[gpaw-users] 'TypeError' error in plane-wave mode calculation

[Marcin Dulak]

Hi,

On 07/17/2014 04:17 PM, Alfredo Palace Carvalho wrote:
> OK, the calculation is running now. Thanks.
>
> Yeah, the empty except was really dumb. I didn't think of removing it 
> in order to get a meaningful error message. Sorry!
> That was the laziest solution I could come up with in order to get a 
> .gpw file even when the calculation fails because the maxiter 
> iterations is reached.
> Yes, I know there's that "observer" stuff I could use. Did I stressed 
> the word "lazy" before? :)
> Thanks for your help.
>
> Now, for another issue (sorry, I don't stop!): shouldn't the 
> plane-wave calculation take less memory than a finite-difference one? 
> I know that for this particular calculation I'm using a much larger 
> box than it would be needed for the small molecule in question. But I 
> think that's beside the point (or maybe not!), the fact is that the 
> exact same system (same box size and all) takes much more memory in PW 
> (about 6x according to gpaw's estimates, I don't know how accurate 
> they are, but even if inaccurate, there's no question it's a lot more).
i guess for this size of cell the grid mode will be faster (and you can 
use davidson which will converge faster for more difficult systems than 
rmm-diis).
On the other hand you may need to go to grid spacing ~0.16 if you really 
want to converge energy differences for systems with O, N,
where in PW mode 550-600 eV should be enough.
The memory estimate in PW mode can be significantly underestimated (a 
factor of 2), and even more when calculating the forces. with GPAW-0.10.0.
There was a very recent commit in the trunk that improves this.

Best regards,

Marcin
>
> Cheers,
> Alfredo
>
> On 2014-07-17 13:02, Marcin Dulak wrote:
>> Hi,
>>
>>  On 07/17/2014 12:56 PM, Alfredo Palace Carvalho wrote:
>>
>>> Well, it doesn't even get to the point to start the SCF iteration
>>> procedure. It blows just before that part begins.
>>> I'm attaching the output.
>>  the error should say that band-parallelization is not allowed with
>> dav:
>>  ValueError('Davidson eigensolver does not support band ...
>>  or
>>  RuntimeError('use fewer bands or more basis functions')
>>
>>  This error is discarded by the empty except (empty excepts are
>> dangerous),
>>  and you get only the error showing that writing of gpw fails.
>>  Use: eigensolver='rmm-diis'.
>>
>>  Normally for a molecule you could use non-periodic calculation, but
>> the structure looks shifted
>>  (probably obtained from some periodic calculations).
>>  If this is a molecule (non-periodic system) i would run it on 16
>> cores like:
>>  ...
>>  system.calc = GPAW(mode=PW(450),
>>  nbands = 64,
>>  xc = 'PBE',
>>  occupations = FermiDirac(0.0),
>>  txt = name + '.out',
>>  eigensolver='rmm-diis')
>>  ...
>>
>>  450 eV may be insufficient for Nitrogen.
>>
>>  Best regards,
>>
>>  Marcin
>>
>
>
>


-- 
***********************************
  
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

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