composition of exciton by ypp

[caowd]

Dear

I am following the suggestion of using "ypp -e s" and "ypp -e a" to calculate the composition of the excitons.
However, after I typed "ypp -e s", there is nothing except the following lines printed on the screen
----------------------------

____ ____ ___ .___ ___. .______ ______
\ \ / / / \ | \/ | | _ \ / __ \
\ \/ / / ^ \ | \ / | | |_) | | | | |
\_ _/ / /_\ \ | |\/| | | _ < | | | |
| | / _____ \ | | | | | |_) | | `--" |
|__| /__/ \__\ |__| |__| |______/ \______/



<---> [01] CPU structure, Files & I/O Directories
<---> [02] Y(ambo) P(ost)/(re) P(rocessor)
<---> [03] Core DB
<---> :: Electrons : 16.00000
<---> :: Temperature [ev]:0.9500E-3
<---> :: Lattice factors [a.u.]: 10.29523 8.91593 16.18626
<---> :: K points : 396
<---> :: Bands : 24
<---> :: Symmetries : 12
<---> :: RL vectors : 93301
<---> [04] K-point grid
<---> [M 0.035 Gb] Alloc qindx_X ( 0.012)
<---> [M 0.046 Gb] Alloc qindx_S ( 0.012)
<---> [M 0.166 Gb] Alloc qindx_B ( 0.119)
<01s> :: Q-points (IBZ): 396
<01s> :: X K-points (IBZ): 396
<01s> [05] CORE Variables Setup
<01s> [05.01] Unit cells
<01s> [05.02] Symmetries
<01s> [05.03] RL shells
<01s> [05.04] K-grid lattice
<02s> [05.05] Energies [ev] & Occupations
<02s> [WARNING]Metallic system
<02s> [06] Game Over
----------------------------

I didn't find any index of excitons.
Before I typed that command, I didn't prepare any input file like ypp.in.
But I did have WRbsWF tag in my BSE calculation. I am not sure what's going wrong.

Could anyone please help me on this?

[Daniele Varsano]

Dear Wen-Dong Cao,
have you diagonalized the BSE matrix. I mean did you used the "d" option instead of the "h" option in the BS solver BSSmod?

Code: Select all

>ypp -e s 

should produce output files containing excitation energies sorted by energy and intensity and relative indexes.
From your standard output it looks ypp did not find the databases containing the diagonalized matrix in the SAVE directory. Other possibility is that you used the -J option
in the yambo run. In this case you need to use the same option for ypp.

Code: Select all

>YPP -e s -J namedir

Best,
Daniele

[caowd]

Dear Daniele Varsano

Thank you for the quick response.

I did use the "d" option and the report file is attached.

I didn't use the -J option.

What would the filename of the databases containing the diagonalized matrix look like?

[Daniele Varsano]

Dear Wen-Dong Cao,
indeed, from the report file I can see that the BS matrix has not been diagonalized and it has not calculated neither.
This is because you have the Chimod variable set as independent particle. So only the IP spectra only is produced and no matrix to diagonalize is needed, and of course
there will be no exciton with the IP approximation.
You need to set:

Code: Select all

Chimod= "Hartree"

Next, please consider that setting:

Code: Select all

 BSKmod= "Hartree"

you are doing an RPA calculations including local field effects. I do not know if you are interested in analyzing these spectra. In order to look at excitons you need to solve the Bethe-Salpeter equations, hence you need to consider the screened electron-hole attraction, this is achieved by setting:

Code: Select all

 BSKmod= "SEX"

The produced databases will be:
/SAVE//ndb.BS_Q1_CPU_*
/SAVE//ndb.BS_diago_*

In line of principle with your input the code should have done the calculation anyway, as with BSKmod= "Hartree" the screening it is not needed, so it is not important in which approximation it has been calculated, and should have produced the RPA results and not the IP, we will fix this condition, anyway this is a quite unusual combination of approximations.

I invite you to look at the theory and the tutorials.

Best,

Daniele

[caowd]

Thank you very much. I will try your advice.

I am indeed not familiar with the BSE theory and I am doing response calculations under RPA+local field effect approximations.

Thank you for your good advice and I will study the BSE theory soon.

[Daniele Varsano]

Ok,

RPA+local field effect approximations.

In this case removing the Chimod= "IP" variable from the input it should work. You can also remove the em1s runlevel and you will save time, as the screening matrix it is not needed for a RPA+local field approximation.

You can also calculate RPA+local field in reciprocal space using:

Code: Select all

yambo -o c

which is totally equivalent provided that you activate the coupling in the BSEmod. Which of the two implementation is faster it depends on your system (ie if you need many bands or many Gvectors in general I would say that the way you are doing is faster)

Best,
Daniele

[caowd]

Dear Daniele Varsano

Thank you very much. Your advice is very good.
I am following your advice and doing a test calculation by yambo.

