Absolute value of the imaginary part of the dielectric function

[Daniele Varsano]

Dear Quxiao,
yambo -p p and yambo -X p are totally equivalent, they do depend on the version of yambo you are using, the last version, if I'm not wrong reads correctly both command lines.
In both cases, you are adopting the plasmon pole approximation where the response function is calculated at RPA level.

For a 2D system the use of a Coulomb truncation is highly recommended as it speed up the convergence wrt the volume of the supercell, which is otherwise very low to converge.
Note that if you use a box shape cutoff also the Monte Carlo integration is needed as coulomb integrals are used to build up the coulomb cutoff potential.

Using explicit keywords you can use the following command:
yambo -r -hf -gw0 p -dyson n
or in a compacted form:
yambo -r -x -p p -g n
also:
yambo -r -x -X p -g n

should be equivalent.

Best,

Daniele

[Quxiao]

Dear Daniele：
Thanks for your help again.
I found that after I considered the truncation, I still used the convergence values of the previous parameters. For example, NGsBlkXp, BndsRnXp, and GbndRage all took the convergence values when the truncation was not considered for calculation. It is much larger than the band gap when truncation is not considered. I guess it is because the RandGvec option I took also needs to be converged. The third direction truncation in CUTBox is slightly smaller than that in the original cell, and the original cell is 28.35 , I take it as 28.3, is it possible ？
The Whether the value of 100RL needs to be increased for convergence test for the RandGvec？please give me some help again.
Thanks
Quxiao
BIT

[Daniele Varsano]

Dear Quxiao,

The third direction truncation in CUTBox is slightly smaller than that in the original cell, and the original cell is 28.35 , I take it as 28.3, is it possible ？

Yes this is ok.

When using the cutoff coulomb technique you are eliminating interaction among replicas and this gives you a larger gap.
If you do not use it and performs different calculation increasing the supercell volume you will see that the gap will increase, the convergence wrt the volume is extremely slow and the cutoff is meant to mitigate this problem.

Bands convergence it is independent from the potential you are using, you can have some dependencies on NGsBlkXp, but I would not expect big changes. RandGvec should not be a problem.

Best,
Daniele

[Quxiao]

Dear Daniele：
I seem to understand a little bit. Does yambo consider the Coulomb truncation more than the VASP calculation of GW? Yambo thinks that only adding a vacuum layer cannot completely eliminate the interlayer Coulomb interaction, so consider the Coulomb truncation more?
Because I found that the GW band gap without Coulomb truncation is roughly the same as the VASP calculation result, and the band gap after adding Coulomb truncation is larger. At first I thought that the parameters needed to be remeasured and converged, but now I want to understand, Think more about Coulomb truncation, right?
Of course, I still want to ask about the parameter of RandGvec. It is enough for the integral to calculate only one G vector at the gamma point, but why does the calculation of 100RL still appear in the example of yambo, and the impact is almost small?
Thank you for your selfless clarification！
Quxiao
BIT

[Daniele Varsano]

Dear Quxiao,
1) I never used VASP so I cannot say much. As far as I know, VASP does not consider any truncation but I can be wrong.
2)

Yambo thinks that only adding a vacuum layer cannot completely eliminate the interlayer Coulomb interaction, so consider the Coulomb truncation more?

+
This is documented in several papers e.g.:
Physical Review B 73 (20), 205119
https://doi.org/10.1103/PhysRevB.73.233103
2D Coulomb cutoff is routinely used (see e.g. the works done with BerkeleyGW in the Louie group, or in the group of Thygesen).

3)

It is enough for the integral to calculate only one G vector at the gamma point, but why does the calculation of 100RL still appear in the example of yambo, and the impact is almost small?

This depends on the size and shape of your BZ. Usually is enough to consider just 1 Gvec unless you have a very large amount of vacuum. Anyway, in the report, you can find the ratio between the integral done with Monte Carlo and as calculated with the trapezoidal rule which is used outside the BZ. For small q they differ, next for large q they tend to provide the same results and the ratio is similar to 1. You can estimate the smaller G along Z and compare it with the integral at the q value on plane similar to that G. Note that 1G vector does not mean at the gamma point, but for all the q points in first BZ. More G indicates that you consider a random integration also outside the 1st BZ:
Best,
Daniele

[Quxiao]

Dear Daniele：
Thank you so much for your reply！
I think I get it, despite the love for yambo philosophy, I still have a lot of lack of knowledge, thanks for your help again.
Best wishes
Quxiao
BIT

[Quxiao]

Dear Daniele：
In order to ensure the accuracy of my calculation, there is still a question that I need to help me with. When I calculate, I set the plane wave truncation (FFTGvecs) to be consistent with QE, but when calculating Exchange self energy (EXXRLvcs) and Correlation part of self energy ( VXCRLvcs), whether the truncation used is also set to be consistent with FFTGvecs, so that convergence can be fully guaranteed? Or are EXXRLvcs and VXCRLvcs automatically generated by default?
Thank you for your patience in helping me！
Best wishes
Quxiao
BIT

[Daniele Varsano]

Dear Quixiao,
EXXRLvcs and VXCRLvcs have a default that of course can be changed by input if needed (memory issue etc.)
In general, it is safe to set them to the MaX Gvecs, i.e. the cutoff of the density used in QE , i.e. 4 times the cutoff of wfc, especially if you are using PBE.
About the Vxc in the report the E_xc is reported and it can be compared to the one reported in the QE output. EXXRLvcs usually converges with less g vectors and convergeces tests can be performed, but if you do not have memory issues you can set to the max g vectors as it is not computationally demanding.

Best,
Daniele

[Quxiao]

Dear Daniele：
Thanks a lot for your help!
With your help, I straightened out the conversion of yambo4.4 and 5.0 to calculate GW, but now I am facing another problem, the command I use to calculate BSE with 4.4 version is yambo -r -b -ob -yd -k sex -V qp, the generated file has the parameter BSENGBlk, but when I converted to version 5.0 and used yambo -r -X s -ob -yd -k sex -V qp, I found that I did not find the corresponding BSENGBlk parameter, which makes me very confused，please give me some help to get the BSENGBlk parameter in yambo.in with the 5.0,
it seems to me that between 4.4 and 5.0 the command to calculate BSE, only the command to generate the inverse dielectric matrix has changed, right?
My second confusion is about BSEmod. When I use the above command to calculate BSE，in the yambo.in file I generated with yambo4.4 before, the automatically generated BSEmod is ‘retarted’, but now I use the corresponding command of yambo5.0 to generate it is indeed ‘resonant’, I noticed this , but I don't quite understand what the corresponding will be and where the difference is
Thanks for your help and have a nice life ！
Best wishes
Quxiao
BIT

[Daniele Varsano]

Dear Quixiao,

Code: Select all

yambo -r -X s -o b -k sex  -y d 

gives you the right input.
It is important you define the kernel "-k sex" after the "-o b" command (ie optics in transition space).

Anyway, by typing "yambo -h" you have the help to use the command lines.

About BSEmod it is possible that the default has changed, you can just change the value of the variable to retarded.
The difference is in the Green function ordering, you can safely set to "retarded" in particular if you are interested in the real part of the response function.

Best,

Daniele