Parameters from Quantum ESPRESSO

Concerns any physical issues arising during the setup step (-i option). This includes problems with symmetries, k/q-point sets, and so on. For technical problems (running in parallel, etc), refer to the Technical forum.

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

haseebphysics1
Posts: 169
Joined: Sat Aug 17, 2019 2:48 pm

Parameters from Quantum ESPRESSO

Post by haseebphysics1 » Sat Oct 05, 2019 8:09 pm

Respected Sir,
I am doing DFT calculations using Quantum Espresso. But I am uncertain about a couple of things before doing massive ground-state calculations so that I can safely go with Yambo BSE steps because these variables also appear in Yambo keywords so I wanted to have them right from the ground state!

Q1: I'm modeling a semiconductor with a well-defined bandgap (DFT is predicting a very close value after applying corrections). Now, I have also introduced doping that will be for sure either p-type or n-type. Moreover, due to semiconductor, I am using the keyword

Code: Select all

occupations               = "fixed"
It worked in pure case but in doping calculation, the following error came in SCF
...
Error in routine electrons (1):

charge is wrong: smearing is needed ...
I think due to addition or deficiency of charge QE is trying to force me to have some sort of smearing and degauss. Right? But after applying some fictitious temperature (degauss value) the fermi position will change. In the fixed occupation, the Fermi level is just at the top of VB (as it should) and in smearing case it is in around the middle. Therefore, how I will be able to relate with pure bandstructure case? My purpose is to lower the bandgap after doping and make conclusions on the basis of the movement of Fermi level, but here fermi level is moving due to smearing too instead of just dopants.

2: I have not used nspin parameter despite Oxygen and other transition metals in the system! Actually, the system total energy was not changing when nspin was 2. Moreover, I have also not used spin-orbit interactions despite having W (Tungsten)and Bi (Bismuth) in the system! Does Yambo consider theses parameters for excited-state calculations?

4: To correct the bandgap I applied Hubbard parameter (DFT+U) but the system total energy is increasing as I am applying U values to single or both the atoms in the system! Is it okay? (I know Yambo will discard U parameter for generating the database).


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

andrea.ferretti
Posts: 206
Joined: Fri Jan 31, 2014 11:13 am

Re: Parameters from Quantum ESPRESSO

Post by andrea.ferretti » Sun Oct 06, 2019 2:54 pm

Dear Haseeb Ahmad,

thanks for writing. Before addressing in more details your questions, please note that some of these topics are probably better suited
for the Quantum ESPRESSO forum (pw_forum).

Let me start from the second point:
(2) I would definitely check the nspin=2 case (don't worry for spin-orbit at first). This is mandatory if you have orbitals that carry unpaired electrons.

(1) If you have a system with impurities I would avoid occupations="fixed" at first, using occupations="smearing" instead. Together with nspin=2,
you'll be able to see where the Fermi level lies and whether you have genuinely partially occupied states or not.
You should pay attention to the value of the smearing parameter (degauss) used (the smaller the better, ideally).

For the sake of interfacing with yambo, I think a small (say 0.001 Ry) but non zero smearing value is ideal (in case you can use it).
In fact, occupations fixed + nspin=2 I think leads to QE asking for tot_magnetization and using two-fermi levels, which would make yambo complain.

Yambo + smearing is ok, though, especially with defects, you want to keep the broadening small.

(3) question missing ?

(4) DFT+U: IMPORTANT: as you say, Yambo does not account for DFT+U, but it is important to realise that this is harmful.
A missing cancellation of Vxc terms will occur eg when computing GW corrections. In the calculation of the response function
the non-local contribution of DFT+U is not included.
You may want to run BSE on top of DFT+U, though I would strongly recommend to double and triple check your results.
All in all, until a proper support of DFT+U is provided, I would try to avoid running Yambo on top of DFT+U calculations

(4bis) I would not worry too much about the DFT total energy increasing after the inclusion of U (actually, I have the feeling that
in the Cococcioni De Gironcoli spirit this is what should happen)

take care
Andrea
Andrea Ferretti, PhD
CNR-NANO-S3 and MaX Centre
via Campi 213/A, 41125, Modena, Italy
Tel: +39 059 2055322; Skype: andrea_ferretti
URL: http://www.nano.cnr.it

haseebphysics1
Posts: 169
Joined: Sat Aug 17, 2019 2:48 pm

Re: Parameters from Quantum ESPRESSO

Post by haseebphysics1 » Sun Oct 06, 2019 8:30 pm

Thanks, dear Andrea.
Haseeb Ahmad
MS - Physics,
LUMS - Pakistan

haseebphysics1
Posts: 169
Joined: Sat Aug 17, 2019 2:48 pm

Re: Parameters from Quantum ESPRESSO

Post by haseebphysics1 » Sat Oct 19, 2019 11:51 am

Dear Yambo developers,

I have another question which is somewhat relevant from DFT ground state run! Since I am uncertain about it, that's why I considered it better to ask here.

First of all, I have already calculated BSE spectra of my semiconducting bulk material. BSEBands was varied to have converged value (among other convergence parameters). But I am now confused about BndsRnXs (polarization function bands).

