This paper describes a fall-back procedure for use with the Gauss-Newton method for nonlinear least-squares problems. While the basic Gauss-Newton algorithm is often successful, it is well-known that it can sometimes generate poor search directions and exhibit slow convergence. For dealing with such situations we suggest a new two-dimensional search strategy. Numerical experiments indicate that the proposed technique can be effective.
|Advanced Modeling and Optimization
|Published - 2009