[Q-e-developers] Hubbard manifold

[Timrov Iurii]

Dear Paolo,


Matteo and I were looking at this. Within pw it is of course useful to keep the atomic wfc on disk during the scf cycle because the occupations are updated each time Kohn-Sham wfc are. However, they can be erased at the end of the calculation. So, for the purpose of +U calculations, while it makes sense to update the way atomic wfc are written and to make it consistent with the new way (one separate file per k-point), it is not crucial. In case a restart is needed (e.g., after a crash) they can be regenerated without any problem.

We think instead that it is important to keep the atomic occupations and have the code read them upon restart. Maybe it is actually a good idea to have an input variable to force it read them (to improve the restart or to have it read the occupations of another run, etc).



a) which codes read that file? for sure, pmw.x, gipaw.x, others?

The other codes we are working on that deal with atomic wfcs are ph - related (dfpt based) - but they are not public yet. In the latest svn version Ph still writes the wavefunctions (the KS evc) the old way (one file per processor, located outside the directory .save), so the style of Hubbard file is still consistent. In any case in our version of the code ph regenerates atomic wavefunctions because it also needs them at k+q. So we don't need atomic wfc to be kept.

Maybe xspectra and Paolo Umari's GW codes need them when using DFT+U?

b) couldn't these code just regenerate the manifold?

Our 5.1 ph-based code does that already (see above)

c) alternatively: shouldn't the format of the Hubbard manifold be the same as for wavefunctions?

yes. In case other codes need to keep them, we think having the atomic (Hubbard) manifold written in .save and in as many files as k-points is a useful thing.

Best,

Iurii and Matteo


________________________________
From: q-e-developers-bounces at qe-forge.org <q-e-developers-bounces at qe-forge.org> on behalf of Paolo Giannozzi <p.giannozzi at gmail.com>
Sent: Friday, September 1, 2017 4:32 PM
To: General discussion list for Quantum ESPRESSO developers
Subject: [Q-e-developers] Hubbard manifold

Currently pw.x saves the atomic wavefunctions with a related Hubbard term into the old "distributed" format (one tmp_dir/prefix.hub* file per processor).
Questions:
a) which codes read that file? for sure, pmw.x, gipaw.x, others?
b) couldn't these code just regenerate the manifold?
c) alternatively: shouldn't the format of the Hubbard manifold be the same as for wavefunctions?

Paolo
--
Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20170906/80260bb8/attachment.html>