Convergence results of Landweber iterations for linear systems (Q403453): Difference between revisions

The paper deals with iterative solving of inconsistent linear systems \[ Ax=b, \] where \(A\) is an \((M\times N)\)-matrix, \(b\) is an \(M\)-vector, and \(x\) is an \(N\)-vector. This problem in the context of image reconstructions is solved by the weighted Landweber iterative scheme \[ x^{(n+1)}=x^{(n)}+\lambda_n V^{-1}A^\ast W(b-Ax^{(n)}), \] where \(\lambda_n\) is the relaxation parameter, \(x^{(0)}\) is an initial guess, \(A^\ast\) is the conjugate transpose of \(A\), \(V\) and \(W\) are two positive definite weight matrices of order \(N\) and \(M\), respectively, which are symmetric or Hermitian symmetric. The Landweber scheme finds one least squares solution minimizing the functional \[ L(x)=\frac{1}{2}\|b-Ax\|^2. \] The Landweber scheme is a preconditioned gradient based method for minimizing \(L\). The paper is well-written and presents relatively simple convergence analysis. Unfortunately, numerical experiments are not presented although the authors announce an application from an image reconstruction.