[Pw_forum] dedicated wfcdir for PWCOND

J. D. Burton jlz101 at unlserve.unl.edu
Thu Nov 19 18:12:59 CET 2009


Dear Gabriele,

>> Thanks for the info. I obviously knew the energy-dependent wavefunctions
>> calculated in PWCOND are different from the Bloch eigenvectors generated
by
>> the periodic pw.x. 

>Not necessarily. I think that the propagating Bloch waves (i.e. those at
real k vector) 
>computed in the leads should coincide with those from pw.x, while the
evanescent states 
>are not present in pw.x due to PBCs along z, obviously.

Let me rephrase: The wavefunctions generated by pw.x are, in general,
fundamentally useless for the PWCOND calculation because they are only
generated on a comparatively sparse k-mesh for self-consistency, whereas
PWCOND uses the propagating modes (i.e. complex kz values, etc) of the leads
at a given input energy (which in general is not on the list of eigenvalues
found in pw.x) on a much more dense mesh of k_perp points.

This is not to say that if you were to use pw.x to generate a wfc at just
the right (real) k_z at just the right k_perp corresponding to just the
right energy that it would be different from that found by PWCOND... it's
just that would be a pointless procedure.

Cheers,
J. D.

************************************
J. D. Burton, Ph.D.
jlz101 at unlserve.unl.edu
University of Nebraska Lincoln
Physics and Astronomy
Office Ph. (402) 472 2499
Mobile Ph. (402) 419 9918
213C Ferguson Hall
************************************
"The job of a scientist is to generate wrong ideas as fast as possible."
-- Murray Gell-Mann


-----Original Message-----
From: pw_forum-bounces at pwscf.org [mailto:pw_forum-bounces at pwscf.org] On
Behalf Of Gabriele Sclauzero
Sent: Thursday, November 19, 2009 10:36 AM
To: PWSCF Forum
Subject: Re: [Pw_forum] dedicated wfcdir for PWCOND



J. D. Burton wrote:
> Dear Gabriele,
> 
> Thanks for the info. I obviously knew the energy-dependent wavefunctions
> calculated in PWCOND are different from the Bloch eigenvectors generated
by
> the periodic pw.x. 

Not necessarily. I think that the propagating Bloch waves (i.e. those at
real k vector) 
computed in the leads should coincide with those from pw.x, while the
evanescent states 
are not present in pw.x due to PBCs along z, obviously.

> I just never noticed that those wfc files made by PWCOND
> were empty! So I guess everything is held in the RAM on worker nodes,
which
> means I just need to be careful about number of processors (available
> memory) vs system size.

You are right. If the system is very big, the memory requirement may be huge
and so you 
should be careful. You can try to reduce the G_perp basis set in the
perpendicular 
directions by playing with ewind and epsproj parameters.

HTH

GS

> 
> Thanks again,
> J. D.
> 
> ************************************
> J. D. Burton, Ph.D.
> jlz101 at unlserve.unl.edu
> University of Nebraska Lincoln
> Physics and Astronomy
> Office Ph. (402) 472 2499
> Mobile Ph. (402) 419 9918
> 213C Ferguson Hall
> ************************************
> "The job of a scientist is to generate wrong ideas as fast as possible."
> -- Murray Gell-Mann
> 
> -----Original Message-----
> From: pw_forum-bounces at pwscf.org [mailto:pw_forum-bounces at pwscf.org] On
> Behalf Of Gabriele Sclauzero
> Sent: Thursday, November 19, 2009 1:47 AM
> To: PWSCF Forum
> Subject: Re: [Pw_forum] dedicated wfcdir for PWCOND
> 
> Dear JD,
> 
> J. D. Burton wrote:
>> Hi everyone,
>>
>>  
>>
>> When running PWCOND, is there a way to keep .wfc* files in a dedicated 
>> directory other than outdir? I know this can be done in pw.x and it 
>> really helps minimize network communication during parallel jobs by 
>> keeping the wfc files on the worker nodes. If it's not there, then I
> 
> Actually pwcond.x in the present implementation does not use the
> wavefunctions produced by 
> pw.x, but only the charge density and the informations in the
data-file.xml.
> The only thing needed to run a standard pwcond calculation is the
> prefix.save directory. 
> If the wavefunction files prefix.wfc are not present, the code will create
> them in outdir, 
> but those files will NOT be used during the calculations (in fact, if you
> look at their 
> size it should be zero).
> If there are already the wavefunction files from the previous pw.x
> calculation, they will 
> be left untouched.
> This means also that you can do the pwcond calculation with any number of
> processors, 
> irrespectively of how many you used for the pwscf one.
> 	
>> suppose I can try to put it into PWCOND myself. unless there is some 
>> fundamental difficulty that I haven't thought of yet.
> 
> If you are volunteering to extend to pwcond the usage of wfc_dir feature,
> you're welcome! 
> Anyway I don't feel there is the need for that at the moment, since, as I
> told you, there 
> is no network communication issue related to wavefunction files in pwcond.
> 
> 
> HTH
> 
> GS
> 
>>  
>>
>> Cheers,
>>
>> J. D.
>>
>>  
>>
>> ************************************
>>
>> J. D. Burton, Ph.D.
>>
>> jdburton1 at gmail.com
>>
>> University of Nebraska Lincoln
>>
>> Physics and Astronomy
>>
>> Office Ph. (402) 472 2499
>>
>> Mobile Ph. (402) 419 9918
>>
>> 213C Ferguson Hall
>>
>> ************************************
>>
>> "The job of a scientist is to generate wrong ideas as fast as possible."
>> -- Murray Gell-Mann
>>
>>  
>>
>>
>> ------------------------------------------------------------------------
>>
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://www.democritos.it/mailman/listinfo/pw_forum
> 

-- 


o ------------------------------------------------ o
| Gabriele Sclauzero, PhD Student                  |
| c/o:   SISSA & CNR-INFM Democritos,              |
|        via Beirut 2-4, 34014 Trieste (Italy)     |
| email: sclauzer at sissa.it                         |
| phone: +39 040 3787 511                          |
| skype: gurlonotturno                             |
o ------------------------------------------------ o
_______________________________________________
Pw_forum mailing list
Pw_forum at pwscf.org
http://www.democritos.it/mailman/listinfo/pw_forum




More information about the users mailing list