I have seen during DFT calculation some authors comment that LDA/GGA functional overestimates/underestimates the band gap/lattice constant of some materials. Before going to the main calculation, it is recommended that we can plot phonon dispersion plot (say) to verify results with literature data so that future calculations goes well.
For example: my intention is to calculate optical properties (say dielectric function) of a material (say GaAs). And I am using a GGA functional which underestimate the band gap from experimental value. At that position, what should I do? Should I use that pseudopotential to run the necessary executables (scf, nscf, epsilon etc)? Or I have to improve the band gap first and make it close to experimental value and after that I should run the executables (scf, nscf, epsilon etc) to find out optical properties?
I am sorry, but there is some confusion in your question and how you are facing the problem. For example, I do not really understand why you need phonon band structure. That would serve to understand if your system is thermodynamically stable.
On the other hand, people often report that the exchange-correlation potentials underestimate the band gap of a material. This fact has a clear physical origin, that is, the xc potentials we have depends on the density, thus tends to delocalise electrons, which in turn reduces the band gap. The only way you have to address this problem is to change the xc-potential. Attempts have been made, for example, via either GW corrections or hybrid xc-correlation functional. Both of these methods are way more computationally expensive than standard LDA/GGA calculations.
Coming back to your physical problem, yes the steps to calculate the dielectric functions are those you said. In my experience, however, there is no way in epsilon.x to adjust the band gap to reduce the impact of the xc-correlation functional.
If you want to obtain a better dielectric and optical response, you need to look at other methods, for example, the solution of the Bethe-Saltpeter equation via, e.g., Yambo that is interfaced with Quantum Espresso.
I hope this helps,
Roberto
@professor Roberto D'Agosta
Thank you so much for your detailed comments. For example, I want to calculate dielectric function using DFT. For that, I choose GGA/LDA (whichever better) and run scf calculation. Lets say, I perform many convergence analysis (Kpoints, ecut etc) on scf and reach ground state. After that, I run nscf with more Kpoints and finally epsilon.x pa kage. At this stage, if I compare my dielectric function with experimental value and find some mismatch (say dielectric function curve is same, but some shifting problem) what reason I can put on surface? Should I say that we couldn't use correct exchange correlation and that's why this error? Or some other reasons might work here?
See the point is that this is not an error as indeed is coming from the quality of the approximation you are making. You should compare trends and maybe relative positions of the peaks (and their physical origin) rather than their absolute position in energy.
That said, it is normal that the spectra are shifted to accomodate for a larger band gap. At the same time, this shift also is somewhat arbitrary as the electron gap might be loosely related with the optical gap.
If you want to co-relate your simulation work with practical work, you have to use different functional like(LDA/PBE/HSE). For different functional there are different types of pseudopotentials. So, you have to try them for each functional. After that, if you get the proper bandgap you have to use that functional and pseudopotential to measure optical properties. You can also use tensile and compressive strain under different functional. Beacuse in practical work we have to go through pressure.
After all of that optimization is necessary to get accurate results for simulation work. To do this kind of work you have to first optimized the lattice constant, k-point and cut off energy. After optimizing the scf file by varying these three parameters you have to run all the other things. You have to find the lowest total energy where the structure is firm. Thanks.
Roberto D'Agosta
Hi Professor, Would you know how is it possible to find out dielectric function which depends both on q points in the Brillouin Zone and frequency ? Actually, if we use Quantum Espresso, we only get based on frequency (not q points). Would we do that using Yambo? If you notice the equation in the screenshot I attached, it needs input of q and frequency dependent dielectric function.
The link of the paper related to the equation (9) (https://journals.aps.org/prapplied/abstract/10.1103/PhysRevApplied.14.024080)
Thank you in advance for your time
Dear Md. Jahid Hasan Sagor ,
To the best of my knowledge, Quantum Espresso is not able to calculate the momentum-dependent dielectric function, as the code that does this calculation performs the integration over the first BZ automatically. Yambo can do that, and I will refer you to the manual for that code for further details. You should be able to run the calculation at the level of DFT without running into the quasi-particle corrections proper of Yambo if not needed.
Do take into account that Yambo works only with norm-conserving pseudopotentials since calculation with ultrasoft is both discouraged and unstable.
I hope this helps,
Roberto
- Badges
- Science topic
- Similar topics
- Filing