Decomposition methods for large linear discrete ill-posed problems (Q2508945)

The authors present decomposition methods for large linear discrete ill-posed problems. The main idea is a splitting of the solution space into a Krylov subspace determined by a standard iterative method of GMRES-type or the least squares QR method and a user-supplied subspace. The latter is of small dimension and can be chosen to help represent the pertinent features of the solution. Numerical tests show that the decomposition methods can yield better approximations of the solution than the standard iterative methods.