My Community

Hello,
This is somewhat related to my last post; this may be too involved of a question for the forum but I wanted to give it a try. I am trying to simulate a slab of chromia (Cr2O3) with a layer of Pt randomly deposited on top (I put in two Pt per Cr2O3 hexagonal unit cell, then relaxed the positions of the atoms). Specifically, I wanted to extract the isotropic magnetic exchange constants by flipping individual spins and taking total energy differences. I have done this before with a pure chromia slab with vacuum, and it was no problem to converge calculations with flipping individual spins while keeping the rest of the spins fixed along +z (see the tar file "Cr2O3slab_noPt.tar", where the moment printout in the OUTCAR where Cr atoms "1" and "4" are initialised as negative, and stay negative).

However, in general for the Cr2O3 slab with the Pt layer on one side, the Cr moments do not stay in directions I initialise them. I'm not necessarily shocked for the Cr which are near to the surface with the Pt layer, as the Pt could affect the magnetic order. What is very disconcerting though is that I can't converge magnetic calculations where I flip the moments on the opposite side of the slab, which should be far enough away from the Pt to not be affected. See the tar file "Cr2O3_Ptlayer.tar", where I am trying to again flip spins 1 and 4, but the OUTCAR converges to have spins 1 and 5 flipped with 4 still pointing along +z. I know the most obvious possibility is that my slab isn't thick enough such that these Cr are still interacting with the Pt, but I don't think this is the case since looking at the layer-projected DOS, the DOS in the center of the slab look bulk-like. I have tried this calculation using a higher density of Pt, using a dipole correction, using ISYM=0 versus ISYM=-1, using a larger vacuum, and no matter what the Cr on the opposite side of the Pt layer do not converge as they did for the pure Cr2O3 slab.

Again I know this is rather an involved question, but if anyone sees something immediately suspect or which I ought to be doing differently, I'd greatly appreciate it.

The below may be not very helpful, but hopefully it might. What you are seeing with an apparent long-range interaction may be physical.

I say that based on the similarity between your setup and the model problem for developing ML potentials with non-local effects, where an adsorbed atom on a slab's surface influences the opposite surface of the slab. See e.g.: https://pubs.acs.org/doi/full/10.1021/a ... ts.0c00689 (Figure 2b and discussion)

(The point of bringing up the reference here is not the subject matter of the ML potentials, but the setup of a problem that looks similar to yours)

Thank you! In any case this seems very interesting, I'll take a look!

Best,
Sophie