The ABINIT Users Mailing List ( CLOSED )

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

Hi,

It looks like your calculation are not converged at all with nstep=20.
It is mandatory for you to increase this parameter. Even the total
energy at step 20 is way before SCF convergency!

I would recommand you to follow the following steps:
1) Do a systematic absolute convergency studie with respect to ecut and kpt
  At present your kptrlen is about 20. This means that your total
energies can't be absolutely converged  down to one mHa. This may be
suitable/acceptable for the starting point of the calulation but not
at all at the end where you expect a convergency of forces better that
5e-5.
2) start your computation wit a first dataset where your parameters
will allow fast caculations.  Don't forget to use "getwfk -1" in your
input file This will spare you lot of caclulations.
3) continue your calculations with a large enough kpt mesh (see
"prtkpt")  and a suitable ecut (absolute convergence down to at most 
0.1mHa seems a minimum).

Following this step, remember that forces are derivatives of the total
energy. Therefor scf convergence of forces is always worst than scf
convergence on total energy. Getting a 5e-6 scf convergence on your
forces at a step of the scf cycle require typically a scf convergency
to about 1e-12 or better on the total energy.
In the output you've send so far the SCF convergency is almost never
better than 1e-3 making the results totaly worthless: your forces and
stress are almost random.

Regards

PMA