[Pw_forum] XSpectra and US or PAW pseudos?

[Adrian Hühn]

Hello Delphine,

thanks for your quick reply! This seems to be a contradiction to my 
observations. Maybe I did not write this in detail, but I was only able 
to get sensible results when calculating diamond USPP spectra with 
v.5.0.2. The same setup fails with the recent v.5.2.0. To be more 
specific: Both PWscf and the XSspectra calculations run on both 
versions, but the data from "xanes.dat" is useless on the recent 5.2.0. 
I uploaded the setups.
http://www.file-upload.net/download-10870211/diamondh_uspp_5.0.2.tar.gz.html
http://www.file-upload.net/download-10870213/diamondh_uspp_5.2.0.tar.gz.html

Concerning my problem with periclase or magnesium pseudopotentials (my 
last trials are with NC): do you think this could also trigger the 
segmentation fault?
I recompiled QE 5.0.2 without parallelization (./configure 
--disable-parallel) but this does not make any difference.

At I deeper look I can see, that the code fails at the same point, 
perhaps when "paw_recon(xiabs)" is accessed.
espresso-5.0.2/XSpectra/src/xspectra.f90:461 or 
espresso-5.2.0/XSpectra/src/xspectra.f90:592
[code]
456     ! 
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$
457     !  Reads reconstruction files
458     ! 
$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ 

459
460     CALL read_core_abs(filecore,core_wfn)
  461     IF ( .NOT. paw_recon(xiabs)%gipaw_data_in_upf_file ) &        
! SEGFAULT !
462          CALL read_recon ( filerecon(xiabs), xiabs, paw_recon(xiabs) 
) !*apsi
[\code]



Am 28.08.2015 um 00:56 schrieb Delphine Cabaret:
> Dear Adrian,
>
> Concerning your first question,  v.5.0.2 contains a bug related to the
> parallelization
> that could be at the origin of the regression you observed.
> This bug has been corrected in the version available on the SVN.
> I suggest you to go on with the v.5.0.2 for the moment.
>
> Regards
>
> D. Cabaret
>
>
> Le 27/08/2015 21:58, Adrian Hühn a écrit :
>> Dear Developers and Users,
>>
>> when using XSpectra everything goes fine when go through the following
>> tutorial:
>> http://www.ictp.it/~rgebauer/Matteo_XAS.tar.gz
>> going further I could nicely reproduce the experimental carbon K-edge
>> X-ray absorption spectrum using PBE.
>> I could also generate spectra using this US pseudopotential:
>>        http://theossrv1.epfl.ch/uploads/Main/NoBackup/pbe.0.3.1.tgz
>> (of course it was also modified to generate the core-hole). Here seems
>> to be a regression when using the latest version 5.2.0. At least I only
>> got consistent results, when using v.5.0.2.
>>
>> Next I wanted to calculate the magnesium K-edge in Periclase. I tested
>> the Mg pseudopotentials from the pslibraries, thereunder the NC and the
>> USPP variants. BOTH lead to a segfault. I already messed with th
>> generation parameters of the Mg pseudopotential and changed the
>> r_paw(1)-parameter in the XSpectra input but no avail.
>> I built QE with GCC 5.2 under Arch Linux on my local machine, but the
>> problem applies at our cluster (using the Intel compiler) as well.
>> Did I miss anything? The Inputs and results are added to the archives.
>>
>> I firstly planned to calculate Mg K-edge spectra of MgO2 species on a
>> Au(111)-surface using PAW pseudopotentials, but I am willing to use USPP
>> or NCPP if they work better with XSpectra.
>>
>> Regards
>> Adrian Huehn
>>
>>
>> Mulliken Institute for Theoretical Chemistry
>> University of Bonn
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://pwscf.org/mailman/listinfo/pw_forum