My question is mainly one about consistency in the smearing method used for a standard "geometry relaxation + band structure (BS) + density of states (DOS)" set of calculations. This is a general query, but I make heavy reference to the Vienna Ab initio Simulation Package (VASP).
For gapped systems, I understand that the tetrahedron method (
ISMEAR = -5 in VASP) is recommended for a high-quality DOS (Ref: the VASP wiki this paper). However, if I were to do a BS calculation with a
KPOINTS file such as
Line_mode KPOINTS file 20 Line_mode Reciprocal 0.0 0.0 0.0 ! \Gamma 0.3333333333333333 0.3333333333333333 0.0 ! K 0.3333333333333333 0.3333333333333333 0.0 ! K 0.5 0.0 0.0 ! M 0.5 0.0 0.0 ! M 0.0 0.0 0.0 ! \Gamma
the tetrahedron method would of course be incompatible and give the following error.
The linear tetrahedron method can not be used with the KPOINTS file | | (generation of strings of k-points)
In this case, it would be necessary, for example, to use Gaussian smearing (
ISMEAR = 0 in VASP).
Hence, my concern is that we can't seem to consistently use the same smearing method for the entire "geometry relaxation (e.g. with
ISIF = 3) + BS + DOS" set of calculations.
As the optimized geometry using both smearing methods is effectively indistinguishable within some small tolerance (based on my tests), my main question is if it is acceptable / standard practice to, for example, perform the geometry relaxation (and BS calculation) with Gaussian smearing (
ISMEAR = 0) and only switch over to the tetrahedron method (
ISMEAR = -5) for the DOS calculations (besides other changes like increasing the k-point density)?