摘要

We extended the Davidson method, which was used to solve the standard eigenvalue problem, to solve the generalized eigenvalue problem and proposed the corresponding block iterative algorithm. Through theoretical analysis and numerical calculation, we found that the block iterative algorithm was doomed to converge after finite iterations if the process of iteration was not divergent. If the dimension of the matrix is n, the number of the eigenvalues and corresponding eigenvectors to be calculated is k, the size of the initial subspace is r(r >= k), the number of iteration is m, then they will fit in with the equation n=r+km. The positive integer root m could be obtained by regulating the size of the subspace.