Page 1 of 1
ERROR in subspace rotation PDSYEVX/ PZHEEVX: not enough eigenvalues
Posted: Sun Sep 10, 2023 11:48 am
by rishikanta_m
Hi,
I am trying to calculate band structure using
MBJ, but I could not solve this error of "subspace rotation PDSYEVX/ PZHEEVX: not enough eigenvalues found 527 32". This error does not exist when
CMBJ is set to some value, but it occurs the other way around.
Is setting fix values for
AMIX and
BMIX, the only solution to solve the convergence issue? If so, how we can we get these values without hit and trial method?
Re: ERROR in subspace rotation PDSYEVX/ PZHEEVX: not enough eigenvalues
Posted: Mon Sep 11, 2023 7:58 am
by merzuk.kaltak
Dear rishikanta_m
The PDSYEVX/PZHEEVX error tells you that vasp failed to diagonalize the Kohn-Sham Hamiltonian in one of the electronic steps. There are several possible reasons why this can happen. Mostly they are related to incorrect input file settings.
However, I do not see any major issues with your input files. You might want to give vasp more than 8 electronic steps (
NELM=8) to find the solution to the Becke-Johnson potential, though.
In fact, you probably need even more than the default 60 electronic steps, because the MBJ functional is not variational and thus can be hard to "converge", see the N.B.s of
MBJ.
Since you found a solution to the KS equations by setting
CMBJ indicates that a fixed value for the average of |grad n(r)|/n(r) is more stable than allowing this parameter c to be determined in each electronic step by the given density. A regular k-point mesh might be helpful, if you insist on determining c in each step.
Also, the
CMBJ, setting can be fine-tuned with the following three parameters
Setting them in the INCAR file might help finding a proper solution as well. Some choices for these parameters are mentioned in the MBJ section of the
METAGGA page.
Re: ERROR in subspace rotation PDSYEVX/ PZHEEVX: not enough eigenvalues
Posted: Fri Sep 15, 2023 2:30 pm
by fabien_tran1
Hi,
Looking into more detail at the error, the problem originates from the Cs atom. The contribution to the average of grad(rho)/rho in the unit cell coming from the Cs atom has nonsense values so that the calculated value of CMBJ is also nonsense. Problems with the Cs atom were already reported in the past, however a recently generated POTCAR file for Cs seems to solve these various problems. Please, use this POTCAR file for Cs (it is attached). Thanks for having reported this problem.