composition of the excited state

[Andy]

Hello Daniele,
thanks again for your answer
Concerning my question "But why is a TDDFT calculation (yambo -o b -k alda) as much time consuming as a BSE calculation (yambo -o b -k sex)?"
I'm a little bit confused now as you refer to my output files which belonged to another question as they are done in gspace. But maybe I just didn't understand your answer? The output in this post shows a "yambo -o b -k alda" and a "yambo -o b -k sex" result and both calculations took nearly the same time.

Essentially you are using independent particle approximation, and this explain why your "Hartree" and "ALDA" are the same.

To get it right - I do a independent particle approximation without local field effects as NGsBlkXd (responsable for LF, http://www.yambo-code.org/tutorials/fan ... ons/h2.php) and FxcGRLc (responsable for creating a real fxc) is set to 1, correct? That's why I also did a calculation with NGsBlkXd=113 - here the 2nd and 4th column are different so that I'm not doing independent particle calculations any longer. As already mentioned there is no difference between Alda and Hartree then. Only if I increase FxcGRLc, I see a difference between Alda and Hartree. So, if I understand you correctly if "XC-kernel RL size" is 1, I've nearly no kernel->RPA/Hartree. If I increase FxcGRLc I can construct a real Alda fxc functional?
If I do a Alda and a Hartree calculation in transition space I can also see clear differences, as you already hinted.

In transition space you are including local filelds (via exchange g vectors) and all the G's for the building of the ALDA kernel.

So in transition space local fields are "automatically" included? If NGsBlkXd helps me in gspace to include such effects for what is the variable NGsBlkXs in transition space then?

If you set the NGsBlkXd to the same values you set the EXXGvec in the transition space you will realize than in G space is even more time consuming.

In transition space I reduced BSENGexx to 1759 - is this the quantity you mean here? If I set NGsBlkXd to 1759, I think I've also to increase the variable FFTGvecs from 1067 to a greater value because I read in the documentation (for NGsBlk):
"A smaller number with respect to FFTGvecs is generally needed to correctly describe the LF effects"
correct?

Coming back to the comparison with the literature - ignoring the green and red line as thy are only independent particle calculations, I focused on the difference between my pink and my blue line. In the literature only the BSE result has a much better visible peak in front of the higher one, whereas in my spectrum the shape of the "first" peak of the BSE curve is comparable to the shape of the Alda result. Is this an artefact because of a not converged BSE result or do I have to include something else? I attached you the corresponding outputfile for the pink and the blue curve. Furthermore I increased NGsBlkXs but this had only a small positive influence.

My professor wanted me orginally to do a TDDFT calculation with the LRC kernel:
- is this kernel not available in transition space because you don't have a "q" Vector here?
- why are the excitation information - which I originally asked for - only available in transition space and not in g-space? Is it just not implemented, yet?

Thanks again for your help and have a nice weekend,

Andreas

[Daniele Varsano]

Dear Andreas,
careful again. In your BSE you are totally neglecting the screening: # | BSENGBlk= 1 you are not essentially including the electron-hole attraction.
I suggest you to take your time and go through the tutorial present in the web page.
Anyway, about the timing, in both calculation ALDA and BSE you need to build up a kernel. In ALDA you are including it automatically, in BSE you are
neglecting the screening so the calculation is faster.

So, if I understand you correctly if "XC-kernel RL size" is 1, I've nearly no kernel->RPA/Hartree. If I increase FxcGRLc I can construct a real Alda fxc functional?

Yes, exact.

So in transition space local fields are "automatically" included? If NGsBlkXd helps me in gspace to include such effects for what is the variable NGsBlkXs in transition space then?

No, they are not automatically included, they are controlled by the BSENGexx variable, and you are setting them correctly.
The NGsBlkXs in BSE is needed to build up the screening, which is calculated at RPA level (in g space), next you have to use the dielectric matrix via ( BSENGBlk) and you are not doing it.

"A smaller number with respect to FFTGvecs is generally needed to correctly describe the LF effects"
correct?

Yes correct, but I would not touch the FFTGvecs, if you do not have memory problem. Lowering that, you can have problem with the normalization of the wave functions. So, when possible, I would not touch it at all.

Furthermore I increased NGsBlkXs but this had only a small positive influence.

Yes, because you are not using it as I told you before, in this way you are including only the (0,0) components of esp-1 that slightly vary when increase the size of the matrix.

- is this kernel not available in transition space because you don't have a "q" Vector here?

No, just because it has not been implemented.

why are the excitation information - which I originally asked for - only available in transition space and not in g-space? Is it just not implemented, yet?

No, because in G space you are solving a Dyson equation X=X^0+X^0(v+Fxc)X , in reciprocal space, and you look for the pole of the X, so you do not have any information about the KS states. In transition space, you are in the base of valence-conduction KS states, and you get the information you want by the eigenvectors exactly in that space.

Hope it helps,

Best,

Daniele

[Andy]

Hi Daniele
thanks again
As you suggested I'll first have a closer look on the yambo tutorials and I will then come back to this topic.

Kind regards,

Andreas

[Andy]

Hi Daniele,
I had a closer look on the tutorials now. I tried to reproduce (without GW correction) the blue curve for the BSE result shown in the literature picture. I tell you my steps - then you can easier see where my mistake might be:

-in order to calculate the statically screened electon-electron interaction I did a "yambo -b" calculation where I reduced NGsBlkXs to 1759. Furthermore I limited BndsRnXs to the range 1 and 32. This calculation gave me a "ndb.em1s" file in the SAVE directory.
-after this calculation I did the "real" BSE calculation with "yambo -o b -k sex -y h". I attached you the output-file.

The position of my peaks are wrong as I didn't use a scissor shift or a GW calculation. However - it disturbs me that the peak between 2 and 3 eV is much smaller than the peak in the blue curve of the literature result. Furthermore the absolute values differ by ~25%. As I used comparable high values for "BSENGexx", "BSENGBlk", "BSEBands" and "NGsBlkXs" I do not expect a convergence issue for the parameters I focused on. Furthermore I included also the "WehCpl" flag. Can you tell me what went wrong? What can I try to reproduce the blue curve?

Thank you very much again and kind regards,

Andreas

[Daniele Varsano]

Dear Andy,
your procedures looks me ok!

Sincerely I would not be disturbed about the intensity of the peak, they have not much physical meaning, what is meaningful is the area beneath the peaks (oscillator strengths). The max value of the intensity depends on the broadening, and you are assigning it artificially via a damping: the input variables BDmRange.
I do not know what values has been assigned in the figure of the literature (most probably 0.1eV), you can check it if it is reported. Lowering the damping (you have 0.3) you will have higher peaks. Another variable that can affect the the absolute max intensities is the k-point sampling: again you should check you have a converged value.
A last comment: the intensity is not insensitive to the excitation peak position: have a look to the f-sum rule (or Thomas-Reiche-Kuhn sum rule).

Best,

Daniele

[Andy]

Hi Daniele,
thanks for your propositions - I did some further convergence tests and found out that a k-point grid of 12x12x12 points reproduces much better the expected features than the 8x8x8 grid used before
Kind regards,
Andreas