Fast Algorithms for Structural Analysis, Least Squares and Related Computations

摘要

New fast algorithms for high speed computation on the modern generation of supercomputers is essential. To meet these challenges new techniques are developed in numerical linear algebra and its applications for implementation on these new architectures. Significantly, applications of this work to practical problems of structural analysis and design and to least squares adjustments, estimation and digital filtering are also being investigated. The current objectives in structural analysis are to develop efficient and stable high speed algorithms for the design and analysis of large complex systems. Interest here is in developing stable alternatives to the often ill conditioned stiffness matrix approach to solving problems in elastic analysis and structural dynamics. For example, a comparative study is developed of the performances of seven alternative methods to the stiffness approach on the Alliant FX/8 and Cray X-MP systems. These methods involve various orthogonal factorization approaches as well as preconditioned conjugate gradient methods which completely avoid formation of the stiffness equations.