[Pw_forum] 'missing eigenvalue'

Stefano de Gironcoli degironc at sissa.it
Thu Jun 5 12:43:04 CEST 2003

Are you sure that the HOMO is non degenerate ?
If the HOMO orbital is degenerate but the number of electrons
in the cluster does not fill it completely this may induce instability
in the case you provide more bands than electrons because from one
iteration to the next the occupied bands could be different.
Another possible problem, related to the fact that the HOMO orbital is
missing, could be related to the fact that the code performs iterative
diagonalization and if the HOMO has different  symmetry from the
current approximate wavefunctions  the diagonalization will not find it.

possible actions:
1) try to start from random wavefunctions instead of atomic ones. This could
help if the missing eigenvalue is due to symmetry of the initial guess for
the wfcs.
2) try to do calculations with a small smearing using gaussian or 
broadening (avoid Methfessel-Paxton in this case since the fact that it is
not positive definite could lead to spurious effects when using very few 
This should clarify what is the level structure for your  cluster and should
tell you if there are problem of degeneracy at the Fermi level.
3) Usually increasing the number of bands should helps in NOT missing 
the fact that you get problems is an hint that there might be problem of 
Write eigenvalues at every iteration (including more bands than Nelec/2) 
to see
how things evolves and see if there are levels that keep changing.
All the best

Stefano de Gironcoli

Raghani Pushpa wrote:

>Dear all,
>I am doing the calculations on clusters. When I do the calculations at 
>zero temperature (no smearing), I have to  restrict the number of bands to 
>the number of filled bands. If I give more number of bands, the program 
>does not converge; I don't understand why is it happening? 
>At degauss=0, the eigenvalues are not same as eigenvalues I get from band 
>structure calculations, I find that in self consistent calculation, the 
>eigenvalue of HOMO is missing and program gives the eigenvalue of LUMO. 
>Could you please tell me why does it happen?

More information about the users mailing list