Regularization of discrete ill-posed problems (Q1826453)

Discrete approximations \( A_n u_n = f_n \) of an ill-posed equation (1) \( Au = f \) with a linear compact operator \( A: X \to X \) in a Hilbert space \( X \) are considered. Here, \( A_n: X_n \to X_n \) is a linear bounded operator in a finite-dimensional Hilbert space \( X_n \), where \( \{X_n,r_n,p_n\} \) is a convergent and stable discrete approximation of \( X \) with restriction operators \( r_n: X \to X_n \) and prolongation operators \( p_n: X_n \to X \). In addition, \( f_n \in X_n \) is an approximation to \( f \in X \). For Tikhonov regularization (2) \( (A_n^* A_n + \alpha I) u_{n,\alpha}^\delta = A_n^* f_n^\delta \) with perturbed data \( f_n^\delta \in X_n \) satisfying \( \| f_{n}^\delta - f_n \| \leq \delta \), an a priori parameter choice \( \alpha=\alpha(\delta) \) in terms of the singular values of \( A_n \) is considered which minimizes the relative sensitivity \( \| u_{n,\alpha}^\delta - u_{n,\alpha} \|/\| u_{n,\alpha} \| \). Here \( \| \cdot \| \) denotes the norm in \( X_n \), and \( u_{n,\alpha} \in X_n \) is obtained by Tikhonov regularization (2) with \( f_n^\delta \) replaced by \( f_n \). For this parameter choice and certain choices of the discretization level \( n = n(\delta) \) and under some additional conditions, it is shown that \( \| u^+ - p_{n(\delta)} u_{n(\delta),\alpha(\delta)}^\delta \| \to 0 \) as \( \delta \to 0 \) holds. Here \( u^+ \in X \) denotes the minimum norm solution of equation (1) which is supposed to exist, and \( \| \cdot \| \) denotes the norm in \( X \). Convergence rates are established for the special case that the discrete approximations are obtained by a truncated singular value decomposition. Finally numerical results are presented.