Gamma or Monkhorst ?

[askhetan]

I did a 1x1 supercell calculation of CeO2 system. It is essentially an fcc lattice. I am confused about the effect of the type of KPOINTS on accuracy and speed of computation

with 13x13x1 Gamma centered points my IBZKPT file has 82 entries for the reciprocal lattice but with 12x12x1 Monkhorst (non- gamma centered) points my IBZKPT file had 216 entries. Two questions arise here:

1. Is the MONKHORST calculation with 216 entries more accurate than the GAMMA centered calculation with 82 entries. Does it also mean that the time taken by KPOINT calculation will also increase about 3 fold?

2. If I reduce the number of KPOINTs in Monkhorst such that the number of entries in the reciprocal lattice is nearly equal to the same in Gamma centered (which is 82), will my calculations converge faster without affecting accuracy since the lattice is FCC??

PLease let me know your expert opinion, because of dearth of experimental data, it is difficult to know which set of kpoonts actually give more accurate solutions

[juhL]

1) yes, the more points in the IBZ, the more expensive your calculation.

2) the 13x13x1 grid should be more accurate than the 12x12x1 grid

3) the reason for the larger number of points in the IBZ in the 12x12x1 grid case is that VASP chooses a kpoint grid that resembles the full symmetry of the system. in your case, symmetry is lost in the case of not gamma centered grids, i.e., all even monkhorst kpoint grids (that are not shifted to the gamma point) will have (much) more points in the IBZ.

i would suggest to take only odd kpoint grids, or to shift the even ones to the gamma point.

[askhetan]

Dear juhL,

Thanks for your reply. does that mean ??

1. More number of K points means more accurate calculations but for the SAME k points - more points in the IBZ (due to gamma/monkhorst difference) means more expensive calculations ? Coz it seems that with 13x13x1 gamma centered grids i can do faster and more accurate calculations than 12x12x1 (non- gamma cenetred) monkhorst grids which requires more IBZ points to represent the symmetry of the system.

2. Is there no restriction like with monkhorst pack one should use even grids and with gamma points one should use odd grids. this is sort of what the vasp manual doesn't talk about but is somehting that i regularly see in literaure

the reason for asking this is that the vasp manual says accuracy only directly depends on thr number of kpts in the ibz which are often themselves unrelated to the k point divisions specified in the KPOINTS file

[juhL]

odd/even kpoint grids are indeed only a matter of time, not of accuracy (although i think you might encounter problems if you compare odd/even kpoint grids). generally, all even monkhorst-pack k-point grids can be shifted in such a way (towards the G point), that the amount of k-points in the IBZ does not increase compared to odd grids.

once again, the reason for your problem is simply a loss of symmetry in the case of even k-point grids. this is due to the symmetry of your chosen system and no general problem.

just read the VASP lectures/hands-on sessions, this issue is described there.