hcp Mg and Monkhorst Pack grids

[vasp_user]

For hcp Mg, I am getting convergence using the Monkhorst Pack scheme with a 12x12x12 grid (total cpu time 28.553 seconds)-- and convergence with the Gamma scheme is not coming until I use a 20x20x20 grid (total cpu time 115.129 seconds).

Since I just noticed the hexagonal lattices subsection under the KPOINTS file section of the VASP guide that can be found at http://cms.mpi.univie.ac.at/vasp/vasp/h ... tices.html, I would like to ask: are my results inaccurate, or is there something I should have done differently? Setting aside the issue of the general wisdom of using Gamma centered grids for hexagonal lattices, is it appropriate to use the Monkhorst Pack grid in this particular case given the results I showed above and the procedure I describe below?

I have included my POSCAR and my INCAR (note GGA = 91) files below. (If you are going to look at the INCAR file, please note that I left some settings as they were when I was allowing the ions to move in previous work? but, for the current KPOINT convergence tests, NSW is set to zero)

Regarding my convention for determining convergence: I found the 30x30x30 grids of Monkdorst Pack and Gamma to have an average energy of -3.0436. I took the absolute value of the difference between the energy of the run with each grid tested, and -3.0436. I defined convergence to be when increasing the size of the mesh no longer produced a difference as defined above [with the absolute value operation] that was larger than 0.002. For the Gamma runs I tested {3x3x3, 4x4x4, 5x5x5, 6x6x6, ?, 30x30x30}; for the Monkhorst-Pack runs I tested {4x4x4, 6x6x6, 8x8x8, 10x10x10, ?., 30x30x30}.

POSCAR

hcp bulk Mg
1
1.6 -2.77 0
1.6 2.77 0
0 0 5.2
2
direct
0.33333 0.66667 0.25
0.66667 0.33333 0.75

INCAR

Title of calculation
SYSTEM = Mg hcp bulk
GGA = 91
# GGA = PE
VOSKOWN = 1

Starting parameters
ISTART = 1 : 0-new 1-cont 2-samecut
ICHARG = 1 : 1-file 2-atom 10-const
INIWAV = 1 : 0-lowe 1-rand

Electronic minimisation
# ENCUT = 400 eV : energy cut-off for the calculation (optional)
# ENAUG = 800 eV : energy cut-off for the calculation (optional)
# PREC = Normal
# PREC = Accurate
# PREC = High
IALGO = 48 : 38-Davidson; 48-RMM-DIIS (8)
NPAR = 2
NSIM = 8
LPLANE = .FALSE. :
LREAL = .FALSE. : projection done in real space (.FALSE.)
# ROPT = ${ROPT} ${ROPT} ${ROPT} ${ROPT} ${ROPT} ${ROPT}
# NELMDL = -5 : delay steps in charge density evaluation (-5)
NELM = 100 : maxium steps of electronic minimisation (60)
NELMIN = 3 : minium steps of electronic minimisation (2)
# EDIFF = 1.0E-03 : criterion for energy convergency in eV
EDIFF = 1.0E-04 : criterion for energy convergency in eV

# ISYM = 0 : the default ISYM turns on symmetry
ISPIN = 1 : ISPIN=1 is the default and it is for non-spin polarized calculations
# AMIX = 0.2 ; BMIX = 0.0001
# MAGMOM =
# NUPDOWN = 4
# NELECT = 145

DOS related values
ISMEAR = 1 : -5-TetraB; -1-Fermi; 0-Gauss; 1..N-MP
SIGMA = 0.175 : width of smearing in eV
LCHARG = .FALSE.
LVTOT = .FALSE.
LWAVE = .FALSE.
# LDIPOL = .TRUE.
# IDIPOL = 3
# NEDOS = 301
# RWIGS = ${RMg} ${RH}
# LORBIT = 1


Ionic relaxation
NWRITE = 1
NSW = 0 : maxium steps of ionic relaxation (0)
IBRION = 2 : 0-MD; 1-quasi-Newton; 2-CG; 3-damped
ISIF = 0 : relax cell shape or volume or internal par
POTIM = 0.5 : time step in fs for MD; scaling otherwise (0.5)
EDIFFG = -0.02 : Positive-energy; Negative-force criterion

#Molecular Dynamics
# ISYM = 0
# SMASS = 2.0 : Nose mass parameter
# POTIM = ${POTIM} : time step in fs for MD; scaling otherwise (0.5)
# TEBEG = ${TEBEG} : Temperature
## TEEND = : Temperature
## APACO = 16.0 : distance for Pair correlation in Angstrom
## NPACO = 256 : slots for Pair correlation
## NBLOCK = 1
## KBLOCK =

[vasp_user]

In the above post ". . ." was converted to "?"

So please imagine ". . ." everywhere you see an out of place "?"

[tlchan]

[quote="vasp_user"]
For the Gamma runs I tested {3x3x3, 4x4x4, 5x5x5, 6x6x6,..., 30x30x30}.
[/quote]
In general, the convergence of even grid and odd grid are different. Maybe, you should consider the convergence of even and odd grids separately.