求解大型稀疏矩阵广义特征值的子空间迭代法加速研究
Accelerated Subspace Iteration Method of Solving Generalized Eigenvalue Problem on Large Sparse Matrix
-
摘要: 采用Wilson移频策略对子空间迭代法进行了加速. 为加速高阶特征值的收敛,对Wilson移频策略进行了改进,给出了详细的移频子空间迭代求解特征值的步骤,讨论了若干移频控制参数的选取. 从给出的对比算例可看出,采用移频算法,子空间迭代法求解特征值明显加速,且随着待求特征值阶数的增加,加速效果更加明显,求解时间与待求特征值数近似成线性关系.Abstract: In this paper, Wilson shift algorithm was employed to accelerate subspace iteration method, the detailed steps were listed and several control parameters were discussed. The results show that the subspace iteration method is accelerated obviously especially with the eigenvalue order increasing. The time need to solve the generalized eigenvalue problem on large sparse matrix is approximately linearly proportional to the number of eigenvalues.
下载: