Some Improvements in the Gradient Search Method of Optimization

摘要

The purpose of this paper is to outline a basic code which employs the gradient search method. Further purpose lies in the presentation of ideas and developments which have been found useful in a variety of optimization problems. In studying the area of optimization, it was realized that knowledge of the response surface would allow the use of various techniques specifically designed to handle definite types of surfaces. Therefore, it was found useful to construct a history of the surface during the gradient search. This history, and knowledge of the contour, could then be employed to determine what action would be most useful in arriving at the optimum. As the name of the method implies, a gradient, which is a vector in the direction of steepest ascent, is calculated; the function is then evaluated at points along this N-dimensional gradient vector. (Author)