strange gw+bse results on two-layered 2D materials

[ljzhou86]

Dear Developers,

Recently, I am using yambo to perform gw+bse calculations on a 2D material, considering the layer-depended excitonic effect. As shown in the follwoing figures, I summarized the main results as follows,

single-layer (1L): DFT gap:0.8508 eV; G0W0 gap: 2.06 eV; the first gw+BSE optical exciton peak: 1.48 eV, so the exciton binding energy: 2.06—1.48=0.58 eV
double-layer (2L): DFT gap: 0.8289 eV; G0W0 gap: 1.9 eV; the first gw+BSE optical exciton peak: 1.96 eV, so the exciton binding energy: 1.9 — 1.96 = —0.06 eV
Triple-layer (3L): DFT gap: 0.8142 eV; G0W0 gap: 1.65 eV; the first gw+BSE optical exciton peak: 1.44 eV, so the exciton binding energy: 1.65 — 1.44 = 0.21 eV


It seems the gw+bse results for 2L are very strange: blue-shifted first optical exciton peak and thus negative exciton binding energy. I have double checked the results and parameters used in my calculations, but I am still confused why I got so strange results for 2L. I attached the pwscf and yambo inputs and r_* log files for 2L calculations. Could you help me to pinpoint this problem?

Thanks a lot,
Liujiang Zhou

[ljzhou86]

Sorry not to attach the input and oupout files, please see it in attachment.

[Daniele Varsano]

Dear Zhou,

from the output you posted I can't see the negative binding energy you mention: the red line (BSE) is always al lower energy than the green line (IP).
In order to check the exact excitation energy you should diagonalise the excitonic matrix in order to have access to the eigenvalues: you have a 20K dimension matrix, even if a bit hard should be still possible to diagonalise (use "d" instead of "h") in the BSSmod variable.
If the trend is not what you expect it is recommendable to check the convergence parameters.
Please note that in the calculation of the screening you did not use the coulomb cutoff potential:

Code: Select all

   [RD./SAVE//ndb.em1s]----------------------------------------
    Brillouin Zone Q/K grids (IBZ/BZ):   61   576    61   576
    RL vectors                   (WF): 10503
   *WRN* Coulomb cutoff potential         :none

and I think it should activated for consistency with the rest of the calculation.

Best,

Daniele