The ABINIT Users Mailing List ( CLOSED )

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

Hi,

Test files and tutorials do not mandatorily have the same goals as
users. In fact most don't. For test files, the purpose is mostly to
use quickly every routines of Abinit; for tutorials the goal is to
demonstrate roughly and quickly the capabilities of Abinit; while
users usually seek for converged calculation at all cost. I think this
is piece of an answer to your question.

The other part is that in simple cases most quantities undergo a
coordinated convergence. That is, you get very small change in
density/potential, and wavefunctions and energy and forces ... at the
same time. Yet in pathological case it may happen that one, or two, or
three, or... of those quantities keep stable (seemingly converged)
while the other don't. The safest criterion for converged GS
calculation is indeed tolvrs. Yet it is not even mathematicaly correct
and safe. You may build DFT problems where all the quantities
monitored for convergence (especialy the energy and the forces)
converge to an arbitrary degree of precision while the problem is
still unconverged.

To sum up : tolvrs is indeed a very good convergence criterion for GS.
It is the very best that I know about. Yet it is not perfect. For
safety you must check by hand that other quantities converge and
ASSUME that the convergence of all the monitored quantities means that
your calculation has reached the GS. So far, it has been true for
every systems dealt with by Abinit.

regards

PMA


On Mon, Jun 8, 2009 at 2:34 AM, <6671011@163.com> wrote:
>
> Personally, I doubt that the GS must use tolvs and the RF tolwfr. In fact, 
> some
> other programe, like CASTEP, can control only the toldfe.  (But maybe what
> Matthieu suggested is safer.)
>
> I guess people who prepared the test files also have different 
> understandings
> on this issue. This explains the inconsistency you met.
>
> A  secure strategy is doing you own tests.
>
>
> Guangfu Luo
>
>
> -----------------------------------------------
> Dear Matthieu,
>
> I am sorry of my later response and ambiguous expression. I have fully
> read netiquette. So,
>
>> you attach an abinit input file to calculate phonons (this is fine).
>> The electron phonon calculation is done later in anaddb.
> The file I attached is derived from telphon_1.in, in order to
> calculation el-phonon interactions and superconducting temperature.
> Note that, the first dataset in it is of a ground-state calculation
> with criterion tolwfr 1.0d-20(suggested by tutorial lesson Elphon).
>
> But in tutorial RF2, also the FIRST dataset of input file trf2_1.in,
> similar ground-state calculation uses the criterion tolvrs 1.0d-18.
>
> Both ground-state scf calculation is to generate a wave function, but
> why the criteria are totally different?
> May be the Elphon calculation needs a relevant more preciser wave
> function in order to compute e-DOS, but in tutorial lesson three,
> t35.in the tolerant criterion is toldfe 1.0d-6. I am really confused.
> What is the insight within it?
>
> Thank you!
>
> Best regards
>
>
>
> --
> B.W. Dong
> State Key Laboratory of Superhard Materials,
> Jilin University, Changchun, China. 130012
>
> tq02ksu@gmail.com [ dbw07@mails.jlu.edu.cn ]
>



-- 
Pierre-Matthieu Anglade