[ase-users] Cores

Marcin Dulak Marcin.Dulak at fysik.dtu.dk
Tue Apr 26 10:20:40 CEST 2011


Hi,

please continue on the mailing list.
dyn or qn is just a name of a variable - this is how you name the optimizer.
It's sometimes safer to avoid symmetry=True (see 
https://listserv.fysik.dtu.dk/pipermail/ase-users/2010-October/000871.html),
As the manual says stay_alive=True saves on writing to disk: see 
https://wiki.fysik.dtu.dk/dacapo/Manual#dacapo-fortran-program-python-interface-communication
Try out the options to see which ones work better for your jobs.

Best regards,

Marcin

Tadeu Leonardo Soares e Silva wrote:
> Dear Marcin
>
> Thank you very much. It works much better! By the way what is the differ 
> between dyn = BFGS and qn= BFGS for geometry optimizations?  Would I need to 
> use symmetry=True and stay_alive=True for my geometry optimizations? 
>
> Sincerely,
>
> Tadeu
>
>
> On Tue, 19 Apr 2011 18:07:54 +0200, Marcin Dulak wrote
>   
>> Hi,
>>
>> Tadeu Leonardo Soares e Silva wrote:
>>     
>>> Dear Marcin
>>>
>>> Thank you very much. But I have a doubt. I would like to make the follow 
>>>       
> job.
>   
>>> #!/usr/bin/env python
>>> import ase
>>> from ase import *
>>> from ase.lattice.surface import *
>>> from ase.calculators.jacapo import *
>>>
>>> slab = fcc111('Pd',a=3.92, size=(5,5,4), vacuum=10.0, orthogonal=False)
>>> slab.center()
>>>
>>> calc = Jacapo(nc='superficiePd.nc',
>>>               pw=400,
>>>               kpts=(3,3,1),
>>>               xc='RPBE',
>>>               dipole=True,
>>>               symmetry=True,
>>>               stay_alive=True,
>>>               ft=0.1,
>>>               atoms=slab)
>>>
>>> #here we select atoms with tag not equal to 1 and 2 to fix
>>>
>>>
>>> constraint = ase.constraints.FixAtoms(mask=[a.get_tag() > 2 for a in 
>>>       
> slab])
>   
>>> slab.set_constraint(constraint)
>>>
>>> dyn = ase.optimize.QuasiNewton(slab, trajectory='superficiePd.traj')
>>> dyn.run(fmax=0.05)
>>>
>>> print slab.get_potential_energy()
>>> print slab.get_forces()
>>> ase.io.write ('superficiePd.xyz',slab)
>>>
>>> The .txt file:
>>>
>>>  KPT:            9  BZ kpoints specified in file
>>>  KPT: k-points in the irriducible Brillouin zone (nkprun) :    5
>>>  KPT: ----------------------------------------------------------
>>>  KPT: k-point      k-point in units of          k-point
>>>  KPT: number      B1        B2       B3         weigth 
>>>  KPT: ----------------------------------------------------------
>>>  KPT:    1   -0.333333 -0.333333  0.000000    0.222222
>>>  KPT:    2   -0.333333  0.000000  0.000000    0.222222
>>>  KPT:    3   -0.333333  0.333333  0.000000    0.222222
>>>  KPT:    4    0.000000 -0.333333  0.000000    0.222222
>>>  KPT:    5    0.000000  0.000000  0.000000    0.111111
>>>  KPT: ----------------------------------------------------------
>>>  KPT: 
>>>  KPT: Chadi-Cohen asymptotic error estimate:  0.000096675384
>>>  KPT: (see PRB 8, 5747 (1973); 13, 5188 (1976))
>>>  KPT: nkpmem :            1
>>>
>>> In this .txt file I would have just 5 five kpoint (100 Pd atoms), then I 
>>> could to use until 15 cores ( 3 cores or processors per k point). But I 
>>>       
> have 
>   
>>> 400 cores in my cluster. Could I to use more cores for this job? I 
>>>       
> belive 
>   
>>> that with 15 cores it will take much.
>>>   
>>>       
>> the job run on many more cores than k-points will be just slower 
>> that a job run on the number of cores equal to the number of k-
>> points - try it. I would set kpts=(4,4,1) (unless you really need (3,
>> 3,1)), so the number of k-points reduced by symmetry would be even,
