Absorption from BSE

Submitted by dinesh169 on Sat, 04/15/2017 - 00:49

Forums 

User questions

Hello Everyone,

I am trying to study GaSe layered structure with 18A vacuum in z-direction. I am trying to calculate absorption spectra using kernel and BSE. Here are the input files

kernel.inp

number_val_bands 10
number_cond_bands 10

screened_coulomb_cutoff 3.0
bare_coulomb_cutoff 10.0

cell_slab_truncation
use_symmetries_coarse_grid
comm_mpi
screening_semiconductor
dont_use_hdf5

absorption .inp

diagonalization

number_val_bands_coarse 10
number_val_bands_fine 5
number_cond_bands_coarse 10
number_cond_bands_fine 5

use_symmetries_coarse_grid
no_symmetries_fine_grid
no_symmetries_shifted_grid

dont_use_hdf5
number_eigenvalues 10

screening_semiconductor
write_eigenvectors 10
use_velocity
gaussian_broadening
energy_resolution 0.015

when i use less bands for number_val_bands_conduction_fine, I get real part of epsilon form (at omega = 0 in absorption_eh.dat from BSE) close to 1 which is expected because of vacuum in the system. to check the convergence I increased number_val/conduction_bands_fine/coarse in kernel and absorption calculation, then the real part of epsilon goes to 400, instead of converging.

The coarse grid is 12x12x1 and fine grid is 50x50x1.

Could you please comment on it? What am I doing wrong.

Best wishes,
Dinesh

Submitted by babarker on Mon, 04/17/2017 - 12:18

Hello Dinesh,

Your Coulomb cutoffs are much too low to draw any meaningful conclusions about convergence. The bare coulomb cutoff should be nearly the same as (better yet, identical to) the kinetic energy cutoff from your DFT calculations, which should be an order of magnitude larger. A screened coulomb cutoff of 3.0 Ry is also about a factor of ten too small. If you used this to calculate epsilon, I would suggest recalculating.

A paper on converging 2D absorption calculations: https://journals.aps.org/prb/abstract/10.1103/PhysRevB.93.235435

A paper on converging 3D bulk semiconductors (supplemental information): http://iopscience.iop.org/article/10.1088/0953-8984/25/10/105503/meta;js...

Best,
Brad

Submitted by dinesh169 on Tue, 04/18/2017 - 03:33

Hello Brad,

Thank you for your reply.

Regarding the screened Coulomb cutoff (SCC), i used the energy of the highest occupied band (500, this is how I choose SCC in my calculations) in the epsilon calculation and keep it same in kernel calculation as well.

Best wishes,

Submitted by babarker on Wed, 04/19/2017 - 14:49

Hello Dinesh,

Using the energy of the highest unoccupied band to determine the screened coulomb cutoff is correct. The problem is that the large vacuum region is going to necessitate using many states. 500 unoccupied bands can be a well-converged value for bulk 3D systems, but a quasi-2D system with vacuum will require more.

For instance, in PRB 93 p. 235435 (https://journals.aps.org/prb/abstract/10.1103/PhysRevB.93.235435), the authors used 6000 empty states and a 35 Ry cutoff.

The "SAPO" mean-field utility that comes along with BerkeleyGW can help generate the plane-wave-like empty states, provided the kinetic energy cutoff is sufficient. It allows for generation of additional empty states in a BerkeleyGW-binary format WFN file without having to run any actual DFT calculations, by generating planewave empty states up to the kinetic energy cutoff, then orthogonalizes them via Gram-Schmidt. (And, optionally, can include resonances when dealing with molecular systems, though that's not relevant to the present case.)

For details on SAPO, check out https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.107.186404

Best,
Brad