ABINIT Discussion Forums

Dear all

I would like to know if response function calculations in metals require extra care. I computed the full phonon spectrum of CdS (insulator) without any problems, but I can't manage to run a single RF calculation on AgS, which is metallic.

I always end up with a segmentation fault, or some file that cannot be read, etc.

Since these calculations require a lot of cpus, I cannot spend too much time testing every single parameter, so I ask for your advice.

One of my concern is the value of occopt. Should I use occopt only in the first dataset (GS calculation)? Since subsequent dataset read the 1st dataset output wavefunction, I shouldn't use occopt right?

Other concern is the value of istwfk. When the GS wavefuntions are read, abinit complains that the values of istwfk differ between the two dataset. Is it a problem?

Finally, about the kpoint mesh. Since AgS is a metal, it requires a dense mesh, and that makes like 1000 kpoints in the BZ. Can I use a coarser mesh in the RF calculation compared to the fine grid of the GS calculation?

Thank you very much for your help

Boris

Boris wrote:Dear all

I would like to know if response function calculations in metals require extra care. I computed the full phonon spectrum of CdS (insulator) without any problems, but I can't manage to run a single RF calculation on AgS, which is metallic.

in principle, you just need to converge nkpt and tsmear, but sometimes the RF require tighter convergence than the GS

I always end up with a segmentation fault, or some file that cannot be read, etc.

this is not normal. Are you sure it's not just an insufficient memory issue? The abinit estimation is not always accurate.

Since these calculations require a lot of cpus, I cannot spend too much time testing every single parameter, so I ask for your advice.

One of my concern is the value of occopt. Should I use occopt only in the first dataset (GS calculation)? Since subsequent dataset read the 1st dataset output wavefunction, I shouldn't use occopt right?

you should have the same occopt everywhere.

Other concern is the value of istwfk. When the GS wavefuntions are read, abinit complains that the values of istwfk differ between the two dataset. Is it a problem?

no, it completes the missing k-points. If it were a problem it would crash or complain with an error

Finally, about the kpoint mesh. Since AgS is a metal, it requires a dense mesh, and that makes like 1000 kpoints in the BZ. Can I use a coarser mesh in the RF calculation compared to the fine grid of the GS calculation?

see above - sometimes it's the inverse, certainly not fewer k-points. 1000 is not that many, actually. Depends on the size of your system. Do you mean in the full BZ or the IBZ?

Matthieu