Calculation of Wannier centers and spreads

[iskarmou]

Dear Marcella,

thanks for your reply.
I tried the Jacobi algorithm as well and I have
seen that in general less iterations were required (with the
same tight convergence).
However the convergence parameter, as you say, is too tight.
The reason I used this value is not really based upon strong scientific
arguments, it's because I have seen several people using tight criteria.
Regarding the effect of convergence parameters on the results, which thing 
do you think that it would
be the most appropriate to check? 
Should I calculate eg the dipoles from the Wannier ions_center trajectory 
file using different convergence criteria?
Or the spreads as well?
Is there another way to check this effect?
And do you know if there is any "optimum" value for the convergence 
parameter for these kind of systems?

Best regards

Yannis 

On Wednesday, July 15, 2015 at 9:53:48 AM UTC+1, Marcella Iannuzzi wrote:
>
>
> Dear Yannis, 
>
> Probably the localisation algorithm requires many iterations to converge.
> Are you sure that you need such a tight convergence for EPS_LOCALIZATION?
> Did you check whether a less tight value really affects the results?
> You can try to replace CRAZY with JACOBI, maybe it converges faster. 
>
> regards
> Marcella
>
> On Saturday, July 11, 2015 at 1:06:49 PM UTC+2, iskarmou wrote:
>>
>> Dear all,
>>
>> I am simulating a mixture of H2O-CO2 and after 
>> an equilibration period of 4 ps I am performing
>> a productive run to calculate the system properties.
>> Among these properties, I want to calculate
>> Wannier centers and spreads.
>> I am using the following method (I attach the corresponding part of the 
>> input file):
>>
>>      &LOCALIZE
>>
>>      METHOD CRAZY
>>      EPS_LOCALIZATION 1.0E-8
>>
>>      &PRINT
>>         &WANNIER_CENTERS
>>           &EACH
>>             MD 1
>>           &END EACH
>>           IONS+CENTERS T
>>           FORMAT XMOL
>>           FILENAME ./ions+centers.xyz
>>           COMMON_ITERATION_LEVELS               3
>>         &END WANNIER_CENTERS
>>
>>         &WANNIER_SPREADS
>>           &EACH
>>             MD 1
>>           &END EACH
>>           FILENAME ./wannier-spreads
>>           COMMON_ITERATION_LEVELS               3
>>         &END WANNIER_SPREADS
>>
>>       &END PRINT
>>     &END LOCALIZE
>>
>> However, the run is significantly slower in comparison with the
>> previous runs, when I was not calculating Wannier centers and spreads.
>> Is there any way, from a methodology and technical point (eg parameter 
>> choice) of view, to speed up my calculation
>> without losing the accuracy in the estimation of the Wannier centers?
>>
>> Thanks in advance
>>
>> Yannis Skarmoutsos
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20150717/1904c78d/attachment.htm>