Why these long or more atom structures doesn't converge?

[SKM]

Hi Admin,
A similar issue faced before, but that topic is locked, so posting here new.
i have two nano crystals linked with ligands and i put vacuum, over one side to simulate the experimental situation.
it has around 350 atoms. I could never converge the geometry properly,. the energy of system is getting minimised continuously (numerically), but if i see the CONTCAR, the atoms are distorted a lot and thus subsequent continuing runs, keep showing lower energy numerically, but because the atoms are distorted, so it keeps trying for minimum potential space and showing some lower energy always i suppose.
How to optimise such structures actually?
INCAR is as below:

PREC = Normal
NPAR=30
IBRION = 2
ISIF = 2
ENCUT = 450
EDIFF = 1.0e-05
EDIFFG = 0.025
IVDW = 11
ISMEAR = -1; SIGMA = 0.2
POTIM = 0.1
IALGO = 38
LREAL = Auto
LWAVE = .FALSE.
LCHARG = .FALSE.
ADDGRID = .TRUE.
ISYM = 2
#SYMPREC = 1e-05
AMIX = 0.01
BMIX = 0.0001
NSW=500
NELMDL = -3
NELM = 220
~

[martin.schlipf]

It is hard to give a general answer to this question. I would approach the problem in the following way

• Simulate the molecule and the bulk independently first. This gives you some idea how to simulate these systems. If they do not work as independent systems it will be even harder to do them in contact.
• Then setup the system you want to simulate. Carefully examine the atomic distances at the surface visually and in the neighbor list output. Make sure that all the distances make sense.
• Initially just relax a few atoms close to the interface as to limit the degrees of freedom. This should make it easier to converge. Take care that all electronic scf finish otherwise you may get tainted forces that can screw up the structure.

[alex]

Hi SKM,

the way I see it is:
- you are facing van der Waals forces in the middle
- for that you missing the dispersion correction
- your EDIFFG = 0.025 is really , make it sth. like 1.e-5
- also adjust EDIFF !

Hth,

alex

[SKM]

Thank you @martin , @ alex, for the inputs.

1. the individual molecule of the ligand was converged.
2. the bulk perovskite is not done independently, but the unit cell from which the bulk was created, was converged. in fact the idea is that the two sides of perovskite bulks represent two nano crystal prototype.
3. @martin:

Initially just relax a few atoms close to the interface as to limit the degrees of freedom. This should make it easier to converge.

. Do you mean, i use "selective dynamics" and only allow the interface atoms ?
4. i did check the inter-atomic distances in OUTCAR, and make sure the closest distance is at least > 0.5
5. @Alex: Yes the EDIFFG = 0.025 is messed-up i think. i wanted the '-ve' sign before this value so that the criteria be shifted to forces instead of energy, but missed out.
6. @Alex: if there is issue of Van derWaals forces in the middle of the structure, wont the IVDW = 11, tag, take care of it?

Regards

[martin.schlipf]

Do you mean, i use "selective dynamics" and only allow the interface atoms ?

Yes

i did check the inter-atomic distances in OUTCAR, and make sure the closest distance is at least > 0.5

0.5 Å is not a large distance. VASP should print out a warning if your atoms are that close. For comparison, the shortest bond length in nature is H2 and that is 0.74 Å long.

[alex]

Hello SKM,

you are right about your point 6.) and IVDW = 11.
I missed that one despite searching for it twice. :-O

Sorry,

alex

[SKM]

Thank you @martin and its OK, @Alex, no worries.

VASP did not throw-up or given any warning about closer atoms in the run. So that part is OK, i suppose.
But 'selective dynamics' run, initially in the sense, at later step, i supposed to do the full model optimisation or convergence. Right?

[martin.schlipf]

If necessary you can run a full optimization later on. You may also decrease POTIM if the initial forces are still very large.