GW to BSE

[kxzhou]

Dear all,
I have studied the content related to BSE on the Yambo website, but I still have some unclear questions and would appreciate your answers.

1. In my initial QE calculations, I set `ecutwfc` to 70, but subsequent tests revealed that 40 is already sufficient for convergence. Will this cause a significant increase in computational load or memory usage during the GW+BSE calculations in Yambo?

2. My GW calculation has not yet converged. Can I start testing the convergence of various parameters for BSE? Are these parameters interrelated?

3. What are the advantages of using the scissor operator compared to directly using ndb.qp? I feel that once the GW ndb.qp is calculated, it can be directly used for BSE calculations, which seems more accurate. Could the potential mismatch between the k-point grids of GW and BSE cause issues?

4. Can the dynamically screened Coulomb potential obtained during the GW_PPA calculation be directly used for BSE calculations? Compared to calculating the statically screened Coulomb potential directly in BSE, which approach is better and requires less computational effort?

Best,
Kexin

[Daniele Varsano]

Dear Kexin,

1) No, as the cutoff in the dielectric matrix is generally much lower than the ones of the WFs. Moreover, if needed you can also lower the cutoff of the FFT grid. Note that 70Ry is a reasonable cutoff.

2) You can check the bands to include in the BSE, while the parameters of the screening are the same you found in GW.

3) The advantage is that, once scissor operator has been verified to be valid, you can avoid calculating QP energies for many bands/k points. Of course this is an approximation and using the calculaed QP is more accurate.

4) Yes you can use it. BSE relies on static screening only, and this matrix is part of the ndb.pp database. This avoids recalculating the static screening.
Usually GW and BSE calculations are done on the same grid, if you need to increase the grid for the BSE you cannot use the ndb.pp database used for GW calculated on a smaller grid.

Best,

Daniele

[xjxiao]

Dear Daniele,
I have read your discussion and also have some questions to ask. Thanks.

1) No, as the cutoff in the dielectric matrix is generally much lower than the ones of the WFs. Moreover, if needed you can also lower the cutoff of the FFT grid. Note that 70Ry is a reasonable cutoff.

I noticed that reducing `FFTGvecs` can decrease the computational cost. It seems that the larger the `ecutwfc` is, the greater the computational cost will be. How can I determine how much `FFTGvecs` can be reduced? Can the difference in the bandgap values from GW calculations be used as a criterion?

FFT.png

2) The advantage is that, once scissor operator has been verified to be valid, you can avoid calculating QP energies for many bands/k points. Of course this is an approximation and using the calculaed QP is more accurate.

Is it possible to interpolate the `ndb.qp` obtained from GW calculations for use in BSE calculations? If so, how can this be done? Sorry I couldn't find the corresponding tutorial on the Yambo official website.

Best,
Xiao

[Daniele Varsano]

Dear Xiao,

1) Yes, you can check the convergence of the QP gap wrt the FFTGvecs. Also in the report you can find the value of the orthonormalisation of the wfs check that it is not much degraded when reducing the cutoff

2) Yes, this is done automatically by yambo when running the BSE.

Best,
Daniele

[xjxiao]

Dear Daniele,

1) Yes, you can check the convergence of the QP gap wrt the FFTGvecs. Also in the report you can find the value of the orthonormalisation of the wfs check that it is not much degraded when reducing the cutoff

Is the "orthonormalisation of the wfs" referring to the "Normalization (few states)" in the figure below?

36691.png

9001.png

I find it a bit strange that when FFTGvecs changes from 36691 to 9001, both the minimum and maximum values increase. Is the reduction in FFTGvecs reasonable?

2) You can check the bands to include in the BSE, while the parameters of the screening are the same you found in GW.

Are the parameters BSENGblk and BSENGexx in BSE also determined by the relevant parameters in GW?
Thanks!
Best,
Xiao

[Daniele Varsano]

Dear Xiao,
1) Yes. Nothing strange, it tells you that with the smaller cutoff you still have a reasonable orthonormalization.
2) No, they are the cutoff in the exchange and correlation kernel.

Best,
Daniele

[kxzhou]

Dear Daniele，
Thanks for your help!

4) Yes you can use it. BSE relies on static screening only, and this matrix is part of the ndb.pp database. This avoids recalculating the static screening.
Usually GW and BSE calculations are done on the same grid, if you need to increase the grid for the BSE you cannot use the ndb.pp database used for GW calculated on a smaller grid.

1) How can the ndb.pp obtained from G0W0 calculations be utilized? Is it possible to proceed as follows: copy it to the BSE folder and then use -J BSE to point to this folder during BSE calculations? Since the energy used to compute this dielectric function does not incorporate G0W0 corrections, would it be less rigorous if only the diagonal terms E_ck - E_vk in the BSE Hamiltonian include quasi-particle energy corrections?

2) When calculating BSE for 2D materials, can rim and rim-w be used to accelerate k-point convergence, similar to G0W0 calculations?

3) How should the following two parameters be selected?

BSEmod_BSKmod.png

Best,
Kexin

[Daniele Varsano]

Dear Kexin,

1) Yambo reads all the databases presents in the SAVE directory and the one indicated with -J. So either you copy in the SAVE or in the BSE directory.
In the BSE equaiton the E_ck - E_vk term is present only in the diagonal.

2) Yes, even if this has not be extensively tested.

3)
BSKmod= SEX this is the BSE equaition.
BSEmod: this is the time ordering: resonant includes the c->v transitions only. Retarted: include also the antiresonant term v->c and include the time-ordering.
Both are in the framework of the Tamm Dancoff approximation.
Coupling includes also the coupling term between v->c and c-v and the matrix is not hermitian anymore, and it is doubled in size.

Best,
Daniele

[kxzhou]

Dear Daniele

4) Yes you can use it. BSE relies on static screening only, and this matrix is part of the ndb.pp database. This avoids recalculating the static screening.
Usually GW and BSE calculations are done on the same grid, if you need to increase the grid for the BSE you cannot use the ndb.pp database used for GW calculated on a smaller grid.

In the GW calculation process, are the generated ndb.pp* files computed based on the DFT bands or the GW-corrected bands?
If they are based on the DFT bands, I am considering whether I should use the ndb.QP obtained from the GW calculation to recompute the static dielectric screening function for subsequent BSE calculations. Would this approach potentially yield more accurate results?
Thanks in advance!
Best,
Kexin

[Daniele Varsano]

Dear Kexin,

it is computed with KS energies, this is the standard GoWo.
You can also insert GW eigenvalues in the Green function and/or in the screening up to self consistency evGW or evGoW or evGWo.
This is the so-called eigenvalue self-consistent GW. It is not guaranteed it provides more accurate results, as it usually overestimates the gap, because you are
anyway neglecting the vertex in the self-energy.
You can find a tutorial here:
https://wiki.yambo-code.eu/wiki/index.p ... alues_only

Best,

Daniele