, and R(A)R(A) is not closed, so problem (1) is ill-posed. Here R(A)R(A) is the range of the linear operator AA. A DSM (dynamical systems method) for solving (1), consists of solving the following Cauchy problem: u˙=u+(B+\ep(t))1Af,u(0)=u0,(2) \dot u= -u +(B+\ep(t))^{-1}A^*f, \quad u(0)=u_0, \quad (2) where B:=AAB:=A^*A, u˙:=dudt\dot u:=\frac {du}{dt}, u0u_0 is arbitrary, and \ep(t)>0\ep(t)>0 is a continuously differentiable function, monotonically decaying to zero as tt\to \infty. A.G.Ramm has proved that, for any u0u_0, problem (2) has a unique solution for all t>0t>0, there exists y:=w():=limtu(t)y:=w(\infty):=\lim_{t\to \infty}u(t), Ay=fAy=f, and yy is the unique minimal-norm solution to (1). If f\df_\d is given, such that ff\d\d||f-f_\d||\leq \d, then u\d(t)u_\d(t) is defined as the solution to (2) with ff replaced by f\df_\d. The stopping time is defined as a number t\dt_\dsuch that lim\d0u\d(t\d)y=0\lim_{\d \to 0}||u_\d (t_\d)-y||=0, and lim\d0t\d=\lim_{\d \to 0}t_\d=\infty. A discrepancy principle is proposed and proved in this paper. This principle yields t\dt_\d as the unique solution to the equation: A(B+\ep(t))1Af\df\d=\d,(3) ||A(B+\ep(t))^{-1}A^*f_\d -f_\d||=\d, \quad (3) where it is assumed that f\d>\d||f_\d||>\d and f\dN(A)f_\d\perp N(A^*). For nonlinear monotone AA a discrepancy principle is formulated and justified.

Keywords

Cite

@article{arxiv.math/0302001,
  title  = {On the discrepancy principle for the dynamical systems method},
  author = {A. G. Ramm},
  journal= {arXiv preprint arXiv:math/0302001},
  year   = {2007}
}