HYBRID ALGORITHMS WITH AUTOMATIC SWITCHING FOR SOLVING NONLINEAR EQUATION SYSTEMS

An optimal switching device is introduced for hybrid algorithms which try to combine the superlinear convergence speeds of Newton model solu tion techniques with the cheapness of slower Gauss-Seidel techniques, We also propose a device for combining second-order and Gauss-Seidel i terative techniques. Tests show that these hybrids lead to powerful im provements in solving nonlinear models, both in terms of accelerating convergence and reducing computational costs and in extending the radi us of convergence in difficult cases.