My yambo input file is
----------------------------
optics # [R OPT] Optics
bse # [R BSE] Bethe Salpeter Equation.
bsk # [R BSK] Bethe Salpeter Equation kernel
bss # [R BSS] Bethe Salpeter Equation solver
FFTGvecs= 4679 RL # [FFT] Plane-waves
ElecTemp= 0.000000 eV # Electronic Temperature
BoseTemp=-1.000000 eV # Bosonic Temperature
BS_CPU= "32,1,1" # [PARALLEL] CPUs for each role
BS_ROLEs= "k,eh,t" # [PARALLEL] CPUs roles (k,eh,t)
BS_nCPU_invert= 0 # [PARALLEL] CPUs for matrix inversion
BS_nCPU_diago= 0 # [PARALLEL] CPUs for matrix diagonalization
WRbsWF # [BSS] Write to disk excitonic the FWs
BSEmod= "causal" # [BSE] resonant/causal/x_coupling/xc_coupling
BSKmod= "Hartree" # [BSE] IP/Hartree/HF/ALDA/SEX
BSSmod= "d" # [BSS] (h)aydock/(d)iagonalization/(i)nversion/(t)ddft`
BSENGexx= 5817 RL # [BSK] Exchange components
BSENGBlk= 1 RL # [BSK] Screened interaction block size
% BEnRange
0.00000 | 20.00000 | eV # [BSS] Energy range
%
% BDmRange
0.10000 | 0.10000 | eV # [BSS] Damping range
%
BEnSteps= 400 # [BSS] Energy steps
% BLongDir
0.000000 | 0.000000 | 1.000000 | # [BSS] [cc] Electric Field
%
% BSEBands
8 | 24 | # [BSK] Bands range
%
----------------------------

The calculations are still running now and so far there is no errors. However, there is a small problem. The ndb.BS_Q1_CPU_* files in the SAVE/ directory seem to be very large (estimated to be almost about 1T when the calculations are complete). Is that normal? And is there any solution to decrease the file size? Because the disck which I am allowed to use in the cluster is not very large.

Wendong

[Daniele Varsano]

Dear Wendong,
The BSE matrix can be very big, as the size is Nc*Nv*Nk_bz.
If you have disk space problem you can avoid to write it on disk, by adding the variable:

Code: Select all

DBsIOoff= "BS" 

This is an optional io variables which you can activate by adding "-V io" when building up your input file.

In this way the BS matrix it is not written on disk. The drawback is that you cannot restart the calculation if it expires before terminating the task, so be sure your queue have enough time for that (you can have a look to the estimated time in the logs).

Best,

Daniele

[caowd]

Dear Daniele Varsano

Thank you very much!!!
I will try this tag.

Wendong

[caowd]

Dear Daniele Varsano

With the tag you suggested, the calculations seem to be OK and the "Kernel loop" stage is finished. But the job stops at the next stage "BSE solver".

And the error message is

Code: Select all

srun: error: cn12101: task 20: Segmentation fault

I am not sure whether this error is due to the large memory cost. Could you please help me about that?

My input file yambo.in is

Code: Select all

optics                       # [R OPT] Optics
bse                          # [R BSE] Bethe Salpeter Equation.
bsk                          # [R BSK] Bethe Salpeter Equation kernel
bss                          # [R BSS] Bethe Salpeter Equation solver
FFTGvecs=  4679        RL    # [FFT] Plane-waves
ElecTemp= 0.000000     eV    # Electronic Temperature
BoseTemp=-1.000000     eV    # Bosonic Temperature
BS_CPU= "32,1,1"                  # [PARALLEL] CPUs for each role
BS_ROLEs= "k,eh,t"                # [PARALLEL] CPUs roles (k,eh,t)
BS_nCPU_invert=  0           # [PARALLEL] CPUs for matrix inversion
BS_nCPU_diago=  0            # [PARALLEL] CPUs for matrix diagonalization
WRbsWF                       # [BSS] Write to disk excitonic the FWs
BSEmod= "causal"             # [BSE] resonant/causal/x_coupling/xc_coupling
BSKmod= "Hartree"                # [BSE] IP/Hartree/HF/ALDA/SEX
BSSmod= "d"                  # [BSS] (h)aydock/(d)iagonalization/(i)nversion/(t)ddft`
BSENGexx= 5817        RL    # [BSK] Exchange components
BSENGBlk= 1            RL    # [BSK] Screened interaction block size
% BEnRange
  0.00000 | 20.00000 | eV    # [BSS] Energy range
%
% BDmRange
  0.10000 |  0.10000 | eV    # [BSS] Damping range
%
BEnSteps= 400                # [BSS] Energy steps
% BLongDir
 0.000000 | 0.000000 | 1.000000 |        # [BSS] [cc] Electric Field
%
% BSEBands
  9 | 24 |                   # [BSK] Bands range
%
DBsIOoff= "BS"

Both FFTGvecs and BSENGexx have been decreased to reduce the memory cost and the report file is attached.
In fact, I have also compared my yambo.in file with that in the GaSb tutorial but I still have no ideas where goes wrong.