Magnetic moments collapse to zero in CoNi FCC (VASP 6.5.1)

[pmoreira]

Dear VASP Support,

I am running spin-polarized calculations with VASP 6.5.1 on an ordered FCC CoNi (2×2×2 supercell) using PBE pseudopotentials. However, both FM and AFM configurations always converge to a non-magnetic solution (local and total magnetic moments go to nearly zero).

Below is the INCAR I used for the FM case:

ISTART = 0
ICHARG = 2
ISYM = 0
ISPIN = 2
MAGMOM = 16*2.0 16*0.6
ENCUT = 500
PREC = Normal
ALGO = Normal
NELM = 60
ISMEAR = 1
SIGMA = 0.2
LORBIT = 11
RWIGS = 1.30 1.25

Could you please advise why the spin polarization always collapses to zero, and how I could resolve this?

Thank you for your help.

Best regards,
Pedro Moreira

[michael_wolloch]

Dear Pedro Moreira,

Please attache a zipped folder of all your input files (INCAR, POSCAR, KPOINTS, POTCAR) and relevant output files (at least OUTCAR, but maybe also std_out) as per the posting guidelines.

I can help you much better when I have all the information.

However, generally it is advisable to set the initial magnetic moments higher than the expected results, as it is mentioned in the MAGMOM page on the wiki.

Thanks, Michael

[pmoreira]

Dear Michel,

The requested files are compressed in NiCo_FM.tar format. I increased the initial MAGMOM values, but I did not get a more satisfactory result.

Best regards,
Pedro

[michael_wolloch]

Dear Pedro Moreira,

Your system is under enormous pressure of a about 6900 kbar. VASP also produces a warning, which is written in your OUTCAR, about ions being very close together.

It is expected that magnetic moments collapse under high pressure, so I think your results are correct for the system.
You might have made a mistake in your supercell construction, however, and the very high pressure might not be intended. After relaxing your cell (with quite loose settings and with ISPIN=1), I obtained a lattice parameter of about 6.96 Angstrom instead of your 5.6. At this volume, the system converges easily to a ferromagnetic solution.

Please let me know if this solves your question, and I can lock the thread. Of course, I am also happy to continue the discussion or answer a follow-up question on the topic.

Cheers, Michael

EDIT:
I forgot to mention that your parallelization setup is far from ideal. I used the same input files as you did to reproduce your results on 64 cores and, by adding KPAR=4 and NCORE=4 I reduced the calculation time from about 5.5 minutes to less than 1 minute. Obviously, our nodes are different, but I would suggest reading our wiki pages on parallelization to optimize your setup.

[pmoreira]

Thanks, Michael. Your answer was very helpful. It solved my question.