On the conditioning of the Hessian approximation in quasi-Newton methods (Q809555)

The authors consider a quasi-Newton approach for locating the minimizer \(x^*\in R^ n\) of a differentiable function f: \(R^ n\to R\). They construct a nonlinear model for the gradient of f along a chosen ray in the variable-space in a way to minimize the condition number bound of the inverse Hessian approximation. Numerical tests, performed on variations of the Broyden-Fletcher-Goldfarb-Shanno method indicate that the proposed approach is effective on the first iterations only.