ABINIT Discussion Forums

Hello:

My ABINIT calculations have been very slow, and I think I may be using too many k-points. For every calculation that I run, I use multiple sets of k-points to check that the total energy is converged with respect to the number of k-points (to within 0.1 kcal/mol). An example is below:

nkpt | etotal (Ha) | kptrlen
6 | -29.408594 | 34
18 | -29.409480 | 56
40 | -29.410157 | 78 *converged
75 | -29.410234 | 101
126 | -29.410215 | 123

ABINIT's default kptrlen is 30, but my systems are metals, not insulators. I tried using the prtkpt utility, but it only showed me nkpt up to 6 and the merit factor only went up to about 2/3.

This just seems like too many k-points. Am I doing something wrong? Should I converge to some other parameter beside the total energy, or do metals just require that much less error in the kptrlen?

Thank you for your advice!
Kale

How big is your unit cell? There is a reciprocal relation (expressed by kptrlen, more or less). Large cells don't need as many k points, smaller cells need more as a rule of thumb, because what matters really is kptrlen. For converging stress you might need kptrlen of 60 or 70 or even more, and if the cell is small, this will lead to a lot of k points. If the cell is large, there won't be so many. abinit is very well parallelized over k points so this is an idea situation to be running on multiple cores.

Thank you for your reply. The dimensions of this particular cell are about 9x9x9 angstroms. The k-points grids I used were simple gamma-centered 3x3x3, 5x5x5, ..., 9x9x9 meshes.

It sounds like I may need to use a high kptrlen, so my number of k-points is not so unreasonable after all.

Thanks again,
Kale