Convergence problem for slab geometry

[hpaudel]

Hi Vasp Users,

I got a problem in convergence for slab geometry of TiO2. I tried to calculate with minimum accuracy, it still did not work. Actually I am trying to optimize it and calculate wavefunction. My slab has 5 layers with 156 atoms. The size of supercell is 21x21x13 in Angstroms. For up to 4 layers jobs were successfully completed. I have put a 15 Angstroms in each direction for shielding the slab from adjacent slabs. Following is my INCAR file.

SYSTEM = rutile
ISTART = 0 # 0 start fr. scratch, no wavefunction WAVECAR need
ICHARG = 2 # 2 start fr. scratch,
# 1 read from CHGCAR,
# 0 Calculate from WAVE function
# 11 read charge density from CHGCAR and keep constant during the subsequent run
LREAL=.TRUE. # reciprocal space projection scheme .FALSE.
# automatic real space projection scheme AUTO?
ISPIN = 1 # non spin polarized calculations 1
# spin polarized calculations 2
EDIFF = 0.0001 # Convergency of ELECTRONIC iteration
EDIFFG = -0.01 # Convergency of IONIC iteration
# "-" mean the condition for FORCE
ENCUT = 500 # Energy cuttoff, need to test your value
LWAVE = .TRUE. # Save wavefunction WAVECAR or not
LCHARG = .TRUE. # Save CHGCAR charge density
NELM = 200 # Maximum number of electronic iteration
NSW = 0 # maximum number of ionic iteration
# 0 or 1 mean there is no relaxation (fix everything)
#NGX=26
#NGY=26
#NGZ=42
IBRION = 2 # -1 Ions won't move
# 2 Do relaxation, ions will move after every ionic iteration
# 0 Molecular dynamics
ISIF = 3 # stress and relaxation
IALGO = 48
POTIM = 0.10 # time step for ionic-motion
TEIN = 0.0 # initial temperature
TEBEG = 0.0
TEEND = 0.0 # temperature during run
ISMEAR = 0 # 0 gaussian smearing
# -5 tetrahedron method with Blöchl corrections
SIGMA = 0.2 # Width of smearing
ISYM = 0

Thank you!!

[Danny]

you might be doing MD (TEIN, TEBEG and TEEND parameters)

Maybe a general hint: Don't try to use all parameters in the INCAR file, there is a chance that they conflict with what you intend to do

[forsdan]

Can you clarify what doesn't work. Do you mean the electronic or ionic convergence? If you referring to the former, the convergence may be improved by playing with the mixing parameters, e.g. you can try a linear mixing (see page 21-22 in http://cms.mpi.univie.ac.at/vasp-worksh ... ectron.pdf ). Also try with IALGO = 38, or ALGO = Fast instead of IALGO = 48.

For ionic convergence you must first of set NSW > 0. In addition, as I wrote in your earlier thread, ISIF should be set to 2.

Furthermore, I don't understand your statement of 15 Angstroms vacuum in each directions. It doesn't fit in your given supercell size if you have a 156 atom slab.

Cheers,
/Dan

[hpaudel]

@forsdan:
The problem is that the job is not completed, it stops after few iterations or sometimes it stops even without doing a step of iteration. About the supercell, the size is 22x22x22 angstroms. 15 angstrom is vacuum thickness.

Thank you.

[forsdan]

I will assume you are talking about the electronic relaxation. The standard procedure is then usually:

1. Check your memory consumption. Are you using more than the available memory? If so, increase the number of nodes or reduce your settings.

2. Check the stdout and stderr files to see if any error messages are given.

3. If no solution can be found with 1 and 2. Add debug-tags to vasp and run it through a suitable debugger.

Cheers,
/Dan

<span class='smallblacktext'>[ Edited Sun Jun 13 2010, 11:06PM ]</span>

[hpaudel]

Thank you, forsdan

[admin]

please do NOT use ISIF=3 for the geometry convergence of a slab calculation.
It does not make sens to relax the volume of a cell if the cell contains a vacuum space