I am new to DFT calculations and I would like to get the electronic and phonon band structures of FeSn (1 kagome layer and 1 Sn layer in between). My question is, is it absolutely necessary to first do a geometry optimisation before performing the electronic and phonon band structures? I will be using the density functional perturbation theory (DFPT) method to calculate the phonon dispersion. The code I shall be using is Elk.
- For the electronic band structure, I'd say it's debatable.
For consistency reasons, it would make sense for example to perform a geometry optimization with PBE and calculate its PBE band structure. So we'd be reporting the PBE band structure of a PBE optimized structure.
On the flip side, it is common for people to do PBE geometry optimization and use a hybrid (HSE06 for example) for the band structure calculation, because doing a geometry optimization with a hybrid is very expensive. So, in this case we'd report a HSE06 band structure of a PBE optimized structure.
Of course, the next relevant question is whether we can use an experimentally obtained structure that might not be the most stable configuration in our DFT calculations. I have seen examples of this for experimental papers where DFT was used to give insight or corroborate the experimental results, but for purely computational studies I find that people usually relax the structures according to the level of DFT theory (LDA, GGA, hybrid, etc) and choice of parameters (k-point mesh, plane-wave cutoffs, etc). This is just my impression, so take it with a pinch of salt.
- For the phonon band structure, I'd say it's necessary.
I think it is a technical requirement to perform the geometry optimization first if you're working in the harmonic approximation where you assume that you are at the bottom of a quadratically shaped basin (with respect to the ionic displacements) in the energy landscape. This is explained clearly by ProfM here and here. This discussion on how low the forces should be to converge phonon energies also seems useful for you to look at. I suspect that one might face convergence issues and maybe artificial soft modes if the geometry optimization is not performed beforehand.