Theorectically same system producing different absorption

[alphahui]

Dear All,

I have come across some very odd results. With the same molecule, R-methyloxirane, in the same cubic cell, but only slightly moved and rotated, the absorption and CD calculated changed dramatically. I am currently using Yambo version 5.2.4 and QE version 7.2. Please see attached the input I used for QE and Yambo. I was testing around with the "BLongDir" setting, to see if set as (1,1,1) could solve the problem, but it is the same as (1,0,0) results. The Yambo input and output with "_0" stands for case with (1,0,0) and "_1" stands for case with (1,1,1). In all calculation I only added 100 extra unoccupied orbitals.

If we look purely at the absorption results:
- Figure of raw absorption output against energy for original system (orig.png) and rotated system (rot.png). The "BLongDir" is (1,0,0)
- Figure of raw absorption output against energy for original system (orig_qbox.png).in Qbox

Due to files limit, I cannot upload the (1,1,1) figures, but they are the same as the (1,0,0). I have also did some test with another program, Qbox, to see if this behaviour is common. Please see attached the input I used.

With the limit, I cannot upload the rotated system image, but it is really close to the orig_qbox.png as well. I have been puzzled by this issue for the past few weeks. Any advice would be great and thank you for the time!

Regards,
Alpha

[Daniele Varsano]

Dear Alpha,

actually if you are rotating the molecule keeping the same cartesian direction for the long wavelenght limit you are probing the absorption in non equivalent direction (the same is true when looking at the (1,1,1) direction). Have you tried to average your results in space? i.e. the average of the three calculations (1,0,0); (0,1,0); (0,0,1).
Probably this can be done using just one input file adding the keyword:

Code: Select all

BSEdips= "trace"

I've seen you inserted the keyword "trace" in BSEprop but this is not recognised there. You can have a look at the end of the output file where the mirror of the used input file variables are reported.

There could be a problem or symmetry breaking for a given polarization direction, but this should not appear if you are using a truncated coulomb potential.
BTW, for a molecule I suggest you to use the "sphere" shape as it is analytical. The "ws cutoff" should work but is not extensively used.

Best,

Daniele

[alphahui]

Dear Professor Varsano,

Thank you very much for the reply. I have followed your advice and after setting 'BSEdips= "trace"', the results doesn't seems to change. I have did the following few tests, all with this new change and some other parameters different:

1. I used my input from original post with the new 'BSEdips= "trace"', then I ran the calculations of (1,0,0); (0,1,0); (0,0,1); (1,1,1) for both the original and the rotated (also moved) system, they are still the same as before. Also, changing the direction doesn't seems to change anything. As the files size for all these tests are too large, I did not include them here.

2. Please see attached the result and input of this run. I try to copy the input from https://wiki.yambo-code.eu/wiki/index.p ... _molecules again with all the new parameters you suggested. Here are also set the BDmRange back to 0.10000. The results of the original and rotated systems now look much more similar, but I believe it maybe only due to the damping applied, blurring away the differences. Thus, I ran another run with the same input, but the damping back to my previous 0.00001.

3. Please see attached the result and input of this run. With the damping reduce. The results are once again same as my previous input. Proving that the "similarity" found in 2 is really just the damping issue.

In conclusion, changing the parameters discussed, the problem persisted. Or may I ask if the subtle difference in 2 is actually accepted as some sort of error, and the result of 2 is already enough? As originally, I set really low damping to try to obtain the "raw data", then apply the gaussian broadening myself. But after applying the broadening myself, the same system with slight differences in location and rotation would show significant differences. Based on this, should I use the damping in yambo instead of applying them by myself? Thank you for the time on my silly question.

Regards,
Alpha

[Daniele Varsano]

Dear Alpha,

Let me clarify a few points.
First, a correction: R-methyloxirane is a chiral molecule and has no symmetries other than identity. Absorption along different directions should provide different results, since your results shows the spectrum is independent of direction, the trace averaging was already being performed correctly before my suggestion — the BSEdips= "trace" keyword was redundant, and I apologize for the confusion.

Regarding the differences between original and rotated geometry: the fact that the low-energy onset is identical is reassuring. The differences at higher energies are likely a numerical artifact stemming from how Yambo groups quasi-degenerate transitions — even meV differences in KS eigenvalues between the two setups can change how oscillator strengths are summed within degenerate clusters. Could you verify that the KS eigenvalues of the two systems are numerically identical?
Finally, regarding the broadening: I would recommend using the damping directly in Yambo rather than post-processing with a Gaussian. With an extremely small damping like 0.00001 Ha you are over-resolving numerical noise, making spurious differences appear significant.

Best,
Daniele