Correct KPOINT mesh for Relaxation

[cameronu]

Hi,

I am running a relaxation using ISIF = 3, but am finding that as I increase the KPOINT mesh using the Monkhorst Pack type, this does not converge in the CONTCAR output.

What mesh type and number of kpoints should I be using for this type of calculation?

See INCAR_REL file below:

'SYSTEM = MASn(2+)I3
!Start Parameters
ISTART = 0 (job:0-new 1-cont 2-samecut)
!ICHARG = 2 (Charge:1-file 2-atom 10-const)
!INIWAV = 1 (Random initial wavefunction,0-lowe)

!Fixing Sn4+
!NELECT = 48

!Parallel Options
!!LPLANE = .TRUE. (Real space distribution)
NPAR = 80 (Max is no. nodes)

!Electronic Relaxation
PREC = Single (Precision level for FFT grid)
LREAL = Auto (Projection operators: automatic)
LASPH = .TRUE.
ALGO = Normal (Electronic minimisation algorithm: 38/48)
ENMAX = 600.00 eV (Plane-wave cutoff)
NELM = 100 (Max number of SCF steps)
NELMIN = 5 (Min number of SCF steps)
EDIFF = 1E-10 (SCF convergence criteria)
POTIM = 0.1
GGA = PE (PBE)

!NBANDS =400
SAXIS = 0 0 1
LORBIT = 11
NEDOS = 500

# MD (do little writing to save disc space)
NSW = 200 ; NWRITE = 0 ; LCHARG = .FALSE. ; LWAVE = .FALSE.
TEBEG = 300 ; TEEND = 300
IBRION = 2
ISIF = 3

!Van Der Waals
LVDW=.TRUE.

# canonic (Nose) MD with XDATCAR updated every 50 steps
# SMASS = 3 ; NBLOCK = 50 ; POTIM = 1.5

# micro canonical MD with temperature scaling every 50 steps
# good for equlibration but usually better to use Nose thermostat
#SMASS = -1 ; NBLOCK = 50 ; POTIM = 1.5

[admin]

ISIF=3 not always gives reliable results.
Much safer approach is ISIF=2
using the shape of the cell as determined
by crystallographic methods

[guiyang_huang1]

ISIF=3 may be not working correctly sometimes due to the compilations or some other reasons.

For example, the gpu vasp does not work for ISIF=3 and HSE06.

ISIF=4 and changing the lattice constant can be tried.