BSE & RPA

[fatimazahra]

Hi
1) I did the calcul RPA. I did test the convergence of the value NGsBlkXd to find the optimal value. if I do the test with RL, the results are identical for all value of NGsBlkXd (3RL, 5RL, 9 RL, 15RL ...)
If I do the test with Ry, the resultat change when I change NGsBlkXd.

2) this value of gap (5.8 eV) obtained by GGA or GW :

% KfnQP_E
5.80000 | 1.000000 | 1.000000 | # [EXTQP BSK BSS] E parameters (c/v)
thank you in advance.

[Daniele Varsano]

Dear Fatima:

1) I did the calcul RPA. I did test the convergence of the value NGsBlkXd to find the optimal value. if I do the test with RL, the results are identical for all value of NGsBlkXd (3RL, 5RL, 9 RL, 15RL ...)
If I do the test with Ry, the resultat change when I change NGsBlkXd.

3,5,9,15 RL are very small values, and possibly you cannot appreciate the variation. If you do the same in Ry you are changing the number of RL vector by order of magnitude. In the report you have the conversion so you can look at how many RL do correspond your input values in Ry.

2) this value of gap (5.8 eV) obtained by GGA or GW :
% KfnQP_E
5.80000 | 1.000000 | 1.000000 | # [EXTQP BSK BSS] E parameters (c/v)

This value it is in input file and not in output. This variable is the scissor operator in a BSE calculations, and usually has to be set to the GW gap to correct the DFT value, if the scissor approximation is valid. Otherwise you need to calculate QP corrections for all the energy level and read the db.QP database using the

Code: Select all

KfnQPdb

variable.

Best,
Daniele

[fatimazahra]

Dear Daniele
1) if I used Ry, I find the same result of RL.
2) For studying the convergence of the G0W0 gap versus by checking how the direct and indirect band gap behave as a function of NGsBlkXd.
for BSE & RPA we check the output (result of dielectric function)
Thank you

[Daniele Varsano]

Dear Fatima,

1) if I used Ry, I find the same result of RL.

As it should, it is just different way to specify the size of the eps matrix to be inverted. RL (reciprocal lattice) is the number of G vectors, Ry it's an energy cutoff, ie include all the G vectors such that |G^2|/2 < Ecut

2) For studying the convergence of the G0W0 gap versus by checking how the direct and indirect band gap behave as a function of NGsBlkXd.
for BSE & RPA we check the output (result of dielectric function)

Sorry, but I can't understand what is the point.

Best,

Daniele

[fatimazahra]

Dear Daniele
how can I know the value of NGsBlkXd is optimal?
Thank you

[Daniele Varsano]

Dear Fatima,
This is very system dependent, so the only way is to perform several calculations increasing its values (eg 50 100 500 ... RL etc...) until your results does not change anymore in the order of the accuracy you want to achieve. Of course the bigger the values the heavier is the calculation in terms of cpu time and memory usage. These values depends a lot if you are dealing with a bulk (order of hundreds RL), or with a nanostructure (1D, 2D slabs or a molecule), in this case it could be of the order of thousands RL. But as I told you, there is not a rule and convergences tests are needed. Anyway I suggest you to have a look to the tutorials in the yambo web page, convergences studies are discussed there.
Hope it helps,
Daniele

[fatimazahra]

Dear Daniele
I have already seen the tutorial, but I did not understand, thank you for your help.