Convergence problem with METAGGA=MBJ

[reynaldo.putra]

Hi,

I am conducting a test run using MBJ to compare PDOS with HSE06. I use the potpaw_PBE.64 potentials and ALGO=ALL. I also use tetrahedron smearing with LORBIT=11.

However, even with ALGO=A, I still face convergence problems; the free energy does not come remotely close to converging (I checked using grep TOTEN OUTCAR).

I have two questions.

1. For MBJ runs, do we still need to come to SCF convergence in order to have non-obsolete results?

2. If yes, should I change my convergence criteria, e.g., NELM, EDIFF, etc.?

Here are my INCAR inputs:

NPAR = 16
NWRITE = 2
LPETIM = .TRUE.

INIWAV = 1
#ISTART = 0
ICHARG = 1

LCHARG = T
LWAVE = F # Put at FALSE, no need to save them bcs takes too much space

Electronic relaxation
ENCUT = 480 # ~20% from the max value in POTCAR
NELM = 1500
ISPIN = 2 # 2 for spin polarization, 4x time of calculation
LREAL = Auto

Stopping criterion for ELM
METAGGA = MBJ
ALGO = All # All for HSE, Normal for PBE
EDIFF = 0.000001 # or 1E-6

DOS-related values
ISMEAR = -5
LORBIT = 11
SIGMA = 0.2

And here are the free energies from the last few SCF cycles:

free energy TOTEN = 2019.95973415 eV
free energy TOTEN = 1967.58592570 eV
free energy TOTEN = 1935.47682361 eV
free energy TOTEN = 1920.17815524 eV
free energy TOTEN = 1913.99886720 eV
free energy TOTEN = 1936.55162392 eV
free energy TOTEN = 1924.30708710 eV
free energy TOTEN = 1920.90786269 eV
free energy TOTEN = 1923.31836842 eV
free energy TOTEN = 1924.03491789 eV
free energy TOTEN = 1928.27026255 eV
free energy TOTEN = 1933.90708122 eV
free energy TOTEN = 1936.53564912 eV
free energy TOTEN = 1938.33838727 eV
free energy TOTEN = 1940.77942310 eV
free energy TOTEN = 1949.37214772 eV
free energy TOTEN = 1969.08766398 eV
free energy TOTEN = 1973.09098119 eV
free energy TOTEN = 1974.83347109 eV
free energy TOTEN = 1983.80735762 eV
free energy TOTEN = 2017.59151735 eV
free energy TOTEN = 2002.17188098 eV
free energy TOTEN = 2004.24538826 eV
free energy TOTEN = 2010.16255471 eV
free energy TOTEN = 2030.78061239 eV
free energy TOTEN = 2036.38372063 eV
free energy TOTEN = 2036.38372063 eV

[pedro_melo]

Dear reynaldo.putra,

I will look into this issue. Meanwhile, could you post here your input files (INCAR, POSCAR, POTCAR, KPOINTS), the OUTCAR file for a small case system where you still find this issue? If your system is not so large it would also be acceptable.

Kind regards,
Pedro

[reynaldo.putra]

Dear Pedro,

Thank you very much for offering your assistance.

Here are all the input files

Best regards,
Reynaldo

[reynaldo.putra]

Dear Pedro,

Sorry to bother you again, but what are your thoughts on the calculation? Should I simply increase the SCF cycle limit?

Best regards,
Reynaldo

[pedro_melo]

Dear Reynaldo,

A colleague of mine more experienced in DFT functionals is studying your case right now. It seems that the functional in question leads to a very unstable SCF cycle. One of us will report back once we have more information.

Kind regards,
Pedro

[fabien_tran1]

Hi,

I have made a certain number of attempts to converge the MBJ calculation, but all have been unsuccessful so far. The change in the energy and magnetic configuration along the SCF cycle is very erratic and increasing the SCF cycle limit seems useless. I will try a few more tricks and let you know if convergence can finally be obtained, which seems however hopeless.

Do you know which magnetic configuration is expected for this system? Are there experimental results? Actually, the PBE and SCAN functionals lead to a very fast SCF convergence and to a non-magnetic ground state.

[reynaldo.putra]

Dear Fabien,

Thank you very much for investigating into this issue. I am not sure about the magnetism of the system, which is why I included the spin polarization in the runs. According to some light searching of the literature it is diamagnetic.

Yes, I also did not face any errors with the SCAN functionals, hence why I am confused with the erratic energy change per cycle.

Knowing that it is diamagnetic, what could I expect from running a non-spin polarized calculation of the system? Would there be a significant change?

Best regards,
Reynaldo

[fabien_tran1]

Hi,

I did a non-spin polarized MBJ calculation, but this was also not possible to reach convergence: similar chaotic change in the energy along the iterations.