Correct relaxation steps for (anti)ferromagnetism materials

[Yue-Wen Fang]

For (anti)ferromagnetism materials, a reasonable magnetic ground state is very significant for electronic structure calculations.

VASP seems to give us several ways to acquire it, but which one is the best?

1. Firstly make relaxation calculations from a non-ferromagnetism state, and then make ferromagnetism relaxation calculations from the CHGCAR in non-ferromagnetism calculations. (assuming all calculations get well convergence.)

2. Make ferromagnetism calculations directly.

Question one: which way is better or which way is correct?


Especially, for the anti-ferromagnetism materials such as mott insulators (NiO, MnO), we can take following ways.

1. Firstly make no-magnetism relaxation calculations, then make magnetism relaxation calculations, finally make magnetism relaxation calculations plus Hubbard U.

2. Firstly make no-magnetism relaxation calculations plus Hubbard U, then make magnetism relaxation calculations plus Hubbard U.

3. Make magnetism relaxation calculations firstly, and then make plus Hubbard U calculations.

4. Make magnetism relaxation calculations plus Hubbard U directly.

Question two: which way is better or which way is correct?

[WolverBean]

I can answer for NiO, as I have done this calculation several ways:
If you want the correct antiferromagnetic ground state, you MUST START from an explicitly antiferromagnetic calculation. The fully relaxed geometry for the non-spin-polarized unit cell differs by 10% from the fully relaxed geometry for the correct antiferromagnetic cell (in the PBE functional), so if you relax the unit cell without explicit antiferromagnetism, and then add the magnetic state after, you will get a very different answer.
For this system, the geometry is also sensitive to the value of the Hubbard U parameter. So if you plan to apply a U value (and for this system, you almost certainly need it), then you should use it in every calculation right from the start.

Option 4 is the best: use explicitly magnetic calculations with Hubbard U turned on, directly, in every step of the relaxation. This unfortunately takes a little longer, but for NiO, it seems there are no shortcuts.
You may already know this, but for anti-ferromagnetic systems, you MUST use the MAGMOM tag to set the moment on each Ni atom (I use +2 and -2). If you use ISPIN=2 and NUPDOWN=0 but without MAGMOM set, the calculation will converge to a NONmagnetic solution, which is at least 1 eV above the ground state antiferromagnetic solution. Using a Hubbard U correction does not prevent this problem. Using NUPDOWN=-1 is even worse, as the calculation will converge instead to a ferromagnetic solution.

Good luck!

[Yue-Wen Fang]

Dear WolverBean,

Thank you for your reply.

I also think option 4 is the best though it sometimes costs much more time in relaxation step than other ways.

Hubbard U is indeed have strong effects on the magnetic moment of metalic atoms in MnO, NiO, CoO systems. You said Hubbard U is always turned on, but have you ever met a problem that the obtained Fermi level is not very explicit in nscf calculation?

Generally, I usually conduct scf calculation plus U without reading WAVECAR from relaxation step , then do nscf calculation reading WAVECAR and CHGCAR from previous scf calculation, maybe my calculation steps are not reasonable so that the fermi level is occupied sometimes incorrectly.


[quote author=2 and NUPDOWN=0 but without MAGMOM set, the calculation will converge to a NONmagnetic solution, which is at least 1 eV above the ground state antiferromagnetic solution. Using a Hubbard U correction does not prevent this problem. Using NUPDOWN=-1 is even worse, as the calculation will converge instead to a ferromagnetic solution.

Good luck![/quote]