ABINIT Discussion Forums

We are following the TDepES tutorial given in : https://docs.abinit.org/tutorial/tdepes/ .
We are able to run the tutorials as given. After running the tutorials, we were trying to change the input parameters to obtain convergence. In the tutorial, the k-pt grid is 2x2x2 and the q-pt grid is 4x4x4. We changed the k-pt grid to 4x4x4 and also 6x6x6. (We did not change any other input parameter in the input file). In both cases, we obtained the following errors (screen shots attached)

check if your qshift is 0 0 0 - it defaults to 0.5 0.5 0.5 (I am lobbying for it to change) which shifts your q grid to values with 1/8, which would not be on your grid.

In any event the 1/6 grid will not work - it is not a supergrid of 4x4x4

Thanks for the quick reply.
We checked the tutorial input file. It gives the k and q-pt variables as follows:

______________________________________________________________________________
ngkpt 2 2 2 # Underconverged : k-grid should be at least 4x4x4 for diamond to be converged.
nshiftk 1
shiftk 0.0 0.0 0.0
ngqpt 4 4 4 # Should be converged upon
qptopt 1
nshiftq 1
shiftq 0.0 0.0 0.0
___________________________________

We are following your suggestion. The problem is likely to be something else. Kindly let us know how to proceed.

We want to replicate the Abinit example results for convergence studies. It suggests a k-pt grid of 6x6x6 and a q-pt grid of 4x4x4 and 10x10x10. In this context, it is not really clear to us what you meant by "1/6 grid will not work - it is not a supergrid of 4x4x4". Could you also please clarify this aspect?

Thanks again for your quick response.

Probably right that the problem is elsewhere.

To be more precise about what I said, if you want to do a random qpt, you can, but you need to do an intermediate step to get the GS wave functions at all k+q points. This is also in one of the rf tutorials. IF, however, your ngqpt is a sub-set of the ngkpt grid (eg 2x2x2q for 4x4x4k) then there is no need, and abinit finds all the k+q it wants in the initial file.

I had not noticed that this calculation of the WFQ is done systematically in tdpes - it is crashing for the first non Gamma q point DS23 (which qpt is it?), and complaining about the WFQ file from DS22. Make sure that iqpt and qpt correspond in DS 21 22 23, and copy your input file at least.

Thank you so much for the reply.

Actually, I do not want to try out random q-pts because, as a beginner, I just wanted to test whether the calculations I am doing is right or not.This is because of the fact that I cannot benchmark any of these random calculations.

For, homogeneous q-pt grids, there are results available both in the "Abinit examples page" and also from literature which can be used for benchmarking. The ABINIT example page points to a reference "Ponce 2014" where the k-pt grid is 6 x 6 x 6 and the q-pt grid is 10 x 10 x 10. Further, in the Abinit example page, they have used q-pt grids that are 4 x 4 x 4, 10 x 10 x 10, 20 x 20 x 20 etc. None of the q-pt grids are sub-grid or super-grid of the 6 x 6 x 6 k-pt grid.

The ABINIT example page states that the k-pt grid can be 6 x 6 x 6 but the q-pt grid can be increased to obtain convergence. I would like to reproduce the results for these combinations since the ZPR shifts are also given in the ABINIT example page.

This is the direction I would like to proceed.

Your suggestion that the q-pt grid has to be a sub-grid of the k-pt grid means that I will have to use very large k-pt grid so that the q-pt grid can also be large. Also, I will not have any benchmark results to compare my results with, since the results are not available in the ABINIT example page. I ran the standard input file in Abinit with k-pt grid = 2 x 2 x 2 and
q-pt grid = 4 x 4 x 4, which was running perfectly fine.

Later, I modified q-pt grid to 2 x 2 x 2, which also gave results.

When we make k-pt grids 4 x 4 x 4 and q-pt grids 4 x 4 x 4, there are errors as shown in my first post.

That is, I am able to get results for two cases: i) 2 x 2 x 2 k-pt and 2 x 2 x 2 q-pt and ii) 2 x 2 x 2 k-pt and 4 x 4 x 4 q-pt.

But I am not getting results for 4 x 4 x 4 k-pt and 4 x 4 x 4 q-pt grids.

I am attaching input files which I had used for the working and non-working cases, so so that you can see all the numbers that I used.

Please advise about how to proceed so that we can reproduce the examples as given in the ABINIT examples page.

Once again, thank you so much for your time and your suggestions.

Arjun

Hello,

setting

kptopt?2 3

fixes it.

Actually, the full calculation of the EPC matrix elements requires that you choose a _unique_ set of k+q wavefunctions (kptopt 3) in the WFQ file. This is a bug in the tutorial - I'll get it corrected. The code refuses to use symmetry to complete the k+q, because the symops used might be different for q1 and q2, implying that the matrix elements have a random phase difference and can not be summed or combined.

Thank you so much for your suggestion.

We were able to make the suggested change and run Example-3 with
tdepes_3.in. Not only that, the next example-4 (for eigenenergy
corrections along high symmetry lines) which uses the input file
tdepes_4.in also had given the same error. After incorporating your
suggestions, it also worked.

Using your suggestion, we were able to run both example-3 (tdepes_3.in)
and example-4 (tdepes_4.in) for cases with kpt4x4x4, 6x6x6 and 8x8x8 and
qpt2x2x2 and 4x4x4.

However, in case of example-3, when I subsequently run the
"temperature_final.py" file, I am getting errors for kpt4x4x4 and 6x6x6
with qpt4x4x4, screenshots of which are attached.

In contrast, in example-4, we were able to run temperature_final.py to get
the temperature dependent eigenenergy corrections after the calculation.

It is not clear to me as to what needs to be done to get proper results
for the example-3 case. I have used the input file as suggested in the
ABINIT examples page and it is also attached.

I request you to let me know how to proceed further.

Once again, thank you for all your suggestions.

Arjun

Hi again,

I am afraid I don't know enough about the python scripts to help you - looks like you don't have the same number of k-points everywhere. Perhaps one of your datasets was run with kptopt 1 instead of 3?