next up previous

2.4.3 Conjugate Gradient     continued...

Minimization performance of CG is shown in Figures 10 and 14 for Rosenbrock's and Beale's functions, respectively. Note that performance for both functions with CG is better than SD, as expected, but the paths are characteristically more `random'. For Rosenbrock's function, for example, a large step is taken between the fourth and fifth iterates, where a distant minimum was detected in the line search. The step length varies considerably from one iteration to the next. Similarly, for Beale's function, the step lengths vary from to 800.