The global convergence of self-scaling BFGS algorithm with non-monotone line search for unconstrained nonconvex optimization problems (Q2644350)

The Broyden-Fletcher-Goldfarb-Shanno (BFGS) method for continuous multivariate optimisation is a quasi-Newton method using an estimate of the inverse Hessian which is updated each iteration. Self-scaling versions modify the inverse Hessian estimate's update rule to avoid very large eigenvalues. The algorithm proposed here chooses search directions according to a self-scaling BFGS method, and step lengths along those directions according to a nonmonotone line search technique. Global convergence is proved.