BSE Calculation : Dielectric matrix contains only zeros and small bandgap

[asaaravindh2019]

Dear VASP Users,We are computing BSE optical response of TiO2 with HSE06 as the starting points for GW calculations. Typical BSE calculation set up we have used is like the following.

ALGO = BSE, NOMEGA = 50/150, ENCUTGW = 150, NBANDS = 64/256/320, NBANDSO = 8 and NBANDSV = 16, NBSEEIG = 20
WAVEDER is computed with HSE06 prescription switched on in the calculaton with ALGO=EXACT, LOPTICS = .TRUE., LPEAD = .TRUE. In a given set,
the same NBANDS and KPOINTS have been used in the entire set of calculations starting from OPTICS run.


For NBANDS = 64 we could complete BSE calculation without running into any troubles. It yielded oscillator strengths and
dielectric functions correctly. However for NBANDS = 256 and 320, with HSE06 as the starting points for GW, the BSE calculation
with the above set up always yields all dielectric function values and oscillator strengths zero. In the standard output it writes a warning:

| You have a very small band gap. This results in slow convergence |
| of the energy with respect to NOMEGA |
| Usually VASP defaults OMEGAMIN to the bandgap which guarantees an |
| optimal grid for OMEGAGRID=140. |
| The gap was found to be OMEGAMIN= 0.0184 |
| Now OMEGAMIN is set to 0.02 instead, which effectively removes some |
| excitations from the spectrum. |

Both GW0 WAVECAR and preceding WAVEDER files were reportedly read successfully during the BSE calculation.
As we inspected the standard output further it reports E-fermi : 3.0956 eV, which is just very close to the HOMO energy of 3.0954 eV calculated in
the preceding GW0 step. The QP gap as obtained in GW0 gap is 4.1 eV.

Since dielectric matrix contains only zeros in the vasprun.xml file, we have recalculated WAVEDER based on the
reccomendation in the wiki page wiki/index.php/BSE_calculations (Under section common issues).
This did not remove the problem either. All our attempts yield the same warning and all the dielectric function and oscillator strengths are zero.

Can someone look into the problem and help us getting it sorted out. We attach BSE INCAR, standard output and for your reference.

Thanks in advance,
Assa

[henrique_miranda]

Did you use the same number of bands to generate the WAVEDER (the LOPTICS=.TRUE. run) as the number of bands you use in the BSE run?

Perhaps you can try to do step 4 of this tutorial for the 64 and 256 bands cases as a test to check whether you get a correct IP dielectric function:
wiki/index.php/Dielectric_properties_of_Si_using_BSE

Do you obtain the small gap warning in the calculation with 64 bands as well?
Coul you maybe also post the OUTCARs of the 64 and 256 bands runs?

[chandra_sarap1]

Hi Henrique,

I too face the same problem while using 256 bands and showing oscillator strengths zero. I can't reduce the NBANDS as I have 344 electrons. I herewith attach my OUTCAR files using GW and BSE. I kept my NBANDS constants throughout the calculations.

Can you please look into it and help me in sorting out the issue. Please let me know if you need any information from my end.


Thanks and Regards,
Chandra

[ferenc_karsai]

Can you please upload also your POSCAR, INCAR, POTCAR and INCAR file for the preceding DFT calculation.

[chandra_sarap1]

Hi,

Please find the attached input files.


Thanks,
Chandra

[ferenc_karsai]

The size of the system is really huge and hence too large for proper investigations. Can you reduce the size of the system to a minimum where the problem still appears?

[chandra_sarap1]

Thank you for the reply.

I can't reduce the size of the molecule. Reducing the size lead to loss of linearity in the backbone.

Is there a way to consider only valence electrons of the system where I can reduce the number of bands?


Thanks!
Chandra

[ferenc_karsai]

We only use valence states in the calculations.

Primarily the size of the cell is the problem. Running GW and also BSE with 100 atoms is very costly and for trial and error runs inspecting the behaviour of the code simply too costly.
Here in first place we are not interested in physical correctness but to reconstruct the "ill" behaviour of the code.
So please try to construct a minimum size cell that shows the same problem even if you brake the backbone (i guess you will only need to have the same constituents).