We have developed a numerical algorithm of computing the eigenvalue distribution of non-Hermitian matrices with the memory size of O(N), where N is the dimension of the matrix. The algorithm basically computes the norm of the Green's function from its largest singular value, which involves (i) the matrix inversion of non-Hermitian matrices by the biconjugate gradient method and (ii) the calculation of the largest eigenvalue of a Hermitized matrix by the Lanczos method. |
|
|