Convergence issues in spinpolarized calculations of adsorbates on TM surfaces

[kolb_surfcat]

We are studying p-block metal adatoms on transition metal surfaces.

Our results from the spin-plarized ionic relaxation give a total energy of ~ -378 eV for the cell. Restarting the calculation with the final structure leads to convergence to either the same total energy of ~ -378 eV or an energy in the range of ~ -371 eV. Which final energy is observed appears to be random. We can only find this behavior for large unit cells of 64 atoms, anything below does not show adsorption energies comparable to the ~ -378 eV result.


We tried several different smearing methods (ISMEAR = 0, 1) and electronic temperatures (0.05 - 0.2), as well as k-point samplings (1x1x1, 2x2x1, up to 4x4x1) and cutoff energies (400-600 eV). So far we cannot find a clear cause or fix for this issue.

The INCAR we use :

general:
System = Test
ISTART = 0 ; ICHARG=2
ENCUT = 400
ISMEAR = 1; SIGMA = 0.2;
NSW = 400
IBRION = 2
POTIM = 0.4
EDIFF = 4d-04
EDIFFG = -4d-2
NPAR = 4
ISPIN = 2
MAGMOM = 64*0 1*5

Does anyone have experience with similar issues and could provide an idea how to fix this ? Thanks in advance for any help !

[WolverBean]

One possibility in a spin-polarized system is that there is more than one possible distribution of spins that can lead to the same total spin, and that the relaxation algorithm is occasionally triggering a switch from one spin configuration to another. In my experience, even when you set MAGMOM for the initial conditions, VASP doesn't strictly maintain that MAGMOM setting during relaxation. I'd recommend seting LORBIT=10 and examining the magnetic moments on each atom in your OUTCAR file: see if one energy corresponds to a +5 moment on your ad-atom and 0 moments elsewhere, and if your other energy corresponds to (say) a +2 moment on your ad-atom and +1 moments on each of three adjacent surface atoms. This behavior is especially likely in your system, as metal surfaces tend to quench spin, so your ad-atom probably doesn't want to stay in a configuration with 5 unpaired spins.
Unfortunately, while I can suggest what to look for, I don't know of a general solution. The MAGMOM tag just doesn't seem to play well with structural relaxation, and I have yet to find a good workaround.