> On 08/04/2011 02:17 AM, [email protected] wrote:
>> 2. You can eventually extract total energies, make a plot and
>> look at it, but attention! It is IMPOSSIBLE to decide whether the energy
>> is "converged" or not without specifying the relevant order of magnitude
>> of DIFFERENCES you are interested in to study, like e.g. the effect of
>> atom displacement, changing magnetic configuration, imposing pressure,
>> etc.
>
> For total energies, this can be controlled by
> DM.Require.Energy.Convergence and DM.Require.Energy.Convergence.  But in
> fact, in default case, the siesta don't use this convergence criteria.
> Regards
> --
> Hongyi Zhao <[email protected]>

This seems to be a misunderstanding. The convergence entries you refer to,
along with some others, determine when the program converges
after a certain number of iterations and stops
FOR THE GIVEN PARAMETERS as MeshCutoff, k-mesh density, basis size, etc.
This does not give you any hint as to whether these parameters
are "good". Whatever was referred to before (and in the mailing list,
and in the tutorials) in the present context as convergence with respect
to these calculation parameters is different.
What was meant is the following:
you take some MeshCutoff, converge the calculation,
write down the value. You change MeshCutoff, you converge the calculation,
write down the value. Then you have a look at a sequence of converged
values you wrote down and try to figure out which value of MeshCutoff
is sufficient, or "converged", in a sense that its further changing
does not affect the result. The problem hereby: you cannot decide on this
unless you know IN ADVANCE which accuracy you in fact need, i.e.,
meaningful energy DIFFERENCES for your problem.
This was the meaning of my comment.

Best regards

Andrei Postnikov


Responder a