[Pw_forum] structural relaxation for a M-M-M system

[Nicola Marzari]

Dear Sen,


here are a few suggestions, from the more general ones to ones
specific to your system.

1) list your affiliation. PWSCF has a long history of misbehaving
with anonymous users.

2) when things do not work, always try a smaller and simple system,
and try to figure out one problem at a time.

3) even if in your case it won't make a difference, I would feel
more comfortable if you used a slightly larger smearing (0.025 Ry),
"m-v" smearing, and one order of magnitude tighter threshold on the
electronic convergence.

4) with these parameters in hand, and with a fixed unit cell, let the 
code run for a few atomic iterations, starting from the coordinates
that were giving you

>      Total force =     0.095460     Total SCF correction =     0.000604
>      Total force =     0.082488     Total SCF correction =     0.000306
>      Total force =     0.046922     Total SCF correction =     0.000911

[...] etc...

Does the total energy goes down ? (note that the code is calling total 
energy your variational quantity in the presence of a fictitious 
smearing temperature, i.e. E-TS).

If it does, things are moving in the right direction, and the ions are 
being relaxed. Note that since you have always a certain amount of noise 
in the forces (due to imperfect convergence), and because the total free
energy as a function of ionic positions is not perfectly quadratic, it
would probably be wise to do only 20 to 50 ionic iterations, and then 
restart the problem from scratch with the new relaxed ionic coordinates.

If the total energy goes up even at the very first steps, the reasons 
could be two:

1) your forces have too much noise (unlikely, given your small "Total
SCF correction"). Note that you can always specify a desired threshold 
for the convergence of forces.

2) the step that pwscf takes in the search direction for ionic 
minimization is too large (or too small), and thus it is difficult to 
guess correctly the minimum along the search direction. Change the trust
radius, either the min or the max ones (assuming you are doing relax in
the default bfgs mode). All the parameters are described in INPUT_PW, in
the Doc directory.

3) switch to a more robust but less efficient ionic minimization - the
other option in pw is damped dyanmics

4) once you know how to relax the ions, repeat the calculations with 
different celldm(1), and find the one that gives you the minimum in 
energy as a function of celldm(1). If your energy cutoff is high enough,
that should also give a trace of the stress tensor that is roughly
equal to zero.

5) try hard, before asking again for help on the list. In this field, 
you learn by doing more calculations, and thinking what could have gone 
wrong. Once you have solved the problem, post a short message that
summarizes what made a difference and solved the problem.

Best luck,

			nicola

---------------------------------------------------------------------
Prof Nicola Marzari   Department of Materials Science and Engineering
13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu