Symmetry, kpoints and memory

[askhetan]

If I am doing surface slab calculations in VASP with ISYM=2 (symmetry on), then the brillouin zone has 16 irreducible kpoints when using 11x11x1 gamma kpoints. However, when I use ISYM=0 (symmetry off), then the brillouin zone has 25 irreducible kpoints when using just 7x7x1 gamma kpoints in the KPOINTS file.

I want to know the following
1) Which set of simulations will be, in principle, more accurate? My guess would be the one with 7x7x1 kpoints because it has more points in the irreducible BZ. Please correct me if I'm wrong.
2) Which set of calculations will consume more memory? Again, My guess would be the one with 7x7x1 kpoints because of the same logic.
3) Why do we the use symmetry in calculations if the number of kpoints in the irreducible BZ is perhaps what matters in the end? When using high kpoints, for eg 11x111x1, we are just using symmetry in the end to reduce the number of actual points in the irreducible zone. Is there any advantage in going to higher kpoints ?

I want to know the answers to these questions in the context of GW calculations. Are there some functions in GW method which are memory intensive and are calculated differently. If I use less number of kpoints with symmery off, will it help me in some way reduce the memory requirement. Unfortunately, I do not have enough compute time left to test this hypothesis.

[alex]

Hi askhetan,

I'm sorry to spoil your mood.
1+2) wrong.
Symmetry on: All symmetry operations will also used on the k-mesh. You'll have a much finer grid with 16 k-points and symmetry compared to P1 and 25 k-points. This is indicated by 'weight' in the list of k-points.
3) You can probably answer yourself.

Cheers,

alex

[askhetan]

Dear Alex,

Thanks a lot for the reply. I saw a recent published paper in which the cell dimensions for GW caclulations were just twice those of mine (ofcourse for the same material) but kpoints were 2x2x1. My tests require a minimum of 11x11x1 kpts to show convergence.I guessed that maybe they had switched off symmetry and that helped them achieve convergence already at 2x2x1 kpts.
Is there any place i can look to understand how this is implemented in VASP?

MfG
askhetan

[askhetan]

Just noticed that during GW calculations, VASP automatically sets ISYM=3 which is symmetric only in forces and stresstensor, not charge density. The IBZKPT is pretty dense. (49 for 7x7x1 and 121 for 11x11x1), My convergence tests were not wrong. I actually seem to need 11x11x1 kpts, but don't know how to reduce the memory requirements.