Choice of Unit Cell Symmetry for Optimization

Hi. I have been trying to reproduce some of your data. It seems that I can only get similar band gap and ΔH using your U values and other parameters only when I use a primitive cell and not a unit cell with higher symmetry. Moreover, after lowering a cell of specific space group to a primitive and optimizing it, I can only increase the symmetry to a very different space group compared to the initial.

I would like to know whether you did optimization and band structure/DOS on the primitive and if yes, then why? Also the final space group that you mention besides each structure, is it the same as the initial space group from which you got your primitive cell?

Thanks!

Hello,

If you can let know the exact material you’re interested, I can be more specific.

I am not entirely sure what you mean by using a higher-symmetry cell than primitive. But I can tell in general +U energy minima may lie around slightly “distorted” structures; so the lower the symmetry constraints you have, the higher the chance of arriving at the minimum. Therefore depending on the symmetry constraints you have initially (can vary depending on the cell size you use, initial magnetic moments, even the tolerances given to the DFT code for symmetrization) your structure may or may not be able to relax to a similar minima as MP.

Primitive cells are used as they are computationally cheaper to calculate and in principle should return the identical result as the conventional, larger cell. So all calculations should in general be done using the primitive cell, unless there is a good reason to use a larger cell. For example, there are cases in MP where different magnetic configurations / supercells (e.g. AFM ordering ordering may require using a larger cell) are tested for systems that require DFT+U for the reasons I described above, but that may or may not be the case for the particular system you’re interested in. In general, I would recommend turning symmetry completely off in your calculations.

2 Likes

Thank you for your reply! Turning off symmetry was the key!

I am looking into NiAl2O4. The problem I am facing is I am starting with a Fd3m structure and after minimizing it with symmetry off, I get a Cm structure. I can see here in MP, that for your Fd3m strcuture, you started with ICSD 608815 which is a Fd3m structure. But for the other symmetry structures in MP, there are no ICSD structure given. May I know how you got the other symmetrical structures?

Thanks!

Glad to hear it was helpful. To answer your second question:

ICSD is one of the several sources of material structures we have, and therefore not all structures have corresponding ICSD structures. Sources of such entries may vary, such as from high-throughput prototyping, to structure prediction or user submission. There often is a “User remarks” section at the end of the material’s page, which can help identify the source of the structure in such cases.

Thanks! I actually wanted to ask that in NiAl2O4, Fd3m, for example, the computed lattice parameters given are not equal (and hence may not give a Fd3m structure) but in the primitive section, the lattice parameters given are equal (a=b=c=5.778 and thus leads to a Fd3m structure) and I am not sure how this has been calculated or obtained as the computed lattice parameters given are a=5.751, b=5.770, c = 5.763…

Thanks!