Kybernetika 32 no. 2, 121-138, 1996

Combined trust region methods for nonlinear least squares

Ladislav Lukšan


Trust region realizations of the Gauss-Newton method are commonly used for obtaining solution of nonlinear least squares problems. We propose three efficient algorithms which improve standard trust region techniques: multiple dog-leg strategy for dense problems and two combined conjugate gradient Lanczos strategies for sparse problems. Efficiency of these methods is demonstrated by extensive numerical experiments.


90C20, 90C30, 65K05