[Pw_forum] espresso3.0 different results in serial andparalelle mode...

Cyrille Barreteau cbarreteau at cea.fr
Thu Jan 26 15:30:23 CET 2006


Ok so there is nothing to worry about.
I was just a bit puzzled also by the fact that
it took 14 iterations to converge in serial and
24 in parallel (for the wire).
But maybe it was just bad luck..

  cyrille


-- 
==================================================================
Cyrille Barreteau  | phone : +33 (0)1 69 08 29 51
CEA Saclay         | fax   : +33 (0)1 69 08 84 46
DSM/DRECAM/SPCSI   | email : cyrille.barreteau at cea.fr
Batiment 462       |
91191 Gif sur Yvette Cedex  FRANCE
           ~~~~~~~~~~~~~~~~~~~~~~~~
http://www-drecam.cea.fr/spcsi/index.php
http://www-drecam.cea.fr/Images/Pisp/cbarreteau/cbarreteau_fr.html
http://www-drecam.cea.fr/Phocea/Membres/Cours/index.php
==================================================================

On Jeu 26 janvier 2006 14:42, Gerardo Ballabio a écrit :
> On 01/26/2006 02:07:42 PM, Cyrille Barreteau wrote:
>>   serial
>>      the Fermi energy is    12.4997 ev
>>
>> !    total energy              =   -61.13842278 ryd
>
>>      band energy sum           =     6.08329685 ryd
>
>>   cluster
>>      the Fermi energy is    12.4997 ev
>>
>> !    total energy              =   -61.13842278 ryd
>
>>      band energy sum           =     6.08329678 ryd
>
> I would not mind. It is quite normal for iterative methods to reach
> convergence through different paths as soon as anything changes.
> Between serial and parallel execution there are probably some
> operations that aren't performed in the same order, and as the
> numerical accuracy of computer numbers is finite, this can give
> slightly different results.
>
> It is also normal that the total energy converges to a better
> accuracy than the parts it is composed of. Thus if the convergence
> threshold is 1e-8, you'll get 8-digit accuracy on the total energy,
> but one or two less on other terms. It isn't a problem, but if you
> mind, try changing the threshold to 1e-10 or 1e-12, I'm sure the
> differences will go away (but it'll probably take a few more
> iterations to converge).
>
> Gerardo
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>




More information about the Pw_forum mailing list