Lavrentiev-type regularization methods for Hermitian problems (Q2355299)

The fractional Lavrentiev regularization method, associated to a noisy Hermitian system \(Ax=b\), reads as \((A^{1+\alpha} + \mu I) x = A^{\alpha} P_{R(A)}(b)\). The authors propose to replace the regularization matrix \(\mu I\) by a more general one \(M_{\mu, \alpha}\), such that \(|| M_{\mu, \alpha} ||_F < || \mu I ||_F\) and \(A^{1+\alpha} + M_{\mu, \alpha} \) has no eigenvalue smaller than \(\mu\). Numerical experiments are also presented.