Quasi-Newton methods based on dispersed methods of recovery of the Hessian (Q2773637)

The author studies dispersed algorithms of recovery of the Hessian of a function under minimization in quasi-Newton methods which are stable with respect to linear dependence of descent vectors. He proposes a new quasi-Newton method and, without assumption that the one-dimensional descent is exact, he proves that the method proposed converges after a finite number of steps for quadratic functions. Results of numerical experiments are given that show advantages of the method proposed.