An Improved Computational Approach for Multilevel Optimum Design∗

Abstract

Recently there has been a growing interest in methods that decompose complex optimization problems and employ a multilevel or hierarchical approach. One of the most irksome problems with the hierarchical approach is the discontinuous behavior of derivatives that are transferred from the lower levels of the hierarchy to the upper levels. This paper proposes a hierarchical algorithm that is free of such difficulties. A penalty function method is employed, in combination with Newton's method with approximate second derivatives, to perform the optimization. The penalty function formulation is shown to be natural for multilevel problems. A simple three-bar truss and a simple frame problem are used for demonstration