Parallel computation for the nonsymmetric generalized eigenvalue problem is one of the fundamental problems in large scale engineering computation.A homotopy continuation algorithm is proposed to deal with parallel computation for this eigenvalue problem.Numerical results show that it has advantages of high parallel efficiency and fast convergence speed.

Parallel processing in the famous method of increment for nonlinear finite element structural analysis is given.Some numerical examples are tested on multiprocessor Challenge-L.

By using multiprocessor systems equiped with local memory and shared memory as the hardware enviroment of the application, the implementation of EBE-PCG method is discussed. with especal attention on the problems relating to configuration of tasks,storation and visiting of datas,communication and synchronization. Finally,a structural analysis problem is solved by use of EBE-PCG method.The results show that the algorithm is effective numerical method for the super-large scale structural analysis problems.

Important concepts such as element vector, pseudo-element valor are discussed and an EBE cpmputational method of preconditioned conjugate gradient method (PCG) is presented, namely EBE-PCG method, which does not require the formation of global stiffness matrix and is highly parallizable. Numerical example shows it is very efficient.

An efficient parallel direct integration algorithm is presented. Computational results for a model in YH-1 computer show that for the model of degree of freedom of 1666, the speed of the parallel algorithm is 30 times faster than of traditional serial algorithm, and if the features of YH-1 computer considered and the assemble language used, it is 50 times faster in the same model case.

In this paper, based on the work of L.J.Hayes in 1986, a vectorized algorithm of the global stiffness matrix-vector multiply in finite element structural analysis is developed when we in fact don't form the global stiffness matrix. A numerical example by using the method in iterative solution of finite element equations verify it is very efficient.

This paper presents simiimplicit Runge-kutta type parallel direct integration methods to solve structural dynamic problems using 3-stage-3-order semi-implicit Runge-kutta method and using polynomial preconditioned conjugate gradient method for solving relevant systems of linear algebraic equations. Compared with relevant serical algorithm RK33S on YH-1 computer, parallel algorithm RK33P's speed up is 24~27 when linear systems of equations involved is of order 10³~10⁴.