[Pw_forum] Modification of sort.f90 in v-3.0

David Prendergast davegp at civet.berkeley.edu
Tue Apr 4 20:56:11 CEST 2006

Thanks Paolo,

I am also aware of the dangers of changing libraries. I just thought I 
should mention this to save some other poor souls that might be digging 
around in the save files produced by PWSCF. The best of luck in 
developing the newer (more transparent) output format.

And, let me reiterate that this version and previous versions of the 
code produce consistent output. This posting was not a criticism, just 
an update.

However one might notice some problems exchanging wave functions between 
different versions - something to be avoided in general.


Paolo Giannozzi wrote:

>On Thursday 30 March 2006 11:37, David Prendergast wrote:
>>Please note that there is a significant change to the subroutine
>>hpsort_eps in the espresso-3.0 distribution with respect to previous
>>versions. (look in $ESPRESSO/flib/sort.f90)
>>This routine is used in ordering the list of G-vectors by magnitude
>>($ESPRESSO/PW/ggen.f90) and is important for those who wish to know the
>>mapping between G-vectors and wave function coefficients, as stored in
>>the .save file. 
>>Fixing this bug is a definite improvement, however, it would be good to 
>>list such a major change in the log files provided in the directory
>>$ESPRESSO/Doc/.  This would allow users to more effectively update their
>>own tools between new releases of espresso.
>the change of ordering you mention was not especially advertised 
>because it was not expected to affect anybody. The safe way to 
>read data produced by espresso is (or rather, was) to use the 
>postprocessing codes, modifying them if necessary to perform 
>the desired calculations. If you write your own tools, you are always 
>exposed to the risk of unexpected changes of the file format.
>[ A true story; years ago, during a period spent at IBM Zurich, I
>  tried to read data that had been produced during a preceding stay.
>  The executable that was reading the data was the same that had
>  produced them, but I got meaningless results. Finally I found that
>  the ordering of G-vectors in the data file was not the same of what
>  I was obtaining. Apparently the essl library, used by that code 
>  version for sorting, had been updated and didn't produce any 
>  longer the same ordering of G-shells ]
>You won't be pleased to learn that in the new version of espresso
>the format of the output data file is completely different, but the 
>new format will be much easier to parse and to read, especially 
>for those who know nothing about the internals of espresso. The
>goal is to make data exchange easier than it is now. 

  David Prendergast
  University of California, Berkeley             phone: (510) 642-2635
  Department of Physics                          fax:   (510) 643-9473
  366 LeConte Hall #7300              email: davegp at civet.berkeley.edu
  Berkeley, CA 94720-7300       web: http://civet.berkeley.edu/~davegp

More information about the users mailing list