The ABINIT Users Mailing List ( CLOSED )

["Anglade Pierre-Matthieu" <anglade@gmail.com>]

Hi Joe,

The convergence of wavefunctions beeing not self consistent, the first
computation with tolvrs is useful to ensure  self consistent
convergence.
The numbers nlline and nnsclo are tuned so that for most system where
SCF convergence is easy you get scf convergence together to
wavefunction convergence in some kind of optimal way : you don't waste
too much time in optimising the wavefunction of a non self-consistent
potential and do not have sensible SCF convergence problems arising
from the not perfectly converged wave functions. However there is no
algorithm that enforce this equilibrium. It may happens that a bad
convergence of wave function prevent the SCF convergence (rare) and
frequently the SCF convergence problems makes the convergence of
wavefunctions meaning less.
All in all the procedure you describe is likely the safest. Although
for many cases you could achieve the two first step by simply fine
tuning nline and nnsclo or just using the default value.

regards

PMA


On Mon, Nov 3, 2008 at 9:02 PM, D. R. Hamann
<drhamann@mat-simresearch.com> wrote:
> Dear Joe,
>
> Yes.  Thanks for pointing this out.  I should revise the tutorial
> accordingly.
>
> Best regards,
> Don Hamann
>
> Josef Zwanziger wrote:
>>
>> Dear Colleagues,
>>
>> In the very useful part 6 of the elastic constants tutorial, where a metal
>> is
>> treated, it is pointed out that it is necessary to use tolwfr as the
>> convergence parameter to ensure that the empty bands are also converged
>> (as
>> opposed to using tolvrs, for example). The suggested procedure is 1) GS
>> run
>> with tolvrs, followed by 2) non-self-consistent run with tolwfr, using the
>> density from 1) as input, to generate highly converged wave functions.
>> However,
>> the tutorial file uses iscf 5 throughout, thus, a self-consistent
>> calculation
>> is used also in the tolwfr part. So, what's the best approach? I'm
>> assuming
>> that the point of the initial tolvrs scf calculation is simply to speed
>> the
>> second, highly converged calculation, since the output of the first
>> calculation
>> is not used anywhere else. Should the general approach actually be
>> something
>> like:
>>
>> 1) iscf 5, tolvrs 1.0D-18, prtden 1
>>
>> 2) iscf -2, tolwfr 1.0D-22, getden 1, prtwf 1
>>
>> 3) iscf 5, tolvrs 1.0D-12, rfphon 1, getwfk 2
>>
>> thanks for your help,
>>
>> Joe Zwanziger
>>
>>
>
> --
> D. R. Hamann
> Mat-Sim Research LLC    | Deptartment of Physics
> P.O. Box 742            |  and Astronomy
> Murray Hill, NJ 07974   | Rutgers University
> phone: 908-370-8079     | 732-445-4381
>
> email: drhamann@mat-simresearch.com
>
>
>



-- 
Pierre-Matthieu Anglade