Application of Shift Algorithm in Accelerating Subspace Iteration Method

In this paper, a generalized eigensolver named SSubspace was developed based on subspace iteration method with an efficient, stable and low memory consuming shift algorithm. The detailed steps were listed and several important control parameters were discussed. In addition, it was employed to the generalized eigenvalue problem of singular stiffness matrix and the rigid body mode was solved. Compared with Intel MKL extended eigensolver (FEAST v2.1), the efficiency of SSubspace is higher and memory consumption is less. Theoretically, SSubspace can solve all eigenvalues with the time approximately linear to the number of eigenvalues and is of significance to solve high order eigenvalue and all eigenvalues of large matrix.