large number of k-points

[Simil Thomas]

Dear Koopman team,

 

In one of my calculations, the unit cell size is approximately 14 Å. In the JSON file, I specified a 2x2x2 k-point grid as follows:

    "kpoints": {
        "grid": [2, 2, 2],
        "offset": [0, 0, 0],
        "path": "XGY"
    },

However, upon examining the postproc/wannier folder, I noticed that the k-point grid is set to 8x8x8, significantly slowing down the calculations (especially wannier calculations with nearly 660 bands). The initial wannier folder has the correct k-point grid of 2x2x2. Is there a way to reduce the k-point grid for postproc/wannier/nscf calculations? Or can I manually edit and run it?

With regards

Simil

[Edward Linscott]

Dear Simil,

It sounds like you have the smooth interpolation functionality enabled. If you don't know what this means, it is explained in section 3.1.3 of the SI of our recent JCTC paper. This functionality is controlled by the smooth_int_factor keyword in the ui (unfolding and interpolation) block in your input file -- during postprocessing, the k-grid is multiplied by smooth_int_factor.

I expect in your case you will find that smooth_int_factor has been set to 4. You can decrease this value to 2 or even remove this setting entirely if you like.

For the reasons explained in that document, it is generally a good idea to use as large a smooth_int_factor as possible because (usually) the cost of performing the Wannier interpolation on a larger k-grid is small relative to the cost of performing the rest of the Koopmans calculation, and generally you get a much better band structure. But of course if the wannierisation during the postprocessing starts to dominate the computational cost of your calculation, then decreasing smooth_int_factor is the thing to do.

Hope this helps,

Edward