Multi-step quasi-Newton methods for optimization (Q1334773): Difference between revisions

Quasi-Newton methods update the existing Hessian approximation by means of data derived from the step just completed. The authors propose ``multistep'' methods, employing data from previous iterations, by means of interpolating polynomials. The results of extensive numerical experimentation indicate strongly that computational advantages can occur from such an approach, particularly as the dimension of the problem increases.