Regardless of the type of material, using small smearing often helps in the convergence procedure. However, there are some caveats to look for.
- Always make sure that the smearing value is neither too small nor too large. Run a few convergence tests with different types of smearing to find the appropriate
degauss value and
smearing is somewhat coupled to k-point sampling. So, maybe you want to converge
K_POINTS before converging the
- Smearing is needed for metallic systems to obtain convergence. And it doesn't harm to have a small smearing in semiconductors or insulators too! It affects the electronic energy a little bit but by carefully choosing the
degauss value, you can reduce such an effect. (see answers to this question for details)
- As far as I know, choosing the
smearing type or
degauss value has nothing to do with whether the system is magnetic or not. Usually for any kind of system
gaussian smearing is safe to choose. But as I said previously, maybe test the convergence first. Even though
gaussian is safer, but
mv smearing often gives a substantial advantage.
- If you need to do the DOS calculation,
occupations='tetrahedra_opt' methods are preferred instead of
So, to answer your question precisely, you don't need to use smearing. But probably you will have a hard time achieving convergence if you don't use it. That's why, as a practical method, you can use a small smearing to accelerate the convergence procedure. The code automatically tries to remove any effects due to smearing (fictitious finite temperature effects) after the calculation. Just make sure that this smearing value is not too large. Otherwise, you will get inaccurate ground state properties since the code will have a hard time removing the effects.