[Pw_forum] Silent Calculations

[Stefano Baroni]

Dear S.K.S., dear all:

I maintain that if a linear-response calculation takes longer than an  
ordinary SCF calculation for a same system, this can only be due to  
the fact that the implementation of LRT is not as optimized as that of  
SCF. This may well be the case for codes in Quantum Espresso, at least  
for some systems, but it has NOT to be so by necessity. If it  
continues to be so it is that too many people have misconceptions  
about LRT and/or have not the time/fiber/keennes to improve upon  
existing implementations. Nothing wrong with this state of affairs,  
given the fact that SCF calculations have a much larger scope of  
application. It would be worse if SCF were less optimized than LRT.  
This being said, I believe that general and apodictic statements such  
as those made by SKS should be substantiated by more solid arguments  
than episodic experience. Once again, nothing wrong in reporting about  
one's own episodic experience, provided it is reported as such, and  
not as an absolute truth.

Cheers, Stefano

On Aug 9, 2008, at 8:30 AM, S. K. S. wrote:

>
>
>
> On Sat, Aug 9, 2008 at 1:52 AM, Paolo Giannozzi <giannozz at democritos.it 
> > wrote:
> >  Only for Gamma point phonon  one  can use "Frozen
> >  phonon" method because it is much faster.
>
> not true in general. Each linear-response calculation
> costs approx. as much as a self-consistent one, and
> you need at least as many scf calculations as
> linear-response calculations.
>
> Dear Sir,
> I agree with you that  for small system each
> linear-response calculation
> costs approx. as much as a self-consistent one, and
>  we need at least as many scf calculations as
> linear-response calculations.
>
> However, IN CASE OF  very large system (say about 64 atoms per unit  
> cell),
> I found that  a single representation  takes too much computer time
> compared to a single scf run.
>
> Moreover, in some cluster which have multiple cpu
> for each node and  in some quad or octa core  computer,  always
> phonon paralization  is much slower than manual
> scf paralization (I mean manually divide   your all  scf  
> calculations in
> different  processors).  In case of  Frozen phonon method, one can
> also accomplish the same goal  by using less memory and
> disk space.
> However, for non-zone center phonon Linear response is the best way.
> There is still a scope in PWSCF to make Phonon calculation much faster
> by implementing the representation number parallization.
>
> regards,
> SKS
> JNCASR
> BANGALORE
>
>
> On Sat, Aug 9, 2008 at 1:52 AM, Paolo Giannozzi <giannozz at democritos.it 
> > wrote:
>
> On Aug 8, 2008, at 18:41 , S. K. S. wrote:
>
> >  Only for Gamma point phonon  one  can use "Frozen
> >  phonon" method because it is much faster.
>
> not true in general. Each linear-response calculation
> costs approx. as much as a self-consistent one, and
> you need at least as many scf calculations as
> linear-response calculations
>
> Paolo
> ---
> Paolo Giannozzi, Dept of Physics, University of Udine
> via delle Scienze 208, 33100 Udine, Italy
> Phone +39-0432-558216, fax +39-0432-558222
>
>
>
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum



---
Stefano Baroni - SISSA  &  DEMOCRITOS National Simulation Center -  
Trieste
http://www.sissa.it/~baroni / [+39] 040 3787 406 (tel) -528 (fax) /  
stefanobaroni (skype)

La morale est une logique de l'action comme la logique est une morale  
de la pensée - Jean Piaget

Please, if possible, don't  send me MS Word or PowerPoint attachments
Why? See:  http://www.gnu.org/philosophy/no-word-attachments.html






-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20080810/31e065d6/attachment.html>