constrained total magnetization for collinear (no SOC) calculations?

[sophie_weber]

Hi all,

I think I know the answer to this unfortunately, but I just wanted to confirm for sure. I am looking at properties of magnetic moments on the surface of vacuum-terminated (110) FeF2. Specifically, I am interested in showing that the magnitude of the two Fe moments on each (110) surface become inequivalent, leading to a small non-zero magnetization on the surface. This is already apparent in a simple scf calculation (attached) of the relaxed slab, where the site-projected magnetization in the OUTCAR shows that the magnitude of the two outermost Fe ions on each surface (they are ordered in the POSCAR along the [110] direction, thus these correspond to Fe 1 and 2 for surface 1, and 15 and 16 for surface 2 respectively) differ by a small but non-negligible (~0.01 uB) whereas the moments away from the surface in the bulk are all equivalent to within 0.001 uB

However, ideally I would like to show this more rigorously, i.e. by plotting energy as a function of fixed magnetization in the slab and show that the energy is lowered for a small magnetization compared to being exactly zero. I tried to do this with NUPDOWN, setting it to fractional values, but because this is literally fixing the number of up and down electrons (rather than what's actually happening, which is the degree of itinerancy of same-spin-state sub lattices become inequivalent), it forces the system to become unphysically metallic at some KPOINTS, which of course raises the energy.

Is there a way in VASP to constrain the total magnetization (NOT the number of up minus down electrons) within collinear calculations? I am pretty sure this is not the case; of course, another option is to switch on spin-orbit coupling and do a constrained magnetization calculation where the magnitude, as well as direction, is constrained (I_CONSTRAINED_M=2), but this is a bit overkill, and so far my attempts have been a mess (my penalty energies are order 10 eV, rendering the results meaningless...). I just wanted to make sure there wasn't an easier way to do this that I'm missing.

Thank you in advance!

[merzuk.kaltak]

Dear Sophie,

unfortunately, there is no other alternative to I_CONSTRAINED_M for collinear calculations.

Please note, vasp-5.4.4 does not symmetrize the local magnetic moments correctly (see here).
We propose to update to vasp-6 or use the workaround as described in here.