>>  and then use an even number of cores, kpts, maybe kpts*2 - you need 
>> to test this.
>>
>> Best regards,
>>
>> Marcin
>>     
>>> What do you think about this?
>>>
>>> Sincerely,
>>>
>>> Tadeu
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> On Tue, 19 Apr 2011 10:59:04 +0200, Marcin Dulak wrote
>>>   
>>>       
>>>> Hi,
>>>>
>>>> if you use k-points please have a look at the vasp recommendation 
>>>> from http://www.raunvis.hi.is/~finnboo/bjolfur/ (last section: 
>>>> running in parallel): one processor per k-point (the number of k-
>>>> points is printed in the txt output). Make an experiment to verify 
>>>> that this holds for the configuration of your cluster. If you don't 
>>>> have k-points then you have to make experiments described at 
>>>> https://listserv.fysik.dtu.dk/pipermail/ase-users/2011-March/000976.html
>>>> Usually people run dacapo jobs on 8-32 cores.
>>>>
>>>> Best regards,
>>>>
>>>> Marcin
>>>>
>>>> Tadeu Leonardo Soares e Silva wrote:
>>>>     
>>>>         
>>>>> Dear Marcin
>>>>>
>>>>> How do I do to calculate how many cores should I to use in Jacapo? I 
>>>>>       
>>>>>           
>>> have a 
>>>   
>>>       
>>>>> cluster with 400 cores. Could I to use all cores for a job?
>>>>>
>>>>>
>>>>> Sincerely,
>>>>>
>>>>> Tadeu
>>>>>
>>>>>  +++++++++++++++++++++++++++++++++++++++++++++++++++++
>>>>>
>>>>>  PEQ/COPPE renova o nivel de curso 7, maximo na CAPES:
>>>>>  45 anos de excelencia no ensino e pesquisa de pos-graduação em
>>>>>  Engenharia Quimica.
>>>>>
>>>>>  ************************************
>>>>>
>>>>>  PEQ/COPPE : 45 years of commitment to excellence in teaching and
>>>>>  research in Chemical Engineering.
>>>>>
>>>>>  +++++++++++++++++++++++++++++++++++++++++++++++++++++
>>>>>
>>>>>       
>>>>>           
>>>> -- 
>>>> ***********************************
>>>>
>>>> Marcin Dulak
>>>> Technical University of Denmark
>>>> Department of Physics
>>>> Building 307, Room 229
>>>> DK-2800 Kongens Lyngby
>>>> Denmark
>>>> Tel.: (+45) 4525 3157
>>>> Fax.: (+45) 4593 2399
>>>> email: Marcin.Dulak at fysik.dtu.dk
>>>>
>>>> ***********************************
>>>>     
>>>>         
>>>  +++++++++++++++++++++++++++++++++++++++++++++++++++++
>>>
>>>  PEQ/COPPE renova o nivel de curso 7, maximo na CAPES:
>>>  45 anos de excelencia no ensino e pesquisa de pos-graduação em
>>>  Engenharia Quimica.
>>>
>>>  ************************************
>>>
>>>  PEQ/COPPE : 45 years of commitment to excellence in teaching and
>>>  research in Chemical Engineering.
>>>
>>>  +++++++++++++++++++++++++++++++++++++++++++++++++++++
>>>
>>>       
>> -- 
>> ***********************************
>>
>> Marcin Dulak
>> Technical University of Denmark
>> Department of Physics
>> Building 307, Room 229
>> DK-2800 Kongens Lyngby
>> Denmark
>> Tel.: (+45) 4525 3157
>> Fax.: (+45) 4593 2399
>> email: Marcin.Dulak at fysik.dtu.dk
>>
>> ***********************************
>>     
>
>
>  +++++++++++++++++++++++++++++++++++++++++++++++++++++
>
>  PEQ/COPPE renova o nivel de curso 7, maximo na CAPES:
>  45 anos de excelencia no ensino e pesquisa de pos-graduação em
>  Engenharia Quimica.
>
>  ************************************
>
>  PEQ/COPPE : 45 years of commitment to excellence in teaching and
>  research in Chemical Engineering.
>
>  +++++++++++++++++++++++++++++++++++++++++++++++++++++
>   

-- 
***********************************
 
Marcin Dulak
Technical University of Denmark
Department of Physics
Building 307, Room 229
DK-2800 Kongens Lyngby
Denmark
Tel.: (+45) 4525 3157
Fax.: (+45) 4593 2399
email: Marcin.Dulak at fysik.dtu.dk

***********************************




More information about the ase-users mailing list