Q1: In nscf from QE I selected nbnd = 250, which included 90 empty bands while 160 are occupied (nspin=1)! Are these number of bands enough or should I also go back to have its convergence as well.

Q2: Selecting BndsRnXs value greater then nbnd of QE will be meaingless, correct? I included all the bands in the screening to aviod its convergence tests and to have more precise value, but I'm not sure how much it can effect my calculation time and system's memory!

Q3: Actual problems occurs when I include spin-orbit coupling (SOC) into the picture, number of bands gets double for non-magnetic non-collinear SOC calculations, i.e, 320 and if I do nbnd = 380 or something bigger like this (high number of empty bands) then the following error comes in Q.E.

Code: Select all

from cdiaghg : error #       987
S matrix not positive definite
Which means I am unable to include too much empty bands in nscf calculation and hence in optical calculations! I read in a online tutorial that Q.E. code also gets angry if you include too much bands.

Any remedy or suggestion will be extremely appreciated in this regard.

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

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: Parameters from Quantum ESPRESSO

Post by Daniele Varsano » Sun Oct 20, 2019 10:58 am

Dear Haseeb,

A1:
Yes, convergences on bands entering in the polarization function (BndsRnXs ) need to be checked.
A2:
Selecting BndsRnXs value greater then nbnd of QE will be meaingless, correct?

Yes, as they are not present!
I included all the bands in the screening to aviod its convergence tests and to have more precise value, but I'm not sure how much it can effect my calculation time and system's memory!
OK, but it is possible you need more. How many bands you need it is something that varies from system to system. The more the bands you need the more memory is needed, anyway you can distribute memory among CPU's by using a parallelization strategy on bands "c" and "v" roles in X_ROLEs variable.
A3:
First, before including more bands, you can include less and see how much the results differ from your actual result, this gives you an indication if your calculation is converged and if not, how far you are from convergence.

About your QE error, there is something you can try to solve the problem:
https://lists.quantum-espresso.org/pipe ... 41445.html

If it does not work you can ask for help in the QE mailing list.
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/

andrea.ferretti
Posts: 206
Joined: Fri Jan 31, 2014 11:13 am

Re: Parameters from Quantum ESPRESSO

Post by andrea.ferretti » Sun Oct 20, 2019 10:14 pm

Dear Haseeb,

Re Q3, the problem comes from a failure of the Davidson algorithm, which is common when asking for a lot of empty states.
In those cases, just revert to diagonalization="cg" (conjugate gradient style diagonalization). This is slower, but designed to provide an answer.

take care
Andrea
Andrea Ferretti, PhD
CNR-NANO-S3 and MaX Centre
via Campi 213/A, 41125, Modena, Italy
Tel: +39 059 2055322; Skype: andrea_ferretti
URL: http://www.nano.cnr.it

haseebphysics1
Posts: 169
Joined: Sat Aug 17, 2019 2:48 pm

Re: Parameters from Quantum ESPRESSO

Post by haseebphysics1 » Tue Nov 19, 2019 8:48 pm

Dear Andrea,

Thanks very much for the QE support as well!

I have also tried "cg" type diagonalization but that didn't work too! And "S matrix not positive" error came at the very start!

I was requesting 750 bands out of which 430 were empty bands. Moreover, I have also tried

Code: Select all

 diago_david_ndim = 2 
as david diagonalization mode but that also ended in the "eigenvalues convergence" problem...

Any suggestion in this regard will be highly appreciated and beneficial.

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

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: Parameters from Quantum ESPRESSO

Post by Daniele Varsano » Wed Nov 20, 2019 9:13 am

Dear Haseeb Ahmad,
The best place where to post your question and receive useful answers is the quantum espresso mailing list.
Have a look to the address reported below for the mailing list usage and subscription:
https://www.quantum-espresso.org/forum

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: 169
Joined: Sat Aug 17, 2019 2:48 pm

Re: Parameters from Quantum ESPRESSO

Post by haseebphysics1 » Thu Nov 21, 2019 11:03 am

Thanks, dear Daniele,

I had already looked into the PW forum as this issue was raised there as well and recently also contacted a Q.E. developer.

Dear Daniele,

Kindly allow me to ask relevant questions to Yambo optical calculation. I know that numbers of bands can only be determined from convergence tests. So, the point that I want to clarify for myself is:

If I do the nbnd = 350 and include 0 - 400 in Polarization function bands, will this be a meaningless calculation? Or Yambo can extrapolate eigenvalues somehow?

And what will be the good nbnd value to have in nscf if I have 320 occupied bands! I could not go beyond 350 bands before running any error. (I am doing scf with no nbnd given)


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

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: Parameters from Quantum ESPRESSO

Post by Daniele Varsano » Thu Nov 21, 2019 12:18 pm

Dear Haseeb,
If I do the nbnd = 350 and include 0 - 400 in Polarization function bands, will this be a meaningless calculation? Or Yambo can extrapolate eigenvalues somehow?
It will be meaningless
And what will be the good nbnd value to have in nscf if I have 320 occupied bands! I could not go beyond 350 bands before running any error. (I am doing scf with no nbnd given)
Hard to answer as it is system dependent. Anyway, 350 total bands over 320 occupied seem to me to be quite a small number.
Sounds strange anyway that you cannot calculate more.

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