Use of preconditioned Krylov subspaces in conjugate gradient-type methods for solving a nonlinear least squares problem (Q2563388)

The problem of nonlinear least squares is solved by a family of globally convergent methods. At each iteration linear search is done in the direction of the Krylov subspace corresponding to the current residual and the preconditioned from the right Jacobian. The search direction can be interpreted as a solution to the linearized system. The convergence conditions are some rather weak conditions on the search directions and the smoothness of the Jacobian.