BSE Calculation: dielectric matrix contains only zeros and ones

[shuo_tao]

Hi All,

I'm having trouble with BSE calculation to find the imaginary part of the dielectric of Silicon-like bulk crystal system.

I followed the guide on this tutorial, the first 3 steps are fine but final steps gives me only ones and zeros in dielectric matrix in vasprun.xml.

Here is my INCAR:

PREC = Normal ; ENCUT = 500.0
ALGO = BSE
ANTIRES = 0
ISMEAR = 0 ; SIGMA = 0.01
ENCUTGW = 350
EDIFF = 1.E-8
NBANDS = 160
NBANDSO = 4
NBANDSV = 8
LOPTICS = .TRUE. ; LPEAD = .TRUE.
OMEGAMAX = 40
PRECFOCK = Normal

I've tried to increase NBANDSO and NBANDSV, also tried to change PRECFOCK to Accurate, but the issue remains the same. Also I noticed in the tutorial, there is no LOPTICS = .TRUE. ; LPEAD = .TRUE. line in step5, but I added it and copied all the *.tmp files from GW0 step as well as the WAVECAR and WAVEDER from DIAG step.


Thanks in advance,
Shuo

[alexey.tal]

Dear shuo_tao,

Could you please provide the input and output files as described in the guidelines.

Also, you should not add LOPTICS = .TRUE. and LPEAD = .TRUE., if the BSE algorithm is selected.

[shuo_tao]

Hi, here are the input and output files.

[alexey.tal]

Thank you for providing the input files. It would be also helpful to see the stdout file.

[shuo_tao]

For stdout file do you mean the VASP log or slurm job output?

[alexey.tal]

Thank you. I meant the log that VASP prints into the standard output, in your script it is written into a log file.

The warning in your log file indicates that the band gap is very small. What band gap did you find in the GW calculation? Could you please provide the GW input and output files?

[shuo_tao]

Sure

[alexey.tal]

Super!
Here are some things that look odd to me:
- The RPA dielectric constant in your GW calculation is 1, which clearly means that something is wrong in this calculation.
- NBANDSGW is too small (see wiki). There is 48 electrons or 24 occupied states in your system, while only NBANDSGW states updated in your GW calculation. I would recommend using the default value NBANDSGW in this case.

What was the gap in the DFT calculation? Could you please provide the input and output files?

[shuo_tao]

0.0557 eV direct gap for DFT band, 0.0163 eV indirect gap for GW0 band.

[alexey.tal]

There are partially occupied states in the OUTCAR file, which means that the system is metallic. It might be because the band gap is small and the SIGMA value is too large. Also, I assume this is the DFT calculation you used for plotting the bandstructure. However, I meant that it would be useful to see how you set up the ground state calculation for the GW step, it would correspond to Step 2 of the tutorial. Since the RPA dielectric constant is 1, I would assume that something is wrong with the DFT calculation.

[shuo_tao]

Ok, right now I don't have access to that directory, since our cluster is down for maintenance.

But I basically use the same INCAR from tutorial except for different KPAR, ENCUT and NBANDS

System = SiGe
PREC = Normal ; ENCUT = 500.0
ALGO = EXACT ; NELM = 1
ISMEAR = 0 ; SIGMA = 0.01
KPAR = 4
NBANDS = 160
LOPTICS = .TRUE. ; LPEAD = .TRUE.
OMEGAMAX = 40

[alexey.tal]

The INCAR from the tutorial is fine. However, the NBANDS is probably too small for the GW calculation with 12 atoms in the cell. It is hard to identify the problem without seeing all the files, so I would appreciate if you could share these files when the access is restored.
Perhaps you should try to reproduce the results from the tutorial first and then try to run the calculation for your system.

[shuo_tao]

Hi,

I redid the whole calculation and increase NBANDS to 768 and NBANDSGW to 48, however the issue remains the same.

I put the SC and DIAG calculations in the attachment.

[alexey.tal]

The number of k-points in SC and DIAG is different because in the SC calculation the symmetry is switched off (ISYM=0). This mismatch is probably the reason why the dielectric function is not calculated correctly. Why do you need to switch the symmetry off?