Yambo Community Forum

Dear developers,

I have done some BSE tests with a hydrogen-adsorbed graphene structure, and I used yambo -b -o b -y h -V qp to initiate the calculations. The variable "BndsRnXs" has been tested with two different values: 1 | 20 | and 1 | 50|, and I got probably strange spectrum curves. So, my question is that is this result reasonable? and what is the reason for this result? The yambo.in is posted as following:
------------------------------------------------------------------------------------------------------------
em1s # [R Xs] Static Inverse Dielectric Matrix
optics # [R OPT] Optics
bse # [R BSK] Bethe Salpeter Equation.
bss # [R BSS] Bethe Salpeter Equation solver
KfnQPdb= "none" # [EXTQP BSK BSS] Database
KfnQP_N= 1 # [EXTQP BSK BSS] Interpolation neighbours
% KfnQP_E
0.300000 | 1.000000 | 1.000000 | # [EXTQP BSK BSS] E parameters (c/v)
%
% KfnQP_W
0.000 | 0.000 | 0.000 | 0.000 | # [EXTQP BSK BSS] W parameters (c/v)
%
KfnQP_Z= ( 1.000000 , 0.000000 ) # [EXTQP BSK BSS] Z factor (c/v)
BSresKmod= "xc" # [BSK] Resonant Kernel mode. (`x`;`c`;`d`)
BScplKmod= "none" # [BSK] Coupling Kernel mode. (`x`;`c`;`d`;`u`)
% BSEBands
15 | 20 | # [BSK] Bands range
%
BSENGBlk= 101 RL # [BSK] Screened interaction block size
BSENGexx= 6473 RL # [BSK] Exchange components
XfnQPdb= "none" # [EXTQP Xd] Database
XfnQP_N= 1 # [EXTQP Xd] Interpolation neighbours
% XfnQP_E
0.300000 | 1.000000 | 1.000000 | # [EXTQP Xd] E parameters (c/v)
%
% XfnQP_W
0.000 | 0.000 | 0.000 | 0.000 | # [EXTQP Xd] W parameters (c/v)
%
XfnQP_Z= ( 1.000000 , 0.000000 ) # [EXTQP Xd] Z factor (c/v)
% QpntsRXs
1 | 21 | # [Xs] Transferred momenta
%
% BndsRnXs
1 | 50 | # [Xs] Polarization function bands
%
NGsBlkXs= 100 RL # [Xs] Response block size
% LongDrXs
1.000000 | 0.000000 | 0.000000 | # [Xs] [cc] Electric Field
%
BSSmod= "h" # [BSS] Solvers `h/d/i/t`
% BEnRange
0.00000 | 10.00000 | eV # [BSS] Energy range
%
% BDmRange
0.100000 | 0.300000 | eV # [BSS] Damping range
%
BEnSteps= 1000 # [BSS] Energy steps
% BLongDir
1.000000 | 0.000000 | 0.000000 | # [BSS] [cc] Electric Field
%
---------------------------------------------------------------------------------------------------------
NOTE THAT the only difference in my two test calculations is the value of "BndsRnXs" as mentioned above. Attached please find the BSE plot.
Many thanks in advance!

WEI Wei
postdoc
Institute of electrochemistry
Ulm University
Germany

Hallo Wei

That parameter (Polarization function bands) is related to the screening in W. Clearly including 20 bands is not sufficient since using 50 bands leads to huge differences in the (absorption?) spectrum. You should continue the convergence study I guess. In parallel you should also see the behavior with NGsBlkXs, since the two parametrs are related.

Best
m

Dear Myrta,
Thank you very much for you reply!
It seems that the adsorption spectrum converges with "BndsRnXs" set at 1 | 50 |, since I got almost the same specturm with several increased velues of this variable. Attached please find the specturm plot with "BndsRnXs" set at 1 | 200 |. So, questions come:

1) Is this spectrum correct with excitonic effects form a BSE calculation? It seems that the spectrum with 1 | 20 | of "BndsRnXs" more reasonable.

2) I have tested variables of "BndsRnXs" and "NGsBlkXs" within BSE calculations upon LiF adsorption specturm, with larger AND smaller values than the suggested ones in the tutorial, it gave the same results. So, in other words, it seems that the spectrum should not change with respect to different "BndsRnXs" and "NGsBlkXs" provided the other variables converged. Since I got significantly different spectrum plots, there may be some other not-converged variables in my current calculation, isn't it?

