GW correction is NaN for some k-point grids

[Christian Koenig]

Dear Daniele,

Sure, I'll try version 4.5 as well. To which values did you change the input parameters?

wf_collect=.true. should be default and no other options are implemented. At least I didn't get any error messages from p2y.
Regarding the pseudopotential I don't think I have any other choice (it's supposed to be norm conserving and fully relativistic).


Best,

Christian

[Daniele Varsano]

Dear Christian,

my parameters were reduced to the minimum to run on my laptop.
I reduced the kinetic energy cutoff to a ridiculous value in the scf/nscf, next I considered 93 bands in the X and Sigma, but considering very few blocks in dielectric matrix. All these changes should not be relevant to the problem you have. In order to have a fast calculation of the dipoles I neglected the non-local commutator (this is done by change the name to the ns.kb_pp_pwscf file.

Please note that you can have norm-conserving and fully relativistic pseudo without non-linear core correction, but most probably you need to generate them, e.g by using the Don Hamman code that you can find here (http://www.mat-simresearch.com). In any case this is not related at all with the NaN you observe.

Anyway, let's what happen with the 4.5 release as many bugs were fixed.

Best,
Daniele

[Christian Koenig]

Dear Daniele,

The calculation with the latest version (4.5.1) just finished with the same results, i.e. NaN for 24x24x1 k-points.


Best,

Christian

[Daniele Varsano]

Dear Christian,
ok, we will need to reproduce the error in order to spot the problem.
Can you try to prepare a very fast test that shows the problem, even if unphysical? For instances reducing the kinetic energy cutoff,
and even better if the problem appears also for a smaller k point sampling. The last test you post actually it is not ideal for debugging purpose. The ideal one would be something running in a few minutes with few CPUs. At the moment I do not even have 40cpu's available.

Many thanks,
Daniele

[Christian Koenig]

Dear Daniele,

Sorry for the delay in getting back to you. I tried again to prepare an example which runs fast but all of these would not reproduce the problem.
The DFT cutoff (ecutwfc) seems to be a key factor here and I fixed the problem by changing it from 50 Ry to 40 or 60 Ry. This worked fine for my calculations and also for the example, so I think this information might be useful.


Best,

Christian