Excitonic effects in metallic systems ?

[Daniele Varsano]

Dear Haseeb,
in order to add smearing to your spectra the variable is BDmRange, while DmRngeXs provides a smearing for the calculation of the screening.
Yambo reads the databases present in the SAVE directory and in the "dir" indicated by -J "dir".
If you specified the -J "dir" in a first run the BSE database are stored in "dir" so in the second run you need to specify the same directory.
In any case, you can see what Yambo is doing by looking at the report and log files.
Best,
Daniele

[haseebphysics1]

Dear Daniele,
Thanks once again for your kind reply.

Sorry for misleading info! Actually I was varying BDmRange not DmRngeXs. I mistakenly wrote DmRngeXs. I'm not specifying any -J option for these calculations so I'm hoping everything will go the SAVE folder, but when I run with different BDmRange (again without specifying -J) calculations started from scratch (BSE). This was strange for me so I was asking...

Moreover can you please tell that what we can do if we are not getting the intensity of the peak close to experimental (or other's people theoretical work)? I have changed BDmRange already to remove some unwanted peaks but it further lowers the intensity and I'm very much away from Si absorption curve.

Attached is the pic obtained so far...

Regards,
Haseeb Ahmad
MS Physics,
LUMS, Pakistan

[Daniele Varsano]

Dear Haseeb,

so I'm hoping everything will go the SAVE folder,

You can easily check it by looking at the SAVE directory and check if the databases are effectively there.

but when I run with different BDmRange (again without specifying -J) calculations started from scratch (BSE). This was strange for me so I was asking...

You can check in the report what is happening.
When a database is read it is indicated with RD and the parameter are reported. If something is inconsistent with the input file it should appear a warning (WARN) or an error (ERR). In the latter case, the calculation is repeated.

I have changed BDmRange already to remove some unwanted peaks but it further lowers the intensity

That's reasonable, what is conserved is the area beneath the peaks.

About the intensity, in the pic you posted you have very large smearing and as stated above this will lower the intensity of the single peaks.
I suggest you to not play with the smearing and check your convergence parameters (k-points, bands in BSE, band in the screening etc.).
Most probably you are not including enough bands. I can see from the report that you are considering Silicon as a supercell (8 atoms cubic) so you have 16 occupied bands, instead of 2 atom in FCC.
Consider that a fast calculation in the minimal cell with these parameters already provides a reasonable absorption spectrum, and it needs 3 occupied and 3 unoccupied bands. In the case of supercell you should multiply it at least by 4.

Code: Select all

optics                       # [R OPT] Optics
em1s                         # [R Xs] Static Inverse Dielectric Matrix
bss                          # [R BSS] Bethe Salpeter Equation solver
bse                          # [R BSE] Bethe Salpeter Equation.
bsk                          # [R BSK] Bethe Salpeter Equation kernel
Chimod= "hartree"            # [X] IP/Hartree/ALDA/LRC/BSfxc
BSEmod= "causal"             # [BSE] resonant/causal/coupling
BSKmod= "SEX"                # [BSE] IP/Hartree/HF/ALDA/SEX/BSfxc
BSSmod= "h"                 # [BSS] (h)aydock/(d)iagonalization/(i)nversion/(t)ddft`
BSENGexx= 1000          RL    # [BSK] Exchange components
BSENGBlk=27            RL    # [BSK] Screened interaction block size
Gauge= "length"              # [BSE] Gauge (length|velocity)
KfnQPdb= "E < 03GoWo_PPA_corrections/ndb.QP"              # [EXTQP BSK BSS] Database
% BEnRange
  1.00000 | 6.0000 | eV    # [BSS] Energy range
%
% BDmRange
  0.10000 |  0.10000 | eV    # [BSS] Damping range
%
BEnSteps= 101                # [BSS] Energy steps
% BLongDir
 1.000000 | 0.000000 | 0.000000 |        # [BSS] [cc] Electric Field
%
% BSEBands
   2 |  7 |                 # [BSK] Bands range
%
BSHayTrs=0.001
% BndsRnXs
   1 |  20 |                 # [Xs] Polarization function bands
%
NGsBlkXs=27            RL    # [Xs] Response block size
% DmRngeXs
  0.00100 |  0.00100 | eV    # [Xs] Damping range
%
% LongDrXs
 1.000000 | 0.000000 | 0.000000 |        # [Xs] [cc] Electric Field
%

Best,
Daniele

[haseebphysics1]

Thanks very much for the detailed reply.

I can see from the report that you are considering Silicon as a supercell (8 atoms cubic) so you have 16 occupied bands, instead of 2 atom in FCC.

From basic solid-state physics course, I remember that FCC contains 4 atoms per unit cell? Are you using reduced cell somehow to reduce calculation time? Moreover, I have not made the supercell, I just downloaded the cif from COD and run the calculation on that. Visualizing the cif just showed FCC lattice in diamond structure (4 FCC atoms and 4 on tetrahedral sites). Therefore, I never thought about supercell in these calculations?

And how I can make two-atoms FCC Si lattice and also, in general, reduce the number of atoms of any lattice for future calculations of heavier systems (considering symmetry has already been applied).

Regards,
Haseeb Ahmad,
MS- Physics,
LUMS, Pakistan

[haseebphysics1]

Dear Sir,

I also think the problem is in BSEBnds and k-points because these two values I couldn't converge properly! I have also attached the results (plots) of my convergence tests.

K-points don't seem to be converging even with 20x20x20 k-mesh and its calculation took a long time (1.5 days)! And also calculations time also gets a lot bigger if I use BSEBands large as it should!

Since I've done convergence tests while plotting the BSE spectra each time so my strategy was to vary the parameter which I want to converge while keeping all other parameters fixed

but such that calculation time remains very reasonable

(5-10 minutes!).

Regards,
Haseeb Ahmad,
LUMS, Pakistan

[Daniele Varsano]

Dear Haseeb,
the procedure you are using to converge parameters is ok.
Anyway, my suggestion is to consider the minimal cell for Silicon which implies to deal with 2 atoms per cell instead of 8.
This will speed up your calculation by one order of magnitude.
Best,
Daniele

[haseebphysics1]

Dear Daniele,

Kindly try reply to my previous post (2nd last), if you find some time to do so, please!

From basic solid-state physics course, I remember that FCC contains 4 atoms per unit cell? Are you using reduced cell somehow to reduce calculation time? Moreover, I have not made the supercell, I just downloaded the cif from COD and run the calculation on that. Visualizing the cif just showed FCC lattice in diamond structure (4 FCC atoms and 4 on tetrahedral sites). Therefore, I never thought about supercell in these calculations?

And how I can make two-atoms FCC Si lattice and also, in general, reduce the number of atoms of any lattice for future calculations of heavier systems (considering symmetry has already been applied).

Regards,
Haseeb Ahmad,
LUMS, Pakistan

[Daniele Varsano]

Dear Haseeb,
this is not related to the code. As you can argue the smaller the number of atoms (electrons) the faster is the calculation.
Silicon has a diamond structure and it can be obtained with two atoms in the primitive cell, try to extract it as an exercise.
Note that Silicon is used in many tutorials of QE where you can find the lattice set up:
https://www.quantum-espresso.org/resour ... cture1.pdf
https://eamonnmurray.gitlab.io/modellin ... 2/#silicon

Best,

Daniele