next up previous contents
Next: 11.7 Relaxation Up: 11 Examples Previous: 11.5 Dimers

11.6 Molecular - Dynamics


About which files do you have to worry:
Use the makeparam utility to create the file. For a molecular dynamics PREC= Low is definitely sufficient. The INCAR file might be similar to
   PREC = Low
   ENCUT  = 150 eV   energy cutoff (opt)
   IALGO = 48 ; NELMIN = 3
   LREAL  = .TRUE.   real space projections
   BMIX   = 2.0      mixing parameter
   ISYM   = 0        switch of symmetry
   EDIFF  = 1E-3

 Ionic Relaxation
   NSW    =    100    number of steps for IOM
   NBLOCK = 1 ; KBLOCK =   100
   SMASS  =    2.0    Nose mass-parameter (am)
   POTIM  =   3.00    time-step for ion-motion
   TEBEG  =   573     temperature

   APACO  = 10.0     distance for P.C.

 DOS related values:
   ISMEAR =    1  ; SIGMA = 0.1
   EMIN   =   -20 ;  EMAX = 5
Use IALGO=48 (i.e. residual minimization) for large molecular dynamic runs. You should switch on the real space projection (LREAL=.TRUE.), and require at least 3 electronic iterations per ionic step (NELMIN=3).

Special consideration requires the parameter BMIX, usually the (quite heuristic) formula

gives a reasonable estimation for the optimal BMIX parameter. The remaining parameters can be looked up in the INCAR section (Sec. 7).

After doing one iteration it is only necessary to copy CONTCAR to POSCAR and restart VASP. Usually a shell script is used for this task. An example shell script can be found on the vamp account in the file  vamp/scripts/iter.

If you are using time steps which are large in comparison to the vibrational frequencies you should use the Verlet algorithm instead of the predictor corrector algorithm. Normally the predictor corrector algorithm improves the energy stability, but because it is an "high order" algorithm it is also less stable for large time steps. For instance we found that the Verlet algorithm is superior for H tex2html_wrap_inline5141 O if tex2html_wrap_inline5151 fs (this means that one period for bond stretching takes approximately 10 time steps, the drift decreases by almost a factor of 10 with the Verlet algorithm). To use the Verlet algorithm change step.o in the makefile to stepver.o. As a rule of thumb always use the Verlet algorithm of you see strong oscillations in the total energy!

Mon Mar 29 10:38:29 MEST 1999