The ABINIT Users Mailing List ( CLOSED )

["Anglade Pierre-Matthieu" <anglade@gmail.com>]

Hi,

One last advice : when compiling and then running the code you must be
very carefull to use every "tricks" described by François Bottin on
page 18 of his presentation at the 3rd IADW. This is very important
and unfortunately seems not described by the "help menu" of the
configure script.

best

PMA

On Tue, Jun 10, 2008 at 3:02 PM, Xuan Luo <xuan.us@gmail.com> wrote:
> Hi Francois and PMA,
>
> Thank you very much for your input.
>
> I will try again and hope I can make it.
>
> Thanks a lot!
>
> Rose
>
> On Tue, Jun 10, 2008 at 3:09 AM, BOTTIN Francois <francois.bottin@cea.fr>
> wrote:
>>
>> Hi PMA,
>> You're right.
>> PAW calculations can be run using Band-FFT-kpoint parallelization for a
>> long time.
>> As concerns spin polarization, this functionality didn't work one year
>> ago.
>> However, this one is now available, at most in 545 prefered production
>> version.
>> Starting from this version, the following error message:
>> if (nsppol /= 1) then
>>       write(message,'(a,a,a,a,i3,a,a,a,a)')ch10,&
>>   &   ' *chkinp* : ERROR -',ch10,&
>>   &   '  The value of nsppol is found to be ',nsppol,ch10,&
>>   &   '  Spin-polarized calculation is not yet allowed in the case of
>> band-FFT parallelization.',ch10,&
>>   &   '  Action : put nsppol = 1 in your input file'
>>       call wrtout
>> <http://www.abinit.org/package/robodoc_5.3.5/01manage_mpi/wrtout_F90.html#robo1099>(6,message,'COLL')
>>       call leave_new
>> <http://www.abinit.org/package/robodoc_5.3.5/01manage_mpi/leave_new_F90.html#robo609>('COLL')
>> end if
>>
>> is then removed from chkinp.F90, which means that spin-polarized
>> calculations are allowed in band-FFT-kpoint parallelization.
>>
>> Just one remark:
>> This does not use the *band-by-band* parallelization of CG.
>> This call the *blocked* eigensolver LOBPCG (see all the input variables
>> needed, as adviced by PMA)
>>
>> Regards,
>> Francois
>>
>>
>> Anglade Pierre-Matthieu a écrit :
>>>>
>>>> Would you please give me further information on "one can run bandfft
>>>> parallelized
>>>> jobs with PAW" ?
>>>>
>>>>
>>>
>>> That's what I thought.
>>> But it appears (from your message) that you tried and failed.
>>> So I must have mistaken right ?
>>>
>>> Nevertheless, after checking again, it appears that F. Bottin
>>> presentation (the one I have indicated to you) displays result of so
>>> called bandfft parallelization using PAW...
>>> I don't know if it can deal with spin-polarized cases...
>>>
>>> In order to check I've compiled Abinit, take one of the PAW test case,
>>> add bandfft parallelisation options (look carefully at the doc about
>>> "npband"),"nspden 2" and "spinat ...", choose  correct "ixc" ...
>>> And apparently it is running . So I'm a little bit confused by your
>>> sentence :
>>> <<I found it is limited to  "No spin-polarization, No PAW calculations."
>>> >>
>>>
>>> best
>>>
>>> PMA
>>
>> --
>> ##############################################################
>> Francois Bottin                    tel: 01 69 26 41 73
>> CEA/DIF                            fax: 01 69 26 70 77
>> BP 12 Bruyeres-le-Chatel         email: Francois.Bottin@cea.fr
>> ##############################################################
>>
>
>



-- 
Pierre-Matthieu Anglade