I'm a little confused about how the DOS calculation works. I assumed that task 10 is a primarily post-processing task and hence that it is relatively fast. So I've done a converged self-consistent ground state calculation, and then want to use the following input file to do a non-self-consistent calculation on a denser grid, then calculate the DOS.
The problem is that this is taking a very long time. The nscf calculation part goes fine and finishes, but it seems to get stuck on task 10 for so long that none of my calculations have even finished (even with the wplot parameters turned down so low as below). Time wise, the one below failed to finish when given 48 hrs to run and MPI parallelised over 192 cores. Sorry I can't give any more specific timing info.
Am I missing something here? Any help is appreciated.
wplot is dependent of the k-point. If you look at the logfile, the calculation of DOS and band structure (task 20) is only using one core. I think this is because of the huge number of k-points that you have combined with only one core to integrate all of the k-points.
I usually leave wplot to default, it should be sufficient. I tried increasing the nwplot to 1000 and my calculation also took a long time.
Best,
Cenna
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
So is my non-self-consistent calculation necessary? Or is it sufficient to do a normal, self-consistent, converged ground state and then rely on wplot to do the finer interpolation of the Brillouin zone?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Depends on what you need. If you want DOS, band structure, etc. then you need the non-SC calculation.
Or is it sufficient to do a normal, self-consistent, converged ground state and then rely on wplot to do the finer interpolation of the Brillouin zone?
If you do a normal (i.e. small k-point sampling) then you'd get a normal DOS. Increaseing the parameters in wplot would just make the points smaller but I don't think this would increase the actual shape or resultion of the DOS. It would just take longer to calculate and you'd get like hundreds of points for the same data.
I think for resultion you should rely on k-point sampling AND swidth. You can read about it in previous posts in the forum.
Best,
Cenna
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi All,
I'm a little confused about how the DOS calculation works. I assumed that task 10 is a primarily post-processing task and hence that it is relatively fast. So I've done a converged self-consistent ground state calculation, and then want to use the following input file to do a non-self-consistent calculation on a denser grid, then calculate the DOS.
The problem is that this is taking a very long time. The nscf calculation part goes fine and finishes, but it seems to get stuck on task 10 for so long that none of my calculations have even finished (even with the wplot parameters turned down so low as below). Time wise, the one below failed to finish when given 48 hrs to run and MPI parallelised over 192 cores. Sorry I can't give any more specific timing info.
Am I missing something here? Any help is appreciated.
Harry
Last edit: Harry K 2018-01-30
Does anyone have any tips on this? Perhaps I have a misunderstanding of wplot
Hi Harry,
wplot is dependent of the k-point. If you look at the logfile, the calculation of DOS and band structure (task 20) is only using one core. I think this is because of the huge number of k-points that you have combined with only one core to integrate all of the k-points.
I usually leave wplot to default, it should be sufficient. I tried increasing the nwplot to 1000 and my calculation also took a long time.
Best,
Cenna
So is my non-self-consistent calculation necessary? Or is it sufficient to do a normal, self-consistent, converged ground state and then rely on wplot to do the finer interpolation of the Brillouin zone?
Dear Harry,
Depends on what you need. If you want DOS, band structure, etc. then you need the non-SC calculation.
If you do a normal (i.e. small k-point sampling) then you'd get a normal DOS. Increaseing the parameters in wplot would just make the points smaller but I don't think this would increase the actual shape or resultion of the DOS. It would just take longer to calculate and you'd get like hundreds of points for the same data.
I think for resultion you should rely on k-point sampling AND swidth. You can read about it in previous posts in the forum.
Best,
Cenna