A Two-dimensional Search used with a Non-Linear least Squares Solver

This note describes a modified search strategy for use with a Gauss-Newton method for nonlinear least-squares problems. If a standard line search along the Gauss-Newton vector p is unable to make much progress, a new search direction is constructed which lies in the plane of p and the steepest-descent vector. Numerical experiments show that a quadratic model of the objective function in this plane can yield effective corrections when the basic Gauss-Newton technique experiences difficulty.