A sparse matrix finite element technique for iterative structural optimization

Abstract A sparse matrix symbolic factorization technique for iterative structural optimization is developed for repeatedly solving matrix finite element equations of the form Kz = f and eigenvalue problems of the form Ky = λMy where K and M are symmetric, positive definite, sparse matrices. The objective is to develop an iterative technique for arbitrary sparse matrices, so that there is no need for selecting the best node numbering sequence for minimizing the bandwidth of the global stiffness matrix. A preprocessing step is used to define pointer arrays that will be used in each structural reanalysis to construct the global stiffness matrix from element stiffness matrices, reorder the array containing the nonzero entries in the lower triangular part of the global stiffness matrix, and numerically factor a symmetric permutation of that matrix. Efficient methods of finding solutions of matrix finite element equations and eigenvalue problems for iterative structural optimization are shown to be feasible, utilizing sparse matrix symbolic factorization techniques and the subspace iteration method. The method is tested in plate deflection and column buckling problems and applied to several shape optimization problems.