QuantumATK Forum
QuantumATK => General Questions and Answers => Topic started by: ams_nanolab on May 11, 2016, 12:58
-
For 1 D / 2D / 3D material what should be ideal k-point setting for optical spectra? For DOS it should be dependent on the dimension of the material , but will that hold for optical spectra as well?
-
The ideal k-point setting is system dependent, but yes, it is also always dependent on the dimension of the material. Ideally, you should check that the optical spectrum is converged with respect to the number of k-points, and also the number of bands included.
-
I would guess that the number of k-grid points needed for calculating the optical spectra are to be at least comparable to that typically used for DOS calculations. As a general rule, the actual k-grid for any type of calculations has to be chosen based on a convergence test for a specific physical quantity of interest.
-
How to carry out the convergence test?
-
One should increase the number of k-points to see how the physical quantity of interest (e.g., optical spectrum) changes upon increasing the k-mesh density. Note that one needs to substantially increase the density, e.g., not just using 5x5x5 instead of 4x4x4. Otherwise, you may not see much of a difference.
This kind of convergence tests are usually done for several different meshes. "Substantially" and the amount of testing effort are vaguely-defined things indeed, but this is where one needs to experiment to get reliable computational results within the tolerance you think is good enough for the problem of your study.
A good initial guess for k-point sampling usually comes with experience in doing calculations for different physical systems and quantities. The convergence speed with respect to k-point sampling can differ significantly for different physical quantities as well as physical systems (e.g., metals vs semiconductors).
-
Okay :)