My Community

I have a question regarding Berry phase calculations. I am calculating the Berry phase for a rare-earth based magnetic oxide. The symmetry of the system is triclinic (P1) . I start with a scf calculation on a 5x3x5 k-point grid and then start the Berry phase calculation along all three directions. All the atoms are fully relaxed and the system is insulating. Since I do not have a symmetry direction for the band structure, I calculate the energy eigenvalues at the k-point sampled and the system is definitely insulating.

When I do a Berry phase calculation along IGPAR = 1 and NPPSTR = 10, I get an error message :
Berry-Phase term: -0.37398 electrons
Error in subroutine BERRY: did not find all determinants

Matrix CMK is not an nxn matrix for
ISTR = 2 j = 0.

However, for the same direction if I change NPPSTR to 7, the error disappears and I get a value for the polarization.

My question is:

1. How does one choose the k-points for NPPSTR?
2. Usually one associates the above error with a metallic state; does that mean that when I change the kpoint distribution on the string, I run into a metallic state which was not sampled during my scf run. Is this just a technical error ?


Other details:
vasp version: 4.6.28

my INCAR file:
PREC = high
ENCUT = 400
ISMEAR = 0
SIGMA = 0.005
EDIFF = -1e-6
NELMIN = 5
NELM = 200
AMIX_MAG = 0.2
ISPIN = 2
MAGMOM= 1*0.0 7*0.0 8*5.0 24*0.0
LORBIT = .TRUE.


# relaxation
#EDIFFG = -0.1e-3
#NSW = 100
#IBRION = 2
#POTIM = 0.1
#ISIF = 3

LDAU =.TRUE.
LDAUTYPE = 1
LDAUL = -1 -1 2 -1
LDAUU = 0.0 0.0 6.0 0.0
LDAUJ = 0.0 0.0 0.8 0.0

LBERRY = .TRUE.
IGPAR = 1
DIPOL = 0.0 0.0 0.0
NPPSTR = 10

K-point: 5x3x5

Any suggestions would be appreciated
Regards,

Sahu

usually it is sufficient to set NPPSTR to 6-8. Concerning the eigenvalue spectrum of that run: please check the eigenvalues and occupancies of each k-point as written in the OUTCAR file to see if a band really crosses the Fermi level (gap) it may of course be true that bands cross the gap in a region of the BZ which has not been sampled during the scf-procedure, especially if the gap width is very small, but as this behaviour changes from NPPSTR 7 to 10, I rather think there is some error in this particular run. (please also check the electronic convergence of this step)