Yambo vs Gaussian

[martinspenke]

Dear Daniele,

many many thanks, very informative.
So when i increase the size of the box (let say from 30 to 40 a.u.), i have to increase the number of bands accordingly, too (let say from 300 to 500 bands) ?
Does this hold for dielectric matrix , too ? (So then i have to increase the dimension of dielectric matrix when increasing the size of box)

Should qp energies be also converged regarding charge oscillations as stated in your paper ???

Another point :
I used for ground state calculations in QE ibrav = 2 , which is FCC and very fast but when using ibrav = 1 , calculations are very slow and i have a bit different eigen-energies.
Should one better use ibrav = 1 or 2 in QE for box ???

Bests
Martin

[Daniele Varsano]

Dear Martin,
yes, if you want to compare two calculations with different box size, they have to be converged at the same level of accuracy.
So both bands (in G summation and dielectric matrix have to be increased). The same is true for the Gblk of the dielectric matrix. For the latter it is done
automatically if you indicate the Gblk in energy cutoff units (Ha, Ry, mHa) and not in number of Gvectors (RL). For the bands, you can have a look at the energy of the higher bands and include the number of bands of the bigger box calculation consistently.

About box geometry, in line of principle it should be equivalent, if you are considering an isolated molecule, I would use ibrav=1 as it is easier to visualize and control for instance the coulomb cutoff parameter. In general if you are using the coulomb cutoff, it is thought for cubic cells. Moreover if you get different energies in the ground state, I would say that two volume cell are not equivalent.

About charge oscillation it is not clear to me to what you mean.

Best,
Daniele

[martinspenke]

Dear Daniele,

is it a must to have this condition : 2Rc > system_size + spill out
or can it also be : 2Rc = system_size .

By which commands can i adjust the Gblk in energy cutoff units (Ha, Ry, mHa) and not in number of Gvectors (RL) ???

Bests
Martin

[Daniele Varsano]

Dear Martin,
if you use a sphere geometry "sphere xyz",
the condition is Rc > system_size + spill out this is because you want all the electrons of your system do interact each other. Then the density decrease exponentially so the condition can be softened. The factor 2 enter if you use a box, you have to double in in put the Lc for a reason which is internal to the code. If your geometry allows you to use the sphere, I would use it instead of the box, as the sphere is analytical.

Next it is important that the size of the cell is big enough to avoid interaction between images.
Roughly speaking: S=system size + spill out ; L=box side
S < Rc < L-S

Please note that this are not exact condition, the exact condition are stricter (but not needed to be fulfilled here): you can have a look to these papers:
Phys. Rev. B 73, 205119 (2006)
Phys. Rev. B 56, 14972 (1997)

In order to change the units, just put in the input (Ha, mHa, Ry) instead of RL that appears by default, eg:
NGsBlkXs = 5 Ha

Best,
Daniele

[martinspenke]

Dear Daniele,

your cell should be large enough that system-system distance is larger than 2Rc

Does this hold for only Box or for sphere xyz, too ? (i mean here this factor 2 is again for the box gemetry due to code, too ???)
I am using sphere xyz geometry.

How large should be this spill out ?

Bests
Martin

[Daniele Varsano]

Dear Martin,

a) the factor two it is for the box.
b) it is system dependent, anyway consider few au (3 or 4).

Best,
Daniele