I am trying to assign the correct amount of time to my phonon calc.
Here is an example:
I have a very low symmetry 20 atom unit cell. I am going to specify rfatpol 1 20
How many iterations of the 2nd deriv energies will there be?
3n+3? or more?
I ask this because I did a calc w/ 20 atoms, knew it would be a little less than 1 hr per iteration and assigned 63 hrs to the job.
But, then the iterations contiuned on to a 64th iteration. And the job timed out.
Figuring out how many iterations of ETOT for a phonon calc?
Moderators: mverstra, joaocarloscabreu
-
ilukacevic
- Posts: 271
- Joined: Sat Jan 16, 2010 12:05 pm
- Location: Dept. of Physics, University J. J. Strossmayer, Osijek, Croatia
- Contact:
Re: Figuring out how many iterations of ETOT for a phonon ca
Hi!
The number of iterations depends on many other parameters besides the number of atoms. For me it is very hard to estimate it.
You can set the maximal number of iterations so that calculation finishes in assigned job time. But then you risk the precision.
Also, have in mind that you have to calculate the scf procedure for each atom at each q-point you set up. If you figure out how much time it takes to calculate the complete scf procedure for one atom at one q-point, then you can multiply it by the number of atoms and by the number of q-points to estimate the total time for response part of the calculation. And it would still be a very rough estimation, since the response scf depends on atomic species (pseudopotential) and q-point coordinates.
You have "some" testing to do.
Igor
The number of iterations depends on many other parameters besides the number of atoms. For me it is very hard to estimate it.
You can set the maximal number of iterations so that calculation finishes in assigned job time. But then you risk the precision.
Also, have in mind that you have to calculate the scf procedure for each atom at each q-point you set up. If you figure out how much time it takes to calculate the complete scf procedure for one atom at one q-point, then you can multiply it by the number of atoms and by the number of q-points to estimate the total time for response part of the calculation. And it would still be a very rough estimation, since the response scf depends on atomic species (pseudopotential) and q-point coordinates.
You have "some" testing to do.
Igor