Feasible Newton-type methods for large scale problems

A class of Newton-like methods is introduced which is feasible for large scale minimization problems. The methods are essentially an imbedding of Gauss-Newton or Newton-type nonlinear algorithms into any of several well known linear iterative methods. The resulting composite algorithms are simple to implement, converge quickly, and require only reasonable amounts of computer storage. Numerical comparisons indicate that for optimal control problems such an approach can be superior to the conjugate direction methods in Rn.