[Pw_forum] Relaxation doesn't converge

[Stefano de Gironcoli]

Dear Pablo,
   the convergence check on stress tests EACH component against the 
threshold
  your stress tensor is not yet converged. sigma_xx and sigma_yy are NOT 
converged
  HTH
   stefano de Gironcoli

Pablo Aguado wrote:
> I'm sorry, but I still can't see the problem ...
>
> forc_conv_thr = 1.0d-3  is satisfied ==> conv_ions = true
>
> Only z is allowed to relax (cell_dofree = 'z') so:
> sigma_xx & sigma_yy should be set to 0 at some point
> sigma = 0.87 Kbar
> press_conv_thr = epsp = 1.0 Kbar
> press = 0.0
>
> so conv_ions.AND.(ABS( sigma(i,i) - press) * uakbar < epsp) is true,
> unless the program is still considering stress_xx and stress_yy
> (cell_dofree is not being taken into account)
>
> Pablo
>
>
> On Wed, Sep 2, 2009 at 1:03 PM, Paolo Giannozzi<giannozz at democritos.it> wrote:
>   
>> Pablo Aguado wrote:
>>
>>     
>>> All I can say is that it seems to be converged
>>>       
>> "it seems" /= "it is"
>>
>>     
>>> press_conv_thr = 1.0d0
>>>       
>> this is internally called "epsp"
>>
>>     
>>>          total   stress  (Ry/bohr**3)                   (kbar)     P=   -0.84
>>>  -0.00001150   0.00000000   0.00000000         -1.69      0.00      0.00
>>>   0.00000000  -0.00001150   0.00000000          0.00     -1.69      0.00
>>>   0.00000000   0.00000000   0.00000591          0.00      0.00      0.87
>>>       
>> and this is where the convergence is checked: PW/vcsmd.f90, line 212
>> and following:
>>
>>      DO i = 1, 3
>>         !
>>         conv_ions = conv_ions .AND. &
>>                     ( ABS( sigma(i,i) - press) * uakbar < epsp )
>>         !
>>         DO j = ( i + 1 ), 3
>>            conv_ions = conv_ions .AND. &
>>                        ( ABS( sigma(i,j) ) * uakbar < epsp )
>>         END DO
>>         !
>>      END DO
>>
>> press=external pressure, sigma=stress tensor,
>> uakbar=atomic units to KBar conversion factor.
>>
>>     
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>