My Community

Dear all,

I am facing a problem when calculating a SCF + nonSCF at a larger k-point mesh (using KPOINTS_OPT) in relation
with the parallel options NPAR or NCORE and KPAR.

I set up the k-point mesh with KSPACING=0.14, which gives a mesh of about 7x7x7 (52 kpoints and 343 in the full BZ).
The problem I am facing are the same if I use the KPOINTS file.
I set 64 bands among which 48 are occupied (this a SC), and I use 64 cores.

I also set NCORE=8 and KPAR=8.

With this setup, the SCF converges well (E=-49.72821151 eV) but I subsequently get the error:
| VASP internal routines have requested a change of the k-point set. |
| Unfortunately, this is only possible if NPAR=number of nodes. |
| Please remove the tag NPAR from the INCAR file and restart the |
| calculation.

Hence, I remove NCORE and KPAR, but then the SCF does not converge at all. The energy values are irrelevant (e.g., -22953136.75324744 eV),
the log file indicates "WARNING: Sub-Space-Matrix is not hermitian in DAV” and the job stops with the error in the OUTCAR:
Error EDDDAV: Call to ZHEGV failed. Returncode = 90 2 128

Hence, I set NPAR=number of nodes. Note that there is a contradiction between VASP OUTCAR message and the website (wiki/index.php/NPAR)
for NPAR that states: NPAR = number of cores
So I set NPAR=64 (= # of cores), but then I get the error: SCF not converging and "WARNING: Sub-Space-Matrix is not hermitian in DAV”.

If I set NPAR=8, then the SCF converges, but I get the error of change of the k-point set.

So I am stuck here. What should I do?

Thank you for your help,
Best,
Pascal

Dear Pascal,

Thank you for reaching us. Could you please post here the input files you are using, as well as the OUTCAR file? It will help with tracking your issue.

Kind regards,
Pedro Melo

Dear Pedro,

Yes, I have put 2 INCAR and OUTCAR files and the logfiles in the attached archive. The output720069.o and output720283.o logfiles are related to the INCAR.1 and INCAR.2, respectively.

Best regards
Pascal

Interestingly, I want to run a ELASTIC CONSTANTS+PHONONS calculation with IBRION=6, but I get the same error about the change in k-point mesh.

The INCAR and OUTCAR files are attached.

Best regards
Pascal

Dear Pascal,

I would suggest that you read the following pages, where it is explained how to set up the different parallelisation choices and the incompatibilities between them

wiki/index.php/NPAR
wiki/index.php/NCORE

Be careful with the number of cores per socket that you have and the total number of tasks that you ask for. Moreover, I was able to run the system you sent with both input files, with having set NCORE = 1; KPAR = 4; NPAR = 8 on INCAR.2. I think this might have been the issue: you were using the wrong parallelisation options for this type of calculation and should have used NPAR instead of NCORE.

Please test your job with new parallel options taking into account the pages I've sent you and let me know if it worked.

Kind regards,
Pedro Melo

All right, thank you for the hints. I am going to test more carefully these options.

Best regards,
Pascal