[Pw_forum] Disparities between HPC's

[Kucukbenli Emine]

Dear all,
My two cents on this infamous negative dr2 issue:

I am not a fan of commenting it out, it often is a sign of some convergence/pseudo related issue,
unless it is at the very beginning of an scf run..

Now I am looking at Jack's files:

1.The error is indeed at the beginning of a run, after a bfgs movement of atoms
2.his ecutrho = ecutwfc*4, the default value.. which is surely not guaranteed to be a converged value for a PAW pseudo..

By commenting out that error, we prevent him to look further into his systems convergence and check his pseudopotentials.

My suggestion would be any sensible combination of the following:

1.we put a check and only for the first steps, ignore if this error occurs, but report dr2 anyways
2.write a more explanatory error message or put this in FAQ with  more explanation
3.encourage users to report these pseudopotentials because i think it is most of the time due to a very hard pseudo..
4.sometimes it is possible that the assumption that we make while writing paw contribution to dr2, that exchange could be neglected, is not accurate enough, and no matter how well behaving your pseudo is, high convergence, you might stil have a small negative dr2. only after one makes sure that that is the case, there could be an input option to ignore this error, so that people dont have to modify the code that way.

I would love to hear what you think.
ciao
emine kucukbenli, postdoc, theos, epfl, switzerland

Quoting Paolo Giannozzi <giannozz at democritos.it>:

> I sent the link to the related svn commit:
>
>> http://qe-forge.org/gf/project/q-e/scmsvn/?action=browse&path=%2Ftrunk%2Fespresso%2FPW%2Fsrc%2Fscf_mod.f90&sortby=file&r1=9744&r2=9743&pathrev=9744
>
> P.
> --
> Paolo Giannozzi, IOM-Democritos and University of Udine, Italy


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20130207/1411c3b6/attachment.html>