About the convergence of transmissionspectrum calculation

[lknife]

Dear experts,

I want to calculate the transmission spectrum (or IV curve) of a 2D-device using DFT-GGA method. First of all, I made a k-point scan for kB (periodic along B direction) from 9 to 19 with kA=1 and kC=150. I got the kB dependency of the transmission current, seen in the attached file.
I want to know, what kind of conclusion can I get from the figure, kB=9 is almost sufficient, or kB=19 is still not sufficient?

I appreciate you for your kind help!

[Petr Khomyakov]

The conclusion is that the current is  somewhat sensitive (within 40-80 nA around 740 nA) to the k-point sampling. If you want the current to be computed with higher accuracy, you should continue increasing the k-point density. 

[lknife]

Dear Petr Khomyakov,

Thank you for your reply. I increased the kB up to 25 and got the attached figure. Do you think that kB=25 is almost sufficient for the transmission spectrum calculation?

Thanks again for your kind help!

lknife

[Petr Khomyakov]

Now  the current seems to be converged much better. It looks fine to me, but at the end it is only you who decides on (or set) the tolerance for the current calculation accuracy. 

[lknife]

Thank you very much for your valuable comments! It seems that a k-point scan or a mesh-cutoff scan should be needed even though they might take a long time for doing so. It's boring but necessary.

[lknife]

Dear Petr Khomyakov,

For the same device configuration, I redid the k-point scan using ATK SE: Extended Huckel method. It's much faster than the DFT method, about 1/20 of the time consuming. However, the converged current calculated using Huckel method is different with that calculated using DFT method, seen in the attached figure: the red plot was calculated using DFT method, and the blue one was calculated using Huckel method.

I have a question: Even though that the result calculated using DFT method might be more accurate, can I use the Huckel method for the simulation of transmission spectrum? In other words, to what extent can I trust the result obtained using Huckel method?

Thank you very much!

[Petr Khomyakov]

There is nothing stopping you from trying to use the Huckel method, but it should be benchmarked (at least for some points on the IV curve) against more accurate methodology such as based on density functional theory, to insure that the Huckel calculations are accurate enough for your system of study.  It is impossible to answer your question without doing any benchmarks.