Band gap accuracy

[niloufar]

Hello dear developers,

I'm trying to plot the band structure of Cs2PbBr3 and although I converged all the parameters that mentioned in wiki website, I got a way bigger band gap than the one in experiments. I had also converged cut off and k points in PW calculation in quantum ESPRESSO before that.
Should I increase the K points in nscf or change the parameters in Yambo input?

Thanks a lot,
Niloufar Dezashibi

[Daniele Varsano]

Dear Niloufar,

k points convergence it is much harder for GW calculations than DFT calculations, in general.
About other parameters (empty bands, cutoff in screening), I cannot say, as you should provide more information about your calculations.
Best,
Daniele

[niloufar]

Thank you for your quick response.
I attached my nscf and gw input files.
It would be nice of you if you take a look at it.

Thanks again

[Daniele Varsano]

Dear Niloufar,
some comments:
1) are you sure that 35Ry is enough for your ground state, it seems to be a quite small value. In any case, keep the VXCRLvcs equal to its maximum (cutoff of the density), probably you did, but I cannot check it as you did not post the report file.

2) The number of bands both in BndsRnXp and GbndRnge should be checked to be at convergence (50/60 are not much, so it is possible that the calculation it is not at convergence). The same for NGsBlkXp. Note that bands and NGsBlkXp are not independent (the higher the number of bands, the higher should be the cutoff in the screening).

3) The k point sampling instead it quite large, so it is probably converged. In any case, in order to accelerate the convergence of the Sigma_x wrt k points you may consider to use the random integration method (activate with -r in input). There you can consider 1 or few G vectors and around 3.000.000 random q points.
You can check the k points sampling convergence once the other parameters are converged as this is independent from the bands and energy cutoff in the screening.

Best,

Daniele

[niloufar]

Dear Daniele,
wow! Thank you very much for your comments.
I will check them and find the bands again.

Best,
Niloufar

[Daniele Varsano]

Maybe it is straightforward, but you can check convergencies just for few k points and bands (e.g. the gap) and finally run the calculations for all the k points and the bands you need the band structure with the converged parameters. This allows saving a lot of time.

Best,
Daniele

[niloufar]

Oh! Unfortunately I just saw your post! and actually I consumed a lot of time for band number and block size convergence ( still didn't run all the inputs) and I'm here because the band gap still seems about 1 ev more than experimental results. I don't have any idea how to make it correct or how I should change the parameters. I'm asking you now because I feel I'm wasting time on something wrong.... I attach the result of my conservancy test ( It's just the bandgap and what I've gotten till now) . It would be nice of you, if you take a look at it.

Thanks a lot @>-

[Daniele Varsano]

Dear Niloufar,

The file you attach it is not very explicit:
I assume that 2,3,4,5,6,7 is the cutoff in the screening matrix (is that Ry or RL?)
40-50-60... are bands in X, G or both?

Also, there is something suspicious in your convergence, you get exactly the same gap up to 10^-4 passing from 50 to 70 bands. It sounds quite strange, if you want to send the report files we can have a look.

Assuming they are Ry and you are varying the number of bands both in X and G there could be many reasons why a GW calculation does not agree with the experiments, starting from the structural model, temperature effects etc.., and also to what kind of experiments (which technique) are you comparing with? It could be also insightful to compare with previous calculations if any exists in the literature (I think so.).

Best,
Daniele

[niloufar]

Thanks again for your help .
I attached my input and report files.

[Daniele Varsano]

Dear Niloufar,

in your nscf calculations you included 50 bands as I can see in the report file:

Code: Select all

[RD./SAVE//ns.db1]------------------------------------------
  Bands                           :  50
  K-points                        :  35
....

This means that you cannot ask for more than 50 bands.
This is why even if you ask for more bands, as yo can see in the report, all the databases (screening and qp correction):

Code: Select all

 [RD./60b_4Ry//ndb.pp]------

Code: Select all

 [WR./60b_4Ry//ndb.QP]

are calculated using 50 bands, as also reported at the end of the report where you can see the mirror of the input file with the variable values that have been used.

This is the reason you have the same value for the calculation of bands higher than 50.

Considering you have 36 occupied bands, a total of 50 bands in the sum over states is most probably not converged, and you need to add more bands in the nscf calculation.

Best,
Daniele