[Pw_forum] conv_thr for forces for relaxing ions

Madhura Marathe madhura at jncasr.ac.in
Wed Apr 22 11:15:12 CEST 2009

 Dear Stefano,

 Thanks for the quick reply. This was really necessary to confirm since
the calculations are expensive. And I am more interested in the ground
state structure, may not use the actual value of the force for further


> In PWscf the forces are calculated  from the  Hellman-Feynman theorem
> (strictly true only at the stationary point) plus a correction that
> (approximately) accounts for the residual lack of self-consistency.  As
> such when the correction is large compared to the Hellman-Feynman term
> one should be careful.
> In your case I would say that the forces are  now probably very small
> and  therefore the structure is reliable.
> I would say that the actual (small) value for the force are probably
> correct within a few percent since the correction term is (according to
> the value quote) of the order of 10 % of the total (and hopefully is not
> completely wrong).
> Hope this helps,
> stefano de Gironcoli -SISSA and DEMOCRITOS
> Madhura Marathe wrote:
>>  Dear all,
>>  During one of the ionic relaxation calculations, I got the error
>> message,
>> "SCF correction compared to forces is too large, reduce conv_thr". I am
>> using conv_thr = 1.0d-8 which is generally sufficient, so I increased
>> the
>> parameter upscale (from 10.0 to 100.0) in &IONS to reduce the the
>> threshold during relaxation. This lead to the convergence till the last
>> ionic iteration which was converged to sufficient accuracy, when again
>> the
>> same error message occurred. The forces then are
>>      Total force =     0.000136     Total SCF correction =     0.000016
>>      SCF correction compared to forces is too large, reduce conv_thr
>>  There has been a recent discussion on the topic. From that, I gathered
>> that the subsequent relaxations after this error message are not
>> reliable.
>>  So my question is whether the forces are reliable in this last
>> iteration?
>> Or do I need to further reduce the conv_thr and re-run the whole
>> calculation?
>>  Thanks for the help,
>>  Madhura.
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum

Madhura Marathe,
PhD student, TSU,
JNCASR, Bangalore.
Phone No: +91-80-22082835

More information about the users mailing list