IBRION=2

[fabella]

Hello,
I am using the IBRION=2 method "tag" inorder to relax the forces in a distorted supercell. I have been using different values of NSW (# of ionic steps). How do I know or when can I guess that the NSW value I have used is sufficient ... when my forces are .05 eV/A ? ... will it stop prematurly if small forces have been reached or does it keep going and going like the energizer bunny?

Thanks
W

[fabella]

Ok ... I have run a few test calculations with IBRION=2, I used increasing NSW = 3,5,10, and 20 ... the forces decrease using this method and the output structure for each ionic step is saved in XDATCAR (the final in CONTCAR). If one wants to relax until VASP says the "reached required accuracy - stopping structural energy minimisation" the NSW can usually be set at 20 or so. (?) I found this was within a sufficient range ... it doesn't keep going like energizer bunny.

So, can anyone add their knowledge about this process?

Also, what does the 'total drift' measure ... ?
total drift: 0.011980 0.067707 0.121824

Is this any good ?

[tjf]

You need to read about EDIFFG in the manual: http://cms.mpi.univie.ac.at/vasp/vasp/node98.html . It's pretty simple.

How many steps are required depends entirely on your system and your starting geometry. Unless you're going to run into a queue time limit or somesuch, it's virtually impossible to set NSW too high. (But if you do set it high, keep a weather eye on the calc to make sure it hasn't gone awry.) I habitually set NSW to several hundred.

That drift is high, but again it depends on the system and what you want from it. Your drift should be smaller than the forces you're interested in.

[admin]

concerning the choice of NSW I would suggest the following (if you run on a system which sets time limits to your jobs)
1) do a small test (NSW=10) and have a look into the job.stdout
how much CPU time this requires. If your sysyem is well-behaved, the number of electronic scf steps for one ionic step usually decreases, so you have an estimation for an upper time limit. Then adjust NSW of all further runs to the CPU time you can use in one job.
2) to continue form one job to the next,
--) keep WAVEFUN
--) cp CONTCAR to POSCAR
--) set ISTART = 1 and copy WAVEFUN back to the batch worker before re-starting vasp.

3) concerning the total drift of forces : if this is considerably larger than the forces on the atoms themselves, this may be due to two reasons
--) your FFT grids are too small (--> set PREC=Accurate)
--) if you do molecules in a large box: unless you fix the coordinates of at least one atom, the molecule may start drifting through your unit cell.

[fabella]

I have set PREC=Accurate, run the same calculation.

Here is my "total drift of forces":

total drift: 0.000178 -0.000188 0.003371

...BTW ... how do you fix coordinates in relaxation?

W

[tjf]

Selective dynamics. In POSCAR. Read the manual.