Inquiry Regarding Issues with VASP 6.5 Upgrade

[sunw_ujn]

Hello, Developer,

I hope this message finds you well. I recently upgraded to VASP 6.5 and have been encountering several issues that appear intermittently. I have carefully verified the number of MAGMOM values and confirmed that they adhere to the required standards（SOC）. Interestingly, when using versions of VASP prior to 6.5, these errors do not occur. I am quite puzzled by this discrepancy and would like to understand the underlying reason and how I might adjust my parameters to resolve this issue.

Thank you for your attention to this matter.

Best regards,

-----------------------------------------------------------------------------
| |
| EEEEEEE RRRRRR RRRRRR OOOOOOO RRRRRR ### ### ### |
| E R R R R O O R R ### ### ### |
| E R R R R O O R R ### ### ### |
| EEEEE RRRRRR RRRRRR O O RRRRRR # # # |
| E R R R R O O R R |
| E R R R R O O R R ### ### ### |
| EEEEEEE R R R R OOOOOOO R R ### ### ### |
| |
| ERROR: You have set 48 value(s) for MAGMOM; however for |
| spin-polarized calculations (ISPIN=2), MAGMOM has to be a list of |
| NIONS values. NIONS=16 in your calculation. |
| |
| ----> I REFUSE TO CONTINUE WITH THIS SICK JOB ... BYE!!! <---- |
| |
-----------------------------------------------------------------------------

general:
System = CrI
ISTART =0
ICHARG=2
ENCUT = 500
ISMEAR =0
SIGMA = 0.1
#IBRION=2
PREC=High
ISYM=-1
#ISIF=3
#NSW = 200
NELM=300
NPAR=12
LREAL=.FALSE.
ISPIN =2
LORBIT=11
EDIFF=1E-6
EDIFFG=-0.01
#MAGMOM = -2 15*2
#NEDOS=9000
RWIGS=1.302
LWAVE=.FALSE.
LCHARG=.FALSE.

LMAXMIX = 4

LNONCOLLINEAR=.TRUE.
LSORBIT = .TRUE.
SAXIS = 0 0 1
GGA_COMPAT = .FALSE.
MAGMOM = 0 0 -2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2

#LORBMOM= .TRUE.
#I_CONSTRAINED_M = 4
#LAMBDA = 10
#M_CONSTR = 0 0 -1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 1

[christopher_sheldon1]

Hi,

Thank you for your question. I can't see anything immediately wrong. Would you be able to upload the POSCAR, POTCAR, KPOINTS, INCAR, and OUTCAR files and I'll try to reproduce it.

Best wishes,

Chris

[sunw_ujn]

FYI

[christopher_sheldon1]

Thank you, I will try to repeat your calculation.

Best wishes,

Chris

[hszhao.cn@gmail.com]

For the example discussed here, below is the solution for VASP non-collinear calculations on my Intel Xeon Platinum 8488C (2x 48-core):

1. Essential Settings:

Code: Select all

#ISPIN=2             # Not needed with LNONCOLLINEAR=.TRUE.
LNONCOLLINEAR=.TRUE.
LSORBIT=.TRUE.
NCORE=16             # Optimized for 48-core socket
MAGMOM = 0 0 -2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2 0 0 2  # Three components per atom

2. Physical Meaning:
- ISPIN=2 is redundant with LNONCOLLINEAR=.TRUE. because:
- Non-collinear calculations naturally include spin polarization
- The full 2x2 spin density matrix:
$$n_{\sigma\sigma'}(\mathbf{r}) = \sum_{n=1}^N \psi_{n\sigma}(\mathbf{r})\psi^*_{n\sigma'}(\mathbf{r})$$
- Handles both charge and magnetization density
- MAGMOM requires three components per atom (x,y,z directions)

3. Performance Settings:

NCORE=16 # Optimal for 48-core socket system:
- Creates 6 balanced groups (96/16)
- Follows VASP recommendation
- Optimizes parallel efficiency

4. Run Command:

Code: Select all

$ module load vasp
$ module list 

Currently Loaded Modules:
  1) lmod                            5) wannier90/v3.1.0-serial-oneapi.2024.2.0
  2) oneapi/2024.2.0                 6) dftd4/devhub-oneapi.2024.2.0
  3) utils/fi_psm3_uuid              7) vasp/6.5.0-oneapi.2024.2.0
  4) hdf5/1.14.4_3-oneapi.2024.2.0

$ mpirun -n 96 vasp_ncl

This configuration provides both correct physical treatment and optimal computational performance.

Regards,
Zhao

[sunw_ujn]

Thank you. When I commented out `ISPIN=2`, the issue indeed no longer occurs in version 6.5. It seems that version 6.5 might be using a different logic when reading the INCAR file. Regardless, it works now, and that's what matters.

[christopher_sheldon1]

Thank you, Zhao. This is very clear and correct. With VASP 6.5, there is a new check in the reader.F (lines 566-578 in 6.5.0) to make sure that MAGMOM is correctly set. The check for LNONCOLLINEAR and ISPIN=2 are opposites, so the calculation is terminated.

You are right in that ISPIN=2 runs collinear calculations, so running noncollinear makes it redundant. In previous versions of VASP (e.g. 6.4.3), it would simply be ignored in the calculation. ISPIN would be reset to 1 and noncollinear calculations run. I've updated the wiki pages for LNONCOLLINEAR and ISPIN to make it explicit that it is important to not put both ISPIN=2 and NONCOLLINEAR=.TRUE. in VASP 6.5.0.

If this answers your question, then I'll close this thread at the end of the week.

Best wishes,

Chris

[hszhao.cn@gmail.com]

Dear Chris,

I've updated the wiki pages for LNONCOLLINEAR and ISPIN to make it explicit that it is important to not put both ISPIN=2 and NONCOLLINEAR=.TRUE. in VASP 6.5.0.

It seems that there is a syntax error in your following note corresponding to the above updated wiki pages:

Important: For noncollinear calculations, ISPIN=1 should be used. ISPIN=2 performs collinear calculations, which noncollinear make redundant. In VASP 6.5.0, the calculation will crash if ISPIN=2 is used in combination with the LNONCOLLINEAR=.TRUE.

Where, which noncollinear make redundant should be written as: which make noncollinear redundant.

Regards,
zhao

[christopher_sheldon1]

Dear Zhao,

Thank you for pointing that out. Redundant was a poor choice of word, I've rephrased it as "incompatible":

Important: For noncollinear calculations, ISPIN=1 should be used. ISPIN=2 performs collinear calculations, which is incompatible with noncollinear. In VASP 6.5.0, the calculation will crash if ISPIN=2 is used in combination with the LNONCOLLINEAR=.TRUE.

Best wishes,

Chris