Yambo Community Forum

Dear developers:

I have a problem related to the convergence test. Take bulk hBN as an example, I first generate the SAVE data with a k-mesh 6 times 6 times 2, then based on this database I perform a series of convergence test related GW calculation (or BSE calculation). Then I obtained reasonable parameters to converge GW gap. Once these parameters I have got, the last convergence test should be done with different PW NSCF k-mesh. My question is the new NSCF k-mesh will affect previous converged variables?

Dear Longjun Xiang,
while the number of bands and dimension of the screening matrix are linked to each other, the k point dependency is rather independent of the other parameters, so you can maintain the convergence parameter found with the lower sampling.

Best,
Daniele

OK, many thanks for your .

Rather than using your Yambopy module, I have written a series of python scripts to perform convergence tests. The following is my strategy (GW runlevel):
1.QE scf (keep converged at DFT level)
2.QE nscf with the lowest DFT converged k-mesh and generate SAVE database.
3.GW_setup
a.initializing
b.find the K-point where the direct band gap is minimum. In the following, I will test the gap (LUMO-HOMO) convergence at this K-POINT.
4.GW_SIGMAX
a.convergence test for variable EXXRLvcs
5.GW_SIGMAC
a.convergence test for variable NGsBlkXp and BndsRnXp with fixed EXXRLvcs
b.convergence test for variable GbndRange with fixed NGsBlkXp, BndsRnXp and EXXRLvcs
5.GW_KMESH
Based on previous converged tags, I will perform the convergence test for different nscf K-mesh. But now I cannot pick up the same K-point to do the test. Do I need to calculate all the K-points and then pick up the global minimum band gap for different nscf K-mesh as the parameter to identify convergence? Or is there any other method to reduce computational cost?

Dear Dr. Longjun Xiang,

But now I cannot pick up the same K-point to do the test. Do I need to calculate all the K-points and then pick up the global minimum band gap for different nscf K-mesh as the parameter to identify convergence? Or is there any other method to reduce computational cost?

What you can do is to perform convergence tests on a point common to all the grid (e.g. gamma) and once the calculation is converged use that grid to calculate the minimum gap. Note that if the "real minimum" is not contained in the mesh, you can always interpolate your bands.

If you want instead to look directly to k points near the minimum, maybe you can check the convergence on the correction to the KS gap instead that on the QP energy gap itself.

Best,
Daniele

Thanks for your quick reply. I found the first point always be Gamma point according to the NSCF output file.

Dear developers, I hope all of you are doing well!

I want to ask about a parameter used in BSE convergence used the Yambopy convergence tutorial, "BSEEhEny"

Q1: Does it have to be mentioned in the input file explicitly? In the tutorial, it is mentioned that it is similar to BSEBands? So, if I have done convergence w.r.t. BSEBands then I don't need BSEEhEny anymore? Right?

Q2: BSEEhEny is in the unit of energy and its syntax is in pair as,
So, even if I opt to go with BSEEhEny, then how should I provide the energy range (since it is not the spectrum energy range)? It must be some sort of cutoff?


Thanks,

Dear Haseeb,
BSEBands and BSEEhEny are alternative.
I suggest you use BSEBands.

When using BSEEhEny you provide and energy windows (Emin and Emax) and all the bands such that Emin < Ec-Ev < Emax are included in the BSE matrix building.

Best,
Daniele