This is helpful when using AMG, where the matrix hierarchy
is computed while coarsening. Another application are
problems, where the coarse matrix is a preconditioner
that does not coincide with the Galerkin-restriction
of the fine matrix, e.g. for DG methods.