基于MPP编程环境的一种并行子空间迭代法

子空间迭代法是科学与工程计算中求解广义特征值问题的有效方法 ,针对向量机和共享内存的多处理机 ,前人已成功地作了并行处理。文中给出了适合 MPP大规模并行计算机的并行子空间迭代法。该算法将广义特征值问题转换为一般特征值问题 ,其计算工作量主要体现在矩阵乘法 ,通过对该方法作并行处理 ,使矩阵求逆及一部分乘法运算转换为各结点机上三角形方程组的并行求解。在大规模并行计算机 PA R95上结合 J8- II机翼的动力特性问题对该算法作了数值试验 ,结果说明所给算法是非常有效的

A Parallel Subspace Method Based on MPP Parallel Computer

Subspace method is an efficient tool for generalized eigenvalue problem in scientific and engineering computing. The former researchers have successfully achieved parallel algorithms of subspace method under the environments of vector computer and shared memory parallel computer. A parallel subspace method suitable to MPP parallel computer is presented. This algorithm converts the generalized eigenvalue problem to standard eigenvalue problem,its computational cost is mainly in matrix matrix operations. The parallel algorithm converts the calculation of the inverse and some multiplication of matrices to the solution of triangular systems in each node computer. With the dynamic analysis of J8 II plane wing, the numerical experiments in PAR95 show that the algorithm is very effective.