[Pw_forum] SCF correction compared to forces is too large, reduce conv_thr

Stefano de Gironcoli degironc at sissa.it
Mon Mar 23 06:40:49 CET 2009


ear Asha gupta,

if your are sure that the system is an non-magnetic insulator then  
there is no point in using smearing and include more bands.
Your smearing_width looks very small to me, especially with  
methfessel-paxton which, being non positive definite, could give  
problems of negative occupations if the k-point sampling is not good  
enough.
If you want to keep the smearing use marzari-vanderbilt ... or even  
plain gaussian smearing ... for an insulator it should not matter  
anyway.
As I said you may want to try mixing_mode='local-TF'.

stefano

Quoting Asha gupta <toashagupta at gmail.com>:

> Dear Stefano,
>     i am using the following:
> &system:  ibrav=0, nat= 98, ntyp=4, ecutwfc = 30.0,ecutrho =
> 180.0,occupations='smearing', smearing='methfessel-paxton',
> degauss=0.003,nbnd=500,
> &electrons: conv_thr = 1.0e-5,mixing_beta = 0.1,
> K_points sampling: 4  4  4  0  0  0,
> the system is insulator and non-magnetic for sure. the number of bands are
> little higher than what is actually required for an insulator. Tha same k
> points sampling for smaller similar system gave  results which tallies well
> reported data.
>   Shall i stop the calculation or let it run to completion? after compleion,
> then i can take the new refined data as my input and reduced conv_thr and
> re-run the calculation again.
> regards
> artee
> department of physics
> Bhavnagar university
> india
>
> On 3/22/09, Stefano de Gironcoli <degironc at sissa.it> wrote:
>>
>> dear artee sharma,
>>
>> conv_thr=1.0e-5 is a rather large value, acceptable if you just want
>> the energy but not if you want forces. 1.0e-6 is the default and for
>> relaxations one rather tends to reduce it...
>>
>> Convergence for large systems (especially inhomogeneous metallic ones)
>> may be slow. mixing_mode='local-TF' usually helps in these cases. are
>> you using it ?
>>
>> Even when the first scf convergence is painfully slow the subsequent
>> ones should be better as potential and wfcs are extapolated from
>> previous steps. check that you are indeed using these extrapolations
>> (should however be the default for relaxation, so probably you do).
>>
>> Sometime when a system does not want to converge it's trying to tell
>> you something ... Is it a magnetic or metallic system ? are you using
>> smearing ? are you sure the number of bands and your k-point sampling
>> is accurate enough ?
>>
>> hope this helps,
>>   stefano
>>
>> Stefano de Gironcoli - SISSA and DEMOCRITOS
>>
>>
>> Quoting artee sharma <mailtoartee at gmail.com>:
>>
>> > Dear Stefano.
>> >    thanks for reply. my conv_thr in input file is 1.0e-5? i have a large
>>
>> > system, takes too much time to convarage, so kept it low. Shall i
>> decrease
>> > it by further one order?
>> >   latest output is showing:
>> >   Total force =     0.023049     Total SCF correction =     0.002483
>> > so i think my system has not converged yet, force is high.
>>
>> > regards
>> > artee sharma
>> > Department of physics
>> > Bhavnagar University
>> > India
>> >
>> >
>> >
>> >
>> > On Sat, Mar 21, 2009 at 11:46 AM, Stefano de Gironcoli
>> > <degironc at sissa.it>wrote:
>> >
>> >> I think it might be better to stop it. According to the message the
>> forces
>> >> you are calculating may be not accurate and any further relaxation based
>> on
>> >> them is not likely to improve your geometry.
>> >>
>> >> However, are you sure your system is not yet relaxed enough ?
>> >> What is your conv_thr and your force threshold ?
>> >>
>> >> stefano de Gironcoli - SISSA & DEMOCRITOS
>> >>
>> >> PS: don't forget to  provide your affiliation
>> >>
>> >>
>> >>
>> >>
>> >>
>> >> Quoting artee sharma <mailtoartee at gmail.com>:
>> >> **
>> >>
>> >>
>> >> ----------------------------------------------------------------
>> >>  SISSA Webmail https://webmail.sissa.it/
>> >>  Powered by Horde http://www.horde.org/
>> >>
>> >>
>> >>
>> >
>>
>>
>>
>> ----------------------------------------------------------------
>>    SISSA Webmail https://webmail.sissa.it/
>>    Powered by Horde http://www.horde.org/
>>
>>
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://www.democritos.it/mailman/listinfo/pw_forum
>>
>



----------------------------------------------------------------
   SISSA Webmail https://webmail.sissa.it/
   Powered by Horde http://www.horde.org/




More information about the Pw_forum mailing list