Multi-Step Skipping Methods for Unconstrained Optimization

When dealing with unconstrained non-linear optimization problems using quasi-Newton methods, updating the approximation to the inverse Hessian is a computationally expensive operation and, therefore, in this paper we investigate the possibility of skipping updates on every second step. The experimental results show that the new methods (i.e. with skipping) give better performance in general than existing multi-step methods, particularly as the dimension of the test problem increases.