forum.alglib.net

Does this procedure work with a degenerate spectrum?

It may be problematic. Nonsymmetric EVD subroutines often have convergence problems on such matrices.

Do you have any specific matrix to try? Can you describe your task in more details?

While I use only a symmetric matrix. I use RMatrixEVD for the sake of versatility. RMatrixEVD works well with degenerate symmetric matrices. I want to get official confirmation.

My matrix is a real matrix of Hermitian operators.

> "degenerate symmetric matrices".

For the purposes of EVD, symmetric matrices are perfect. They have orthogonal eigenvectors, and it eliminates at least one possible failure point for nonsymmetric EVD - inverse iteration part which finds eigenvectors. So, RMatrixEVD performs very well on these matrices.

However, I recommend you to use specialized symmetric EVD functions because they perform much, much faster. And they are more precise at handling clustered eigenvalues (multiple eigenvectors having same eigenvalue).

Good. Thank you.