A globalization scheme for the generalized Gauss-Newton method (Q582820)

The author considers the problem of minimization of the Euclidean norm of a vector-valued nonlinear function F(x) subject to nonlinear equality constraints \(G(x)=0:\) \[ \min \{\| F(x)\|^ 2:\quad G(x)=0\}. \] A globalization scheme for the generalized Gauss-Newton method with damping is described. A penalty function is used to select the damping factor for the line search determining the new iterate. This penalty function is \(\Psi (x,\mu)=\| F(x)\| +\mu \| G(x)\|,\quad \mu >0.\) The global convergence of the algorithm is proved under `natural' conditions and it is further shown that, if the method converges to a solution with a small residual, the algorithm changes to the undamped generalized Gauss-Newton method. The linearized problem arising from a step of the Gauss-Newton method is solved using algorithm LSE described by \textit{C. P. Lawson} and \textit{R. J. Hanson} [Solving least squares problems (1974; MR 51, 2270)]. Numerical results are presented for 5 small problems.