hi

[hpaudel]

I trying to optimize a slab geometry of TiO2 with around 200 atoms. I am sure It gonna take long time. Could you help out please for fastest and safest way of optimizing this geometry?

[forsdan]

Can you just elaborate on what do you intend to do? Are you looking at interfaces between two TiO2 slabs or are you investigating a TiO2 surface and want to calculate surface energies and/or adsorption energies for different atoms/molecules? How many layers do you use in the slab(s)? The details affects how the most efficient relaxation could be performed.

/Dan

[panda]

yes, details please! Can't help with such little information!

btw what do you mean by "safest"? regardless of how you perform a calculation it will be safe, we aren't doing labwork here

[hpaudel]

I have five layers in my slab. The size of supercell is 22x22x14 (I am not sure 5 layers are enough to converge the total energy). I am trying to calculate DOS and band structure of my system (hopefully first I need to have optimized geometry for that). My final aim is to get dipole matrix elements for the nano structure size TiO2 geometry that would requires the wavefunctions of the discrete states within the system. It gonna take long time for such a large system. So I am trying to construct such slab and look the dipole matrix elements for stable geometry.

[hpaudel]

hi, here is my INCAR file for TiO2 slab optimization.

SYSTEM = TiO2
ISTART = 0
ICHARG = 2

LREAL = .FALSE.

ISPIN = 1
EDIFF = 0.0001
EDIFFG = -0.01
ISIF = 3
ENCUT = 500
PREC = Accurate

LWAVE = .TRUE.
LCHARG = .TRUE.
NELM = 200 # Maximum number of electronic iteration
NSW = 0 # maximum number of ionic iteration
#NGX=26
#NGY=26
#NGZ=42

IBRION = 2

ISIF = 2 # stress and relaxation
POTIM = 0.10 # time step for ionic-motion
TEIN = 0.0 # initial temperature
TEBEG = 0.0
TEEND = 0.0 # temperature during run
ISMEAR = -5 # 0 gaussian smearing
# -5 tetrahedron method with Blöchl corrections
SIGMA = 0.2 # Width of smearing

ISYM = 0

[panda]

I think 200 atoms in five layers is sufficient for your calculation, but the best way to check is to of course check your final energy value in eV/atom with the bulk value calculated for a periodic system. What type of cluster are you running the calculations on, and how many processors are you using? This will also have an effect on how long it will take the calculation to converge. Why do you have ISIF = 3 and then ISIF = 2 below that? If this is the INCAR for a single calculation you should only have that once.

[hpaudel]

It was mistake, ISIF written twice. Thank you for that!!

About the cluster:
It is High Performance Computing IBM Stokes Cluster. The job runs in the cluster with 4 nodes with 8 processors in each nodes. Each processor has 2000M. So altogether 32 processor of each 2000M.

[panda]

No problem, I typically use ISIF = 3 for bulk calculation. Do you use ISIF = 2 for slab? that sounds like the specs of one of the clusters I run similar jobs on, and my calculations take between one and two weeks. There are some things you can do to speed things up, like setting the tag IALGO = 48, but I would prefer in the case of the slab calculation to allow it to take as long as necessary with PREC = Accurate as you have done, because you want your results to be as good as possible!

[alex]

Where does the 14 in 22x22x14 come from? If you count in 'layers' of -O_(1/2)-TiO-O_(1/2) you'll end up with 15 layers. Or you just count TiO2 layers and then it's 5 ...

Cheers,

alex

[forsdan]

@ alex: I presume 22x22x14 refers to the supercell dimensions given in Angstroms. 14 Angstroms in the lateral direction seems reasonable for 5 layers of rutile together with a vacuum region.

@ panda, hpaudel: ISIF = 2 should be used for the slab calculations, since you would like to simulate a semi-infinite crystal that has the same in-plane lattice parameters as the bulk phase.

@ hpaudel: I can't see any reason why you shouldn't simple be able to perform a direct minimization with IBRION = 2 and an appropriate POTIM value. It could be helpful from a speed point of view to first do a rough relaxation with PREC = Normal and then switch to PREC = Accurate the last steps. As panda also points out the time for the electronic convergence can also be improved with IALGO = 48, although IALGO = 38 is more stable.

Cheers,
/Dan




<span class='smallblacktext'>[ Edited Wed May 12 2010, 01:04PM ]</span>

[hpaudel]

@alex: 22x22x14 is the dimensions of the supercell in angstroms.


I got a problem. My job for optimization of TiO2 slab geometry with 5 layers did not converge. I used the following INCAR file, with little changed in previous INCAR file.

ISTART = 0
ICHARG = 2
LREAL=.TRUE.
ISPIN = 1
EDIFF = 0.0001
EDIFFG = -0.01
ENCUT = 500
PREC = Normal
LWAVE = .TRUE.
LCHARG = .TRUE.
NELM = 200
NSW = 0

#NGX=26
#NGY=26
#NGZ=42

IBRION = 2
ISIF = 3
IALGO = 48
POTIM = 0.10
TEIN = 0.0
TEBEG = 0.0
TEEND = 0.0

ISMEAR = 0
SIGMA = 0.2

ISYM = 0

Could you please give me some suggestions?

[sharma]

can you send me the email to me ? bharatsharma61@hotmail.com I want to discuss with you about VASP.

[admin]

as Dan pointed out above, for slabs with a vacuum NEVER use ISIF=3, or do you want the slab/vacuum to collapse? there will not be a 'stable' cell dimension along the axis which includes a vacuum.