[ase-users] VASP in ASE

[Janne Blomqvist]

On Wed, Jan 05, 2011 at 11:06:58AM +0100, Marcin Dulak wrote:
> Hi,
> 
> if fact vasp uses non-recommended potentials for more elements, so it's good
> to set the 'setups' parameter for vasp calculator to the "most accurate 
> ones":
>         parameters['setups'] = {
>             'B':'_h', 'C':'_h', 'N':'_h', 'O':'_h', 'F':'_h',
>             'H':'_h',
>             'Li': '_sv', 'Be': '_sv',
>             'Na': '_sv', 'Mg': '_pv',
>             'K': '_sv', 'Ca': '_sv',
>             'Rb': '_sv', 'Sr': '_sv',
>             'Cs': '_sv', 'Ba': '_sv',
>             'Sc': '_sv',
>             'Ti': '_pv', 'V': '_pv', 'Cr': '_pv', 'Mn': '_pv',
>             'Fe': '_pv', 'Ni': '_pv', 'Cu': '_pv',
>             'Y': '_sv',
>             'Zr': '_sv', 'Nb': '_pv', 'Mo': '_pv', 'Tc': '_pv',
>             'Ru': '_pv', 'Rh': '_pv', 'Pd': '_pv',
>             'Hf': '_pv', 'Ta': '_pv', 'W': '_pv', 'Re': '_pv',
>             'Os': '_pv',
>             'Al': '_h',  'Si': '_h', 'P': '_h', 'S': '_h', 'Cl': '_h',
>             'Ga': '_h', 'Ge': '_h',
>             'In': '_d', 'Sn': '_d',
>             'Tl': '_d', 'Pb': '_d', 'Bi': '_d',
>             }
> The page http://cms.mpi.univie.ac.at/vasp/vasp/_elements.html says:
> " Since computers are becoming ever more powerfull, we recommend to use 
> the more accurate potentials whenever possible."

Well, with some reservations, I think:

- the Al_h potential causes ghost states to appear and should not be
  used. 

- If one wants to calculate with a smaller cutoff energy like 400 eV,
  hard potentials such as C_h and O_h with ENMAX=500 might give worse
  results than the standard potentials (in the case of C and O, ENMAX
  = 400 eV)?

> I would recommend also to use:
>         parameters['lscalapack'] = False # sometimes scalapack fails
>         parameters['ialgo'] = 48 # otherwise it fails sometimes for 
> light elements

IALGO = 48 (RMM-DIIS) sometimes gets stuck when starting from scratch
or when the ionic positions have been updated; ALGO='Fast' is a good
compromise, it does the initial iteration(s) with DAV and then
switches to RMM-DIIS.

-- 
Janne Blomqvist