Dear all,
Since the first step in EPW is calculating the electron phonon matrix elements which are computationally demanding, I need to clarify , once and for all, a few things regarding the 'checking for convergence' process.
1 Do the 'decay.H' and 'decay.dynmat' depend on the interpolated q and k grids or they only depend on the coarse grids? That is do we have to check them versus the cahnges in the fine q and k grids too?
2 Regarding the convergence of electronic relaxation time and the phonon self energy, do we have to re calculate the elph matrix elements each time we change either the phonon or the electronic fine grids?
3 Once convergence has been achieved with respect to phonon and electronic fine grids, and we have to check convergence w.r.t degaussw and degaussq, Is there any need to recalculate those matrix elements?
4 Is the parameter 'etemp' related to temperature. That is by setting etemp='300' does it mean the parameters are calculated at 300K? the same question holds: do we have to recalculate the elph marix elements?
5 If etemp does not represent temperature, how can we obtain let us say the elecself energies at various temperarures?
I am sorry to ask all of this but there are a lot of parameters to be checked and for the sake of saving computational time, I am trying to get my head straight around this.
Regards
clarification (convergence)
Moderator: hlee

 Posts: 82
 Joined: Thu May 05, 2016 5:18 pm
 Affiliation:
clarification (convergence)
Physics Department
university of Rondonia Brazil
Porto Velho Rondonia
university of Rondonia Brazil
Porto Velho Rondonia
Re: clarification (convergence)
Dear eliephys78,
It is indeed a good idea to understand that.
No, they do not. They only depend on the initial coarse grid.
You need to make sure that they decay sufficiently for increase real space vectors. You should expect at least 46 order of magnitude decrease. Best plot on a log y scale.
Yes but you only have to redo the interpolation part. So you can restart from reading the epmatwp1 file.
In principle yes. However if you are looking at the electron selfenergy, the you mainly care about degaussw and for the phonon selfenergy mainly about degaussq.
In addition, I would say that they just need to be small enough to not change the results.
Typically you want to use 15 meV for the degaussw (this is very well converged). If you do not care about tiny variation, then you can use 10 meV. Larger than that might affect the results (but will speed up the convergence).
degaussq you do not need to do convergence since most quantities are outputed for different values of degaussq so that you only need to do the calculation once.
This is a bit tricky. eptemp (not etemp) is the temperature that enters into the FermiDirac occupation factors.
In most cases it is indeed the temperature. So etemp='300' means 300K.
Now you can compute multiple temperature at a time (see the corresponding input variables). Note that it might not work for all quantities. Then you have to do 1 temperature per calculations.
In the specific case of mobility, 'eptemp' is not used and 'tempsmin', 'tempsmax' is used instead.
No problem,
Samuel
Regards
It is indeed a good idea to understand that.
1 Do the 'decay.H' and 'decay.dynmat' depend on the interpolated q and k grids or they only depend on the coarse grids? That is do we have to check them versus the cahnges in the fine q and k grids too?
No, they do not. They only depend on the initial coarse grid.
You need to make sure that they decay sufficiently for increase real space vectors. You should expect at least 46 order of magnitude decrease. Best plot on a log y scale.
2 Regarding the convergence of electronic relaxation time and the phonon self energy, do we have to re calculate the elph matrix elements each time we change either the phonon or the electronic fine grids?
Yes but you only have to redo the interpolation part. So you can restart from reading the epmatwp1 file.
3 Once convergence has been achieved with respect to phonon and electronic fine grids, and we have to check convergence w.r.t degaussw and degaussq, Is there any need to recalculate those matrix elements?
In principle yes. However if you are looking at the electron selfenergy, the you mainly care about degaussw and for the phonon selfenergy mainly about degaussq.
In addition, I would say that they just need to be small enough to not change the results.
Typically you want to use 15 meV for the degaussw (this is very well converged). If you do not care about tiny variation, then you can use 10 meV. Larger than that might affect the results (but will speed up the convergence).
degaussq you do not need to do convergence since most quantities are outputed for different values of degaussq so that you only need to do the calculation once.
4 Is the parameter 'etemp' related to temperature. That is by setting etemp='300' does it mean the parameters are calculated at 300K? the same question holds: do we have to recalculate the elph marix elements?
This is a bit tricky. eptemp (not etemp) is the temperature that enters into the FermiDirac occupation factors.
In most cases it is indeed the temperature. So etemp='300' means 300K.
Now you can compute multiple temperature at a time (see the corresponding input variables). Note that it might not work for all quantities. Then you have to do 1 temperature per calculations.
In the specific case of mobility, 'eptemp' is not used and 'tempsmin', 'tempsmax' is used instead.
5 If etemp does not represent temperature, how can we obtain let us say the elecself energies at various temperarures?
I am sorry to ask all of this but there are a lot of parameters to be checked and for the sake of saving computational time, I am trying to get my head straight around this.
No problem,
Samuel
Regards
Dr. Samuel Poncé
Institute of Materials
École Polytechnique Fédérale de Lausanne (EPFL)
Switzerland
Web: https://www.samuelponce.com
Institute of Materials
École Polytechnique Fédérale de Lausanne (EPFL)
Switzerland
Web: https://www.samuelponce.com

 Posts: 82
 Joined: Thu May 05, 2016 5:18 pm
 Affiliation:
Re: clarification (convergence)
Dear Samuel,
Thanks for the valuable information and for your patience. One final thing : How is the convergence of the elph coupling strength (lambda) and superconducting temp (Tc) w.r.t. degaussq checked? Do we have to plot lambda as a function of degauss for various fine phonon grids and observe for which fine phono grid and at which degaussq, lambda stops to vary?
Regards
Thanks for the valuable information and for your patience. One final thing : How is the convergence of the elph coupling strength (lambda) and superconducting temp (Tc) w.r.t. degaussq checked? Do we have to plot lambda as a function of degauss for various fine phonon grids and observe for which fine phono grid and at which degaussq, lambda stops to vary?
Regards
Physics Department
university of Rondonia Brazil
Porto Velho Rondonia
university of Rondonia Brazil
Porto Velho Rondonia