Top Page > Browsing

lapack subroutines for diagonization Date: 2014/08/29 05:31 Name: John Chan Dear Prof. Dr. Ozaki, I have a question on using lapack subroutines for diagonizing Hamiltonian matrix based on nonorthogonal LCAO. I am trying to interpolate bands based on the elements of Hamiltonian and overlap stored in HS.out. I have used zhegv for the diagonization, but it failed that the diagonization gives me zero for all eigenvalues. Is zhegv suitable for such a business? Having a look over the subroutiens of openmx, I noted that it uses different lapack subroutines. For instance, if I understand correctly, it uses dsyevx in the subroutine Eigen_lapack.c after some transfermations. What is the standard way for soluting linear equations AX = lambda BX? And what would be you suggestion for my business? Thank you very much. Best regards, John

Page: [1]

Re: lapack subroutines for diagonization ( No.1 ) Date: 2014/08/29 18:36 Name: Artem  <artem.pulkin@epfl.ch> Hi John, I would suggest you first googling "generalized eigenvalue problem". It can be done in a single line of code if you use Matlab or python numpy for example. I would strongly suggest you starting with that. Then, the way I solved it in my home-made OpenMX code is following wrapper template: F77_NAME(zggev,ZGGEV)("N","V",&n,leftHandSide,&n,rightHandSide,&n,lambdas_a,lambdas_b,NULL,&n,states,&n,work2,&n16,work,&info); I.e. the function you are looking for is ZGGEV. Artem

Page: [1]