The ABINIT Users Mailing List ( CLOSED )

[TORRENT Marc <marc.torrent@cea.fr>]

Willy,

You chose to download a PAW dataset from the atompaw table.

Such datasets are generated on a regular grid (here, more than 4000) points, and:
 - this is not efficient for Abinit; it's better to use datasets on logarithmic grids (less cpu-demanding)
 - it seems that there is a problem with regular grids in the last version of Abinit; this problem was already mentioned to me and I had forgotten it; it's now up in my personnal list...
Regular grids where accepted in older versions of Abinit, but a new developpement seems to have broken this capability.

So, what is the solution ?:
If you absolutely want to use this PAW dataset, you have to convert it on a logarithmic grid.
You have to use the atompaw tool (+atompaw2abinit translater).
Download the input file from http://www.wfu.edu/%7Enatalie/papers/pwpaw/periodictable/atoms/NO/O_GGA_input
Add "loggrid" (without quotes) at the end of the second line
Download the atompaw and atompaw2abinit tools from http://www.abinit.org/PAW/AtomPAW2Abinit-Manual-html/AtomPAW2Abinit.htm
Compile them and use them as:
atompaw < O_GGA_input
atompaw2abinit (reply O.atomicdata to the first question and type enter to the others)
You get a abinit.pawps file which is exactly the same PAW dataset as O_GGA_abinit except that it is defined on a logarithmic grid.

I enclose the result I get with your input file, using the new PAW dataset; O-O bond length seems acceptable.

Also note that the O_GGA_abinit dataset uses only 2 non-local projectors for the PAW on-site parts... this seems to be enough for the dimer, but couldn't be for other systems.

Don't be surprised by the several WARNING about negative densities; they generally appear when the simulation cell contains vaccuum (this is the case here) and should disappear before the end of the computation. You can forget them is the printed negative value is low (10-5, 10-6...).


Marc

Willy Offermans a écrit :

20080710142002.GA4387@wiz.vpn.offrom.nl" type="cite">

Dear abinit friends,

I have installed the abinit versions 5.5.3 and 5.5.4 on my laptop running
debian linux 2.6.22 kernel.

In both cases, I could perform calculations with the 08-O.GGA.fhi PP file and
O-GGA-hard-uspp.paw. The results look promising. At least the O-O
bond length is reasonable.

However, in both cases, I obtained an error msg by using the O_GGA_abinit PAW file.
<snip>

  Fast computation of inverse cubic root failed.
  Action : contact ABINIT group.

</snip>

I was advised to contact the abinit group. I will include the input
files and the log files.

So I cannot yet conclude that 5.5.4 will give a reasonable result.


I hope someone can provide feedback.

On Thu, Jul 10, 2008 at 11:33:57AM +0200, Anglade Pierre-Matthieu wrote:
  

Matthieu, In this case I think you meant "full proof" not "fool proof" right ?

On Thu, Jul 10, 2008 at 11:07 AM, matthieu verstraete
<matthieu.jean.verstraete@gmail.com> wrote:
    

Hello Willy,

alwayts take the latest (5.5.4 and soon 5.5.5): the last number implies it
has had the most number of bug fixes - also incorporating, e.g. bugs which
are found and fixed in 5.6 (not released yet) and which were present in 5.5.

It would be good to hear PAW is stable and platform independent. The
automatic tests do a lot to this end, but obviously are not fool proof.

Matthieu


~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
              NEW ADDRESS/TELEPHONE 15/5/2008
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Dr. Matthieu Verstraete

European Theoretical Spectroscopy Facility (ETSF)
Dpto. Fisica de Materiales,                  Phone: +34-943018393
U. del Pais Vasco,                           Fax  : +34-943018390
Centro Joxe Mari Korta, Av. de Tolosa, 72,   Mail : mjv500@york.ac.uk
E-20018 Donostia-San Sebastian, Spain

http://www-users.york.ac.uk/~mjv500

      

-- 
Pierre-Matthieu Anglade
    

Attachment: O2.out.loggrid.tar.gz
Description: GNU Zip compressed data