Convergence parameters for the screened e-e interaction

[jjmm1974]

Dear all

I am trying to perform a BSE calculation on an aromatic molecule. To do so, I need first to get the screened potential via "yambo -b". Two adjustable parameters appear then, "BndsRnXs" (i.e., the number of bands considered to calculate the screening) and "NGsBlkXs" (the dimension of the matrix representing the response function), if I understood well. I have some questions about this.

1) How to converge these values? My guess is that I have to choose reasonable values of both of them (I know that NGsBlkXs is lower here than for GW calculations) and then go forward to the BSE calculation (i.e., yambo -o b -k sex -y h) and check. If the results are not converged, then I have to come back to the "yambo -b" stage and rerun with other values. Am I right?

2) In relation to the previous, can I use lower values in the BSE calculation than those in the previous W calculation or they must be equal?

3) I have calculated the quasiparticle energies for the system under the PPA, which required the calculation of the static screening as well. I guess that I can use the screening calculated at the PPA stage directly in the BSE calculation, thus skipping the "yambo -b" stage. The price will be a much higher computation time, since the block size is much higher in PPA. Am I right?

Thank you in advance, and thank you for keeping the code and the forum.

Take care

Juanjo Meléndez
Department of Physics
University of Extremadura
Badajoz (Spain)

[Daniele Varsano]

Dear Juanjo,
1) You are right, anyway you can combine the two calculations in a single run:

Code: Select all

 yambo -b -o b -k sex -y h 

2) You can use lower value, I would say safely. Just have in mind that a calculation with BSENGBlk < NGsBlkXs can slightly differs with a calculation with the same BSENGBlk calculated with BSENGBlk =NGsBlkXs. The screening requires a matrix inversion and the same element of eps^-1 differs if it is calculated with two matrix of different size.

3) Yes you can use the pp databse, I would not say that the price is higher as you have already calculated it, and moreover you are more sure to have converged results. In general the convergence parameter they are not so different.

As you are dealing with an isolated molecule and yambo works in plane wave, I suggest you to use coulomb cutoff techniques (yambo -r in version 4.0 yambo -c in version 3.4) in oder to accelerate convergences with the respect the vacuum size. You can have a look to this paper. It is not well documented in the website but there are several post in the forum where you can have a look.

Best,
Daniele

[jjmm1974]

Thanks Daniele, it helps a lot.

Take care

Juanjo Meléndez
Department of Physics
University of Extremadura
Badajoz (Spain)