Error about k-points when IBRION=8

[Limeng66]

Hi there,

I am posting for error when I calculating lattice dynamics using vasp 6.3.1 or 6.4.2. When I set IBRION=8, in the output file it shows "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." I noticed this error was reported for several times, and have tried the given solution methods provided, but most of them don't work.

Following is a brief summary:
1. a small unitcell (8 atoms) with 48 degree of freedom -- SUCCESS
2. a moderate unitcell (80 atoms) with 24 degree of freedom, no NCORE or NPAR -- SUCCESS
3. a moderate unitcell (80 atoms) with 24 degree of freedom, set "ulimit -s unlimited \ ulimit -s" in both terminal and PBS jobsctipt -- FAIL
4. a moderate unitcell (80 atoms) with 24 degree of freedom, no NCORE but NPAR=2/ncore -- FAIL
5. a moderate unitcell (80 atoms) with 24 degree of freedom, no NPAR but NCORE=2/4/12/24/48 -- FAIL
6. a moderate unitcell (80 atoms) with 24 degree of freedom, ISYM=0 -- FAIL
7. a moderate unitcell (80 atoms) with 24 degree of freedom, making KPOINTS 1*1*1 -- FAIL
8. a moderate unitcell (80 atoms) with 24 degree of freedom, making KPOINTS 2*2*2/3*3*3/4*6*4(so it's almost equivalent along a,b,c direction) -- FAIL

Actually I want to use DFPT method to calculate a 320 atom supercell, so the sucessful method 1 and 2 seem unapplicable.
Could you give me some suggestion how to realize my calculation, either modify the TAGs or reset the environment.

With best regards.
Meng

[pedro_melo]

Dear Limeng66,

Thank you for your extensive summary of tests. However, I am having trouble understanding what you mean when you write that the job failed for each occasion. Was it stoped by VASP itself, or by the job manager at your HPC cluster (due to out of memory, out of time, etc)? On either case, could you provide me with examples of the OUTCAR files or the error/log files created by PBS?

By the way, there is an error in the INCAR you provided: you have LRAEL instead of LREAL. Very likely it did not affect your jobs, since it is not a default variable of VASP, and the default value of LREAL is set to FALSE.

Furthermore, when you wrote "48 degrees of freedom" or "24 degrees of freedom" did you mean the number of cores?

Kind regards,
Pedro

[Limeng66]

Dear Pedro,

Thank you for your prompt reply. I am sorry for making it ambiguous. For all failed cases, it shows "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" in the VASP output file. I am not sure wether it is stopped by VASP or HPC; I have just run a case calculation (with same error) and all relevant files are attached.

Thank you for pointing the wrong TAG in my INCAR.

In terms of the "degree of freedom", I borrowed this concept in the vasp output file. It may equal to how many round it will do the electronic calculation (see attached vasp.out).

Best regards,
Meng

[pedro_melo]

Dear Meng,

I spoke with my colleagues. As fas as I am aware, NCORE/=1 is not supported for IBRION=8 (or even 6). The error message you are seeing is a direct consequence of this.

Note that you do not need to specify NPAR if NCORE is already given. As it is stated here https://www.vasp.at/wiki/index.php/NCORE, "The user should either specify NCORE or NPAR, where NPAR takes a higher preference".

Could you try your runs without specifying NCORE in your input file?

Best,
Pedro Melo

[Limeng66]

Dear Pedro,

Thank you for your reply.

I think you may be correct that IBRION=8 (and maybe 6) does not support NCORE!=1. I have tried 2 cases: for the first case, I disabled both NCORE and NPAR, and the calculation can be sucessfully proceeded (see attachment 1); for the second case, I I disabled NCORE then set NPAR equaling to the number of cpus I used on HPC, and the calculation can be sucessfully proceeded.

However, I noticed one interesting thing: When I use a relatively small supercell, i.e., a 2x2x2 supercell containing 16 atoms, I can run the calculation with NCORE set equalling to the number of cpus I used on HPC (see attachment 2). Does this mean there are some potential way to set NCORE!=1 for an larger supercell? I am asking this because I think parallelization is important to speed up my calculation and save the resources.

P.S. I found there are some confusion in vasp software and on vasp wiki. For instance, in the error output file (as those in my previous attachment), it says "set NPAR equal to the number of nodes", but on the vasp wiki it says "the default of NPAR is the number of cores". While on HPC, we usually call a batch of cpus a node, for instance, on the one I's using, there are 48 cpus per node. So I was confused what should be the appropriate value of NPAR based on the error output file.

Looking forward your further suggestion on the parallelization,
Merry Christmas in advance,
Meng

[Limeng66]

Dear Pedro and Admin,

May I re-draw your attention on the parallelization when using IBRION=8?

"I noticed one interesting thing: When I use a relatively small supercell, i.e., a 2x2x2 supercell containing 16 atoms, I can run the calculation with NCORE set equalling to the number of cpus I used on HPC (see attachment 2). Does this mean there are some potential way to set NCORE!=1 for an larger supercell? I am asking this because I think parallelization is important to speed up my calculation and save the resources."

Looking forward your answer,
Meng