The ABINIT Users Mailing List ( CLOSED )

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

> If so, then for most systems this is simply not possible. Then my question 
> is:
> Does anybody know what the effect is on the dynamics of a system of working
> with non-converged forces??????

Every one doing molecular dynamic knows that MD behavior is chaotics.
So Whatever your level of convergence you'll never get the kind of
"exact" dynamic you seem to which. There is at least one criterion
that would avoid you using the 10^-9 toldff: energy consevation. As
long as your energy keeps being conserved you can consider that your
forces are converged "enough". Enough is the important word here. Any
way even if you were able to get absolute convergency of forces to
10^-9 you wouldn't get an exact MD since dft is just an approximation
with it's own error bars...

Note that forces of the order 10^-6 Ha/bohr are already pretty small
meaning that atoms are quite close to equilibrium position. If the max
forces of your MD run have this order of magnitude then this is almost
0 Kelvin MD. Most forces that begins to matter are > 10^-4. So don't
worry about your almost 0 forces.

> Following this, if the forces on my system are of the order of 10-6 Ha/Bohr
> then I need at least a toldff, say 10-9 Ha/Bohr, for a given value of ecut
> for which energy is converged, to avoid seeing my forces changing when
> increasing ecut. Is this what you meant ????

You can at least consider that an absolute convergence of forces to
10-6 is far more than enough for forces of about 10-6 Ha/Bohr because
you'll get some error cancellations. And having 10-7 Ha/Bohr toldff
should already be OK.

--
Pierre-Matthieu Anglade