Two-level preconditioners for regularized inverse problems. I: Theory (Q1805843)

For the operator equation \({\mathcal A}u= b\) with \({\mathcal A}={\mathcal K}^*{\mathcal K}+\alpha{\mathcal L}\) the two Schwarz preconditioners (additive and multiplicative) are compared, where \({\mathcal K}\) is a compact operator in a Hilbert space and \({\mathcal L}\) is positive definite with closed range and \(\alpha>0\). The operators \({\mathcal K}^*{\mathcal K}\) and \({\mathcal L}\) have drastically different spectral properties, inherited by their discretizations what in turn accounts for the ineffectiveness of multigrid methods. In this paper the equation \({\mathcal A}u= b\) is decomposed using a \(2\times 2\) block matrix system and from this system two preconditioned matrices are defined: Jacobi-like additive and Gauss-Seidel-like multiplicative preconditioned Schwarz ones. It is shown that the additive Schwarz preconditioner significantly increases the condition number whereas the multiplicative one improves conditioning.