Primitive vs. supercell noncollinear magnetization

[vahid_askarpour]

Dear VASP Users,

When VASP calculates the magnetization energy (such as JS_iS_j), is this energy converged in the code relative to supercell size? How does VASP determine how many neighbours to include in the interaction energy summation? Alternatively, can differences in the magnetic moments/energy between primitive and supercells be attributed to this convergence issue?

Thanks,
Vahid

[jonathan_lahnsteiner2]

Dear Vahid Askarpour,

VASP can not determine for you how large your simulation box has to be. In principle one should always do a finite size analysis to check if the quantity which you would like to determine is converged with respect to the size of your simulation box. You can do this by repeating your calculation for different simulation box sizes and comparing the obtained results.

All the best Jonathan

[vahid_askarpour]

Dear Jonathan,

Thank you for the quick response and your comment on convergence with supercell size.

Is there a reason then why in the Hobbs, Kresse and Hafner paper (PRB62, 11556, 2000), no mention of any supercell convergence (angles, moments and energy) made for the noncollinear magnetization calculation? All calculations in that paper were made on a single cell. This is the paper that explains noncollinear implementation in VASP. And no mention of supercell convergence is made in any of the examples on VASP Wiki or the VASP manual? One can assume that convergence is a must and need not be mentioned but then one wonders why in the above paper, only a single cell was used for all the materials considered.

Thanks,
Vahid

[jonathan_lahnsteiner2]

Dear Vahid Askarpour,

One should always do finite size analysis for bulk DFT calculations if it is not known what boxes sizes are appropriate for a specific material and the quantity of interest. It is important to verify that the used box is large enough to capture the physics which are important for the material of interest. This is not only important for non-collinear magnetic calculations but for any quantity you can obtain from atomistic simulations. This finite size study will not always be published because the box size might already be known from previous studies. Therefore, usually only the boxes are shown for which it was verified that they are large enough and the quantity of interest can be obtained without finite size effect errors. The paper Hobbs, Kresse and Hafner paper (PRB62, 11556, 2000) is discussing only small clusters of atoms were the largest is a pentamer. So there are no bulk materials discussed in this paper for which a finite size analysis should be done.

All the best Jonathan

[vahid_askarpour]

Dear Jonathan,

While it is true that in general, physical properties should be converged, I have not yet seen a paper on noncollinear magnetization that specifies the size of the supercell used. Papers on thermal conductivity, however, always quote the supercell size for second and third-order force constants so that users can reproduce the data. In case you are aware of at least one paper that specifies the supercell size for noncollinear magnetization of a bulk material, I would very much appreciate to know.

For example, the determination of Berry curvature of bcc Fe by the Vanderbilt group (PRB74, 195118, 2006) uses a primitive cell as seen from their procedure, Berry curvature and band structure. It seems that some authors are very secretive about their supercell size.

Thanks,
Vahid