Abstract
This paper discusses accuracy of estimated gradient and Hessian components in approximate optimization. Several flexible approximate optimization methods can estimate gradient and Hessian components of an estimated surface without using finite difference, and it will be available in approximate optimization process using a gradient-based optimization method. Since its accuracy affects an estimated result of optimum solution or convexity, accuracy of estimations by those flexible methods should be investigated. In this study, accuracy of gradient and Hessian estimation by Kriging method is investigated. At first, in this paper, a formulation for gradient and Hessian estimations by Kriging method is described. Next, accuracy of those estimations is discussed with numerical results.
