RPA paramters

Deals with issues related to computation of optical spectra, in RPA (-o c) or by solving the Bethe-Salpeter equation (-o b). Includes local field effects, excitons, etc.

Moderators: Davide Sangalli, andrea marini, Daniele Varsano, andrea.ferretti, Conor Hogan, myrta gruning

Post Reply
haseebphysics1
Posts: 50
Joined: Sat Aug 17, 2019 2:48 pm

RPA paramters

Post by haseebphysics1 » Fri Feb 14, 2020 12:07 pm

Dear Yambors,

I am doing a optical absorption calculation using IP approximation. And I have some points for clarification.

1: Is IPA same as RPA? Or RPA includes more advances approximations as well?

2: I have read the following tutorial:

http://www.yambo-code.org/wiki/index.ph ... icle_level

Here it is written that "Non-local commutator" can play a huge difference in the computational time, so, I want to know how can I guess that is this term important for my system or not? Is the inclusion of this term makes the results more accurate? If I will have to calculate the spectra with both NL commutator and without it then there is no benefit, I mean, to make calculation without it. Since my system is large that's why I am worried about it.

3: Why we are going to q || (1 0 0) direction in the above-mentioned tutorial to not include the "Non-local commutator"? This is the direction of the applied electric field to the material. Isn't it?

4: In optics runlevel:

Code: Select all

% [b]QpntsRXd[/b]
 1 | 1|                   # [Xd] Transferred momenta
%
ETStpsXd= [b]1001  [/b]             # [Xd] Total Energy steps
I want to know why we want to have just q=0 part? (meaning of QpntsRXd?) And Energy steps 1001 (above tutorial) are just the number of steps. Can't I select 1000 for q=0? And finally which factors decide that we only need to know the long-wavelength spectra?

5: My calculation is taking the enormous time, I am attaching the necessary files as attachement to have your feeedback to optimize things. I can reduce the number of G-vectors (which is the default now). What else can I do, I mean, NL commutator and only q=0 calculations?

Thanks,
Haseeb Ahmad,
LUMS - Pakistan
You do not have the required permissions to view the files attached to this post.
Haseeb Ahmad
MS - Physics,
LUMS - Pakistan

Online
User avatar
Daniele Varsano
Posts: 2177
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: RPA paramters

Post by Daniele Varsano » Fri Feb 14, 2020 3:25 pm

Dear Haseeb,
Is IPA same as RPA? Or RPA includes more advances approximations as well?
Here in literature, there is a bit of confusion, in some works you can find IPA sepctra indicated as RPA.
In Yambo language, anyway, when we speak of RPA we refers to it when Local field effects are included:
see here differences in step 6 and 7:
http://www.yambo-code.org/wiki/index.ph ... tandalone)
Non-local commutator: I want to know how can I guess that is this term important for my system or not?
Hard to say the effect as it is system dependent, anyway take in mind that this term will affect the oscillator strengths only and not the excitation energies. Moreover, note that you need to calculate it once for all.
Why we are going to q || (1 0 0) direction in the above-mentioned tutorial to not include the "Non-local commutator"? This is the direction of the applied electric field to the material
Yes, this is the direction of the applied electric field (light polarization). Note that the direction of the field and inclusion or not of the non-local commutator are not related at all. In that part of the tutorial the non-local commutator is switched off by hand ( mv SAVE/ns.kb_pp_pwscf SAVE/ns.kb_pp_pwscf_OFF) and then the result is compared for absorption in the chosen direction with the case where it was included.
I want to know why we want to have just q=0 part? (meaning of QpntsRXd?) And Energy steps 1001 (above tutorial) are just the number of steps. Can't I select 1000 for q=0? And finally which factors decide that we only need to know the long-wavelength spectra?
Up to you to choose the response for any q point you have in the sampled q-grid. If you are interested in optics you need to select q=0 (which is QpntsRXd=1), think about the momentum carried by a photon, so you want a limit for q->0. If you are interested in electron-energy-loss spectra you can choose any q. Energy steps you can put the number of steps you want to calculate the spectra in the selected interval of energies (EnRngeXd).
My calculation is taking the enormous time:...
Yes, number of G vectors (FFTGvecs) will speed up the calculation of the dipoles. Only q=0 (ie QpntsRXd=1|1).
Reduce number of bands, unless you are interested in higher energy part of the spectrum.
Reduce ETStpsXd (now you are calculating the spectrum for 2000 frequency and probably you do not need this definition, note that the spectrum is smeared by DmRngeXd. Finally, for IPA spectrum you can also switch to a transition space representation which is most probably faster. You should do a Bethe-Salpeter like calculation setting an IP kernel BSKmod=IP

Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

haseebphysics1
Posts: 50
Joined: Sat Aug 17, 2019 2:48 pm

Re: RPA paramters

Post by haseebphysics1 » Fri Feb 14, 2020 11:19 pm

Thanks for providing the answers in detail.

Dear, I am a small question (I don't have learn about this theory so far):
this term will affect the oscillator strengths only and not the excitation energies.
.

So, I want to know what is the trend or relation of the non-local commutator with oscillator strength?

Q: Is oscillator strength increases with the addition of "non-local commutator" or is it inversely proportional?

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

Online
User avatar
Daniele Varsano
Posts: 2177
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: RPA paramters

Post by Daniele Varsano » Sat Feb 15, 2020 2:57 am

Dear Haseeb,
In the case of diamond the effect of the non-local commutator is to decrease the oscillator strength, see Fig. 5.5 of Stefan Albrecht PhD thesis:
https://etsf.polytechnique.fr/system/fi ... brecht.pdf
the same is true for other insulators and for second-order susceptibility:
PHYSICAL REVIEW B 78, 245124 􏰔2008􏰕
I do not know if it can be also mathematically proved.
Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

haseebphysics1
Posts: 50
Joined: Sat Aug 17, 2019 2:48 pm

Re: RPA paramters

Post by haseebphysics1 » Tue Feb 18, 2020 9:31 am

Dear Daniele,

Do we need to keep to include a lot of empty conduction bands for BndsRnXd or only valance bands are needed?

% BndsRnXd
start number | end number | # [Xd] Polarization function bands
%

I have 160 conduction and 250 conduction bands! To avoid convergence tests of this parameter (it takes a lot of time for a single run!), is it wise to take these numbers to stay in the safe limit of convergence?


% BndsRnXd
50 | 350 | # [Xd] Polarization function bands
%

And similarly, with the above-mentioned band range, I could not increase the following parameter's value

FFTGvecs= 20 Ry # [FFT] Plane-waves
NGsBlkXd= 5 Ry # [Xd] Response block size

while ground state DFT was performed with 100 Ry. So, I have already lowered the parameters. What would you recommend for the theses parameter's convergence?

Thanks,
Haseeb Ahmad
MS - Physics,
LUMS - Pakistan

Online
User avatar
Daniele Varsano
Posts: 2177
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: RPA paramters

Post by Daniele Varsano » Tue Feb 18, 2020 1:29 pm

Dear Haseeb,
you cannot include only valence bands in BndsRnXd, otherwise, you do not have transitions.
In the "IP" the number of conduction bands is related to the energy range you want to calculate your spectrum, the more bands you insert
the larger will be the energy of the valence-conduction transitions (Ec-Ev).
Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

Post Reply