Negative eigenvalues for excitation

[subhayan]

Dear developer,

I am running BSE calculation on a molecule but I keep getting negative eigenvalues. The yambo input file contains the value 'BSEmod= "retarded" '. Does this mean I am running with Tamm-Dancoff approximation? If so, I think I should not be getting any negative eigenvalues. Could you please let me know what I might be doing wrong here? I am attaching the input file and the eigenvalue file. Please let me know if you need anything else.

Best regards,
Subhayan

[Daniele Varsano]

Dear Subhayan,
there are some problems in your input file:
1) The first and most severe is that you are solving the BSE on top of Kohn Sham energies and not quasiparticle energies and this is not consistent. You need to calculate quasiparticle energies first either by a GW calculation or modelling that with a scissor operator. This will open your gap and probably will bring the excitation energies to positive values. In the way, you are doing you are including many-body effects in the electron-hole interaction but not in the calculation of the quasiparticle levels. If you look at the BSE equation you have quasi-particle energies ih the diagonal part.

2) You are building a BSE kernel, including only one reciprocal vector in the electron-hole screened interaction: this is for sure far away from convergence.

I invite you to follow step by step the tutorials present in the yambo page: here you can find the last tutorials prepared for a recent school:
http://www.yambo-code.org/wiki/index.php?title=ICTP2020

Please note that in Tamm-Dancoff approximation the eigenvalues are real, but this does not assure they are positive definite. Note that "retarted it is not synonym of Tamm-Dancoff, but actually you are applying the Tamm-Dancoff as you are not considering coupling terms.

Best,
Daniele

[subhayan]

Dear Dr. Varsano,

Many thanks for this. A very quick question in this regard. Should I include more than one reciprocal vectors in electron-hole interaction even when I am solving for a molecule? The input file was generated using the command "yambo -o b -k sex -y d". Except for an addition of the keyword "WRbsWF", all the parameters have been left unchanged. In this case, should I modify the value for "BSENGBlk" by hand?

Best regards,
Subhayan

[Daniele Varsano]

Dear subhayan,
yes, it is a parameter that should be converged, actually, it is the more expensive part as you have to calculate the screening matrix.
For molecules it could be large as you have vacuum in your supercell which implies a strong anisotropy and many g vectors.
I strongly suggest you follow one of the tutorials in the yambo wiki page:
http://www.yambo-code.org/wiki/index.ph ... r_overview
they are done for bulk and nanostructures but the same strategy applies for molecules.

Best,
Daniele

[subhayan]

Ah, okay I understand that we need more reciprocal vectors. I guess I should run a convergence test for this. Could you please let me know why, in the automatically generated (-o b -k sex -y d) yambo input file, there are so many reciprocal lattice vectors for the exchange part but only one for the screened Coulomb part?

Best regards,
Subhayan

[subhayan]

Dear Dr. Varsano,

Thanks for this. Sorry, I actually posted the last message before I saw your reply. I am going through the tutorial in detail.
Thanks again for this

Best regards,
Subhayan

[Daniele Varsano]

Dear Subhayan,
please note that BSENGBlk should be changed accordingly to NGsBlkXs, i.e. cannot be lower than the Gblk included in the calculation of the screening function. As default value, we set BSENGBlk=1 RL, as it is system dependent and there is not a standard value that work for all the systems, as said before it depends on the inhomogeneity of the system.
Best,
Daniele