Iterative solution of linear equations with unbounded operators (Q879067)

The problem under consideration is to solve a solvable operator equation \(Au = f\) with linear, closed, densely defined, unbounded, and not necessarily boundedly invertible operator \(A\) in a Hilbert space. In view of possible non-uniqueness the desired solution is the normal solution (having minimal norm) \(y\). For stable approximation of \(y\) under given noisy data \(f_{\delta}\), \(\| f-f_\delta\| \leq\delta\), author suggests and justifies the iterative process \[ u_{n+1}= aT^{-1}_{a}u_{n}+T^{-1}_{a} A^*f_{\delta}, \quad u_1 \bot N(A), \;a>0, \] where \(T_a=A^*A+aI\). The stopping rule is \(n=n(\delta)\) such that \(n(\delta)\rightarrow\infty\) and \(\delta n(\delta)\rightarrow 0\). The case of self-adjoint \(A^*=A\) is briefly considered. The second method for solving the considering problem is the dynamical systems method: \[ \frac{du}{dt}=-u+T^{-1}_{\varepsilon(t)}A^*f_{\delta}, \quad u(0)=u_0, \] with stopping time \(t_{\delta}\) determined by a discrepancy principle. Numerical inversion of the Hilbert matrices by the presented methods is demonstrated.