The ABINIT Users Mailing List ( CLOSED )

[Fabien Bruneval <fabien.bruneval@polytechnique.fr>]

Dear Pierre-Yves,

The situation is:
your k-points are correctly placed on a regular grid and you want to 
generate ALL the q-vectors, which are differences k_i - k_j to calculate 
the screening.

If your k-points are regularly spaced, you should convince yourself that 
building them with

q_i = k_i - k_1

or with

q_i = k_i - k_2

or whatever

will give the SAME set of q-vectors. So there is no need to consider all 
pairs k_i - k_j. Just keep one fixed and vary the other one.

Therefore, the main point is to have a KSS file with k-point regularly 
spaced in the BZ. Please use a Monhorst-Pack grid, as for a standard DFT 
calculation. If you don't satisfy this requirement, your calculations 
are likely to crash.

I hope I convinced you...

Fabien




pprodhom@chartreuse.cea.fr wrote:
> Dear Fabien Bruneval,
> Firstly, thanks a lot for your response, I am glad that you take time to answer my question, I think I well understand what you said. 
> But I guess that my question was not very clear. In fact it seems to me that all the possible q-vectors are not generated during the step of the screening calculation, because I look into the code and I saw that when it made the q vectors, it uses this: q=kj-k1 but k1 is just the first k point in the IBZ of the KSS file. So the q-vectors q=kj-ki for i=/1 are not calculated.
>
> Then if we want to calculate (in the same run) sigma on a k-point k2 
> different from k1, there is a problem because abinit can't find the q vectors 
> q=kj-k2 because it didn't generate it in the previous step. So the error 
> message appears:
>
>   **error: q = k - kp + g0 not found.
>   kmkp  =    0.5000000000000000       -0.5000000000000000
>      0.000000000000000
>
> Then maybe I could change that modifying the code. That would avoid me to 
> launch the calculation as many times as the number of k-points where I want 
> to make a correction.
> So I want to know if what I have said on q vectors is correct?
>
> Best regards
> Merci Beaucoup
>
> Pierre-Yves Prodhomme
> CEA-LETI/Minatec/D2NT/LSCDP