A New Type of Step Sizes for Unconstrained Optimization

Abstract Step sizes is very important for a global convergence gradient method solving the problems of unconstrained optimization. The new step formulas techniques proposed, key idea used in construction algorithm to approximate Hessian by suitable diagonal matrix, which has been found be most efficient this paper. Under weaker conditions, we define convergences proposed methods. In addition, will show that performance better than descent (GD) method.