HSE06 Calculation stops after 4 electronic steps

[mwuensch]

Hey,
I've recently finished some slab relaxations and wanted to calculate the DoS and band structure. For this purpose I intended to use HSE06, however, no matter what calculation I attempt with it (including ones with a very plain INCAR file), the calculation always freeze after 4 electronic steps. I've tried this both on VSC5 as well as on a local VASP 6.4.2 installation, with the bulk version of the POSCAR as well as the relaxed slab. The CPU cores are still working and the memory is also taken up, so it's not like the calculation just fully stops, it just doesn't progress whatsoever. This also happens with much fewer k points than I put into the attached file (

calcfiles111223hse06.zip

). I've also tried starting the calculation with a converged WAVECAR using the standard GGA, as well as with ISMEAR = -5 and different SIGMA values for ISMEAR = 0, with different or no parallelisation, it changes nothing.

Thank you in advance, I really hope the problem is just a simple user error
Max

[jonathan_lahnsteiner2]

Dear mwuensch,

I checked your calculation. I am not able to reproduce your problem.
You should note that you are using a very high ENCUT = 850. Have you checked by
a convergence analysis if you need such a high energy cutoff.
Also, your NEDOS is quite large for testing. When using the following INCAR file
I was able to execute your calculation on 8 intel cores in a couple of minutes.

Code: Select all

SYSTEM = GaN
LHFCALC = .TRUE.
GGA = PE
HFSCREEN = 0.2
ENCUT = 300          # ENMAX from your POTCAR file
PREC = Accurate
ISMEAR = -5
LORBIT = 11
NEDOS = 400

I also used a smaller KPOINTS file for testing:

Code: Select all

Gamma centered K-mesh
 0
Gamma
 4 4 4
 0 0 0

You may try these cheaper settings on your local cluster and then increase the parameters in your input files as you require. Like this, you will also have the possibility to calculate rough estimates of how long your electronic minimization will take. You can do this by varying a certain parameter and monitoring the average LOOP time in the OUTCAR file.

Also, note the output might not be written immediately to your output files on a compute cluster. But might be collected and only written every couple of steps or at the end of the computation.

I hope this is of help.
All the best Jonathan

[mwuensch]

Dear Jonathan,
Thank you for your reply! The high cutoff value did come from a convergence study, but I'm currently redoing it for the slab structure in case the result looks different.
With the softer parameters that you suggested I did get past the first 4 steps of the calculation! However, I did notice that the first 4 steps were much faster than any of the steps after. I seem to have misjudged how expensive HSE06 is, due to how quickly the first four steps were calculated, hence why I put the 5th step not happening within another 3 days down to being a mistake on my part.
With the 135 k-points I had planned to use, HSE06 won't realistically yield a result within 3 days, it seems, even with 128 cores.

Cheers
Max