I am looking foward to you reply, thank you very much!

And any replies are appreciated!

Hallo Wei

WEI Wei wrote: 2) I have tested variables of "BndsRnXs" and "NGsBlkXs" within BSE calculations upon LiF adsorption specturm, with larger AND smaller values than the suggested ones in the tutorial, it gave the same results. So, in other words, it seems that the spectrum should not change with respect to different "BndsRnXs" and "NGsBlkXs" provided the other variables converged. Since I got significantly different spectrum plots, there may be some other not-converged variables in my current calculation, isn't it?

Please mind that the calculation of LiF in the tutorial is an example with the parameters tuned so that is possible to make the calculation in few minutes on your PC and yet getting the excitonic effects. In this sense LiF is also a "good" system, having strong excitonic effects and allowing to do that. Other systems are much more nasty to converge and only at convergence the spectrum "makes sense". Anyway LiF in the tutorial cannot be taken as an example of converged calculation.
Second, convergence behavior and convergence parameters depend on the system. The general rule is to make a convergence study of all the parameters that may influences the accuracy of the calculation.

WEI Wei wrote:1) Is this spectrum correct with excitonic effects form a BSE calculation? It seems that the spectrum with 1 | 20 | of "BndsRnXs" more reasonable.

I do not know! If the calculation is well converged with all the parameters, then that is the result.
Anyway, when you say "more reasonable" do you have some reference or it is just what do you expect?
To inspect at excitonic effetcs anyway you should compare the results from BSE with the independent particle ones (in the same output files).

Best
m

Dear Myrta,

I think I have known something ... ....
Thank you so much!

Dear Developers,

I have done convergence test for the Perovskite structure in GW calculation I need help for the following with respect to BSE calculation :
1) Do I have to do convergence test again for the parameters which I have done convergence test for them in GW and appear in BSE calculation i.e NGsBLK , BndsRn Xp and FFTGVecs ?

2)Are the following parameters, I have to do convergence test for them
a) BSENGexx
b)BSENGBlk
c) BSE Bands
d) BndsRnxs

3) Are there relation with respect to the selection values and others between
BSE GW
a) BSENGexx and EXXRLvcs
b) BSENGBlk and NGsBlk (Because I know the values of NGsBlk comes from RL shells from report file, what a bout BSENGBlk)
c) last one, BndsRnXs and BndsRnXp.

I waiting for your available suggestion, Really I need to clarify these for me because I don't have reference for the strucure I am dealing with to compare the results. Thank you in advance.

Dear Khaled,
as usual convergence tests are needed. I inveite you to have a look to the documentation here in
order to understand to what the variable stands for.
In particular:

FFTGVecs: This is not strictly a convergence parameter. You can leave it to the default (maximum value according to your cutoff used in the DFT). It is reduced in case of memory issues, so it has to be checked that the quality of the calculation it is not lost.

a) BSENGexx and EXXRLvcs : These are different, the first is the G vector entering in the exchange part of the BSE kernel, the secondo is the HF self energy. Anyway both are not very expensive and you can use a lot of them.
b) BSENGBlk and NGsBlk : Again, the first G vector in direct term of the kernel of the BSE, The second the block considered in the screening for W (I do not undesrstand what you mean with RL shell). BSENGBlk cannot be bigger than NGsBlk (Gvectors considered in the screening). See eq for W_res in the documentation. NGsBlk is the dimension of the matrix for epsilon, BSENGBlk is the sum for building up the W term.
c) BndsRnXs and BndsRnXp. (s) stand for "static", (p) for plasmon pole. Both are the band to build up the polarization. In BSE only the static screening is included. In general id you converged BndsRnXp for the GW corrections, the BndsRnXs for the BSE should be very similar.

Finally, of course BSE Bands gives you the size of the BSE matrix and has to be converged, starting from the bands near to the gap and adding up empty and occupied states till convergence.

As a general suggestion, I invite you to have a look to the documentation and lecture notes, in order to not run calculation as a "black box".
Hope it helps,

Best,
Daniele