A new descent algorithm with curve search rule (Q1764727)

A globally convergent curve search algorithm for solving unconstrained minimization problems is developed. The curves which underly the step direction and step size procedure at each iteration are rational expressions in the curve parameter \(\alpha\). Nominator and denominator depend linearly on \(\alpha\). There exist some similarities with conjugate gradient methods, and Wolfe's line search rules are considered at the step procedures, too. Numerical experiments allow to compare the method proposed with some standard algorithms.