interruption of calculation during the SCF

[marco_infantino]

While using VASP, I encounter very ofter interruption during the SCF.
I have found from internet some suggestions related to the info in the vasp.log , but still I continue to have the same interruption very often in many calcultions.
Since is it not very clear to me which is the cause of the interruption, could you please explain it to me and give me some indication on how to overcome this ?
I would be very glad to receive some indications on that!
Attached there are the files.
Thank you in advance.
Marco
ps : I obtain the exact same error in different machines, so it is not very clear if the error source is science related or hardware related

SCF_vasp_interruption.zip

[fabien_tran1]

Hi,

one can see that the total energy is changing from one iteration to the other by a huge amount. At some point, at the 11th iteration, the density and potential have probably unphysical shapes, leading to nonsense values and the crash.

For the moment the questions are:
1) You mentioned that the crash occurs very often. Does it mean systematically? Only for this system?
2) Does the crash occur also if you use ISTART=0?
3) Is the geometry in POSCAR reasonable?

[marco_infantino]

Hello !
Thank you for your answer. So, for this kind of system it happens actually very often.
I really think that the POSCAR is reasonable. Weel, I haven't tried with ISTART=0, but I cancelled everything except the input files, that from a calculation perspective is actually the same....

[fabien_tran1]

Ok, so try ISTART=0 to see if it helps.

[alex]

Hello Marco,

restarting from scratch is surely a good thing in your case.

Two other thoughts:
Consider using NELMDL to delay mixing and calculation of charge density, e.g., I'm using sth. like
NELMDL = -10 ! Number of NON-selfconsistency steps (negative value: delay only for the start-configuration)
in notorious hard cases. It applies only in the very first optimisation cycle.

You may also want to get rid of NELECT for starters. Just use the neutral system for a good wavefunction and add your extra electron later.
Negatively charged stuff is really a nightmare.

Hth,

alex

[marco_infantino]

Alright dear Alex,
Let me try it!! Thanks a lot for your suggestion!
I just have one simple question : in case I add some adsorbate, I think that starting a geometrical optimization with or without background charge could lead to complete different trajectory, right ?
I know this is a further step, I just wanted to know if a slightly different background charge can actually affect an optimization toward different minima.
Always thanks

[alex]

Hi Marco,

in principle you are right. Considering your slab, I'd say that extra electron ends up there delocalized in some d-band not contributing a lot.

Good luck!

alex