Subroutine for eigenvector calculation

Hi Prof Julian,

Could you please let me know as to :

(1) Which subroutine of LAPACK package is GULP using in order to calculate the eigenvectors and eigenvalues ?

(2) As I am not familiar enough with fortran so for the same subroutine (having the same dynamical matrix ) can I use Python(scipy) in order to calculate exactly the same set of eigenvectors as we are getting from GULP ?
I believe I need to call some specific subroutine inorder to calculate the eigenvectors for my Hermitian dynamical matrix.

(3) If not python then could you suggest me other ways by which I could get the same set of eigenvectors?

Thank you in advance.

(1) Multiple routines are used for eigenvector/eigenvalue calculation depending on the algorithm and whether you’re dealing with gamma or not. The standard routines are dsyevd or zheevd in serial.
(2) I’m sure that scipy will provide eigensolver routines, but you’ll have to ask a Python person for more details.
(3) There are plenty of other maths libraries available to compute eigenvectors - just go to

1 Like

Thank you Prof. Julian for your reply.

I am dealing with a reciprocal space point other than Gamma so the dynamical matrix is a Hermitian matrix. In order to get the same set of eigen values and eigenvectors I did the following (just for testing):
I took the same dynamical matrix given by GULP and took it as an input for my own program which produces eigenvalue and eigenvector using the same “zheevd”(used the same parameters as used in GULP) subroutine as suggested by you earlier. The eigenvalues I am getting is the same but the eigenvetors are coming out to be different.
My aim is to get the same set of eigenvectors as produced by GULP. I skimmed through the “pdiagl.F90” subroutine which calls “zheevd” subroutine which I hope is the one which is being used as one of the subroutines to get the eigenvales and eigenvectors by GULP.
Am I missing something which GULP does after getting the eigenvectors from the “zheevd” subroutine and may be that’s why my eigenvectors are not matching?

You may want to look through the other posts on this forum since it’s a pretty common question. One possibility is if you have degeneracy then the eigenvectors can mix freely and so the answer you get depends arbitrarily on the numerics of the computer you are using. A linear combination is still a valid set of eigenvectors with the same eigenvalue, so no point trying to match, though you can always mix your eigenvectors to get the other combination.

1 Like

Thank you Prof. Julian for your valuable suggestions.