Convergence problem in 11.8b1

[BlackBarrel]

Recently I tried mGGA in bulk calculation. It converges well and yields a good bandgap result, so I think mGGA should give better results in device calculation. However, when I tried 11.8b1 with my old device structures (which means they converge well in 11.2.3 using PBE), severe convergence problem occured in all part of Two Probe calculation using mGGA. Every part of Two Probe calculation is not able to converge, from the electrode to central region SCF. I've tried methods like anneling, pre-conditioner, tuning the iteration parameters and using the previous PBE .nc file as the initial state, but all didn't work. Even the electrode calculation is not able to converge.

Is there anyone else having such convergence problem?

My system is all made up of Si atoms, having 8 atoms in electrode and 50-100 atoms in central region (I've changed it from 144 to 72 and still got such problem) and two metallic regions.

PS: Have 11.8.b1 expired now?

[camelluxin]

expired on my computer

[Nordland]

Yes. 11.8.b1 expired on first of November.

There is two ways to use mGGA. The TB09LDA exhcange correlation potential depends on a parameter C. This parameter can either be determined automatically
with
xc = MGGA.TB09LDA
or it can be set to a fixed value like
xc = MGGA.TB09LDA(c=1.23)

Using the automatic method is well-suited for simple bulk systems, but for more advanced systems it is best to determine this C using a simple precalculation.
So what I would do, is to setup a fast calculation of Si(bulk) and do a calculation of the band gap, and then tune C until it gives me the correct band gap.

Lucky for you, I have much interest in Silicon myself, and I have found that, that C=1.035 using DZP basis set gives a band gap of 1.1147, and that is good enough for me.

[Nordland]

On a second note. It is important to know a little about mGGA.

This is what I have learned.

• Never use C < 1.0, it simply goes haywire
• A larger C always gives a larger band gap
• The effect of C is very strong, so small steps.
• All semi-conductors (I have studied) have a good C in the range from 1 to 1.2/1.3
• If you need a insulator band gap like 10 eV, then C around 1.7 is enough

[Anders Blom]

Very useful advice. Although perhaps directly relevant for this calculation, it is also important to remember that meta-GGA does not work for metals. So, if you have a system with metal electrodes, you will (probably) not be able to use meta-GGA even if your central region is Si or SiO2 or similar. In this case, the suggested approach is to use LDA/GGA+U instead, in which case you may use the meta-GGA calculation as a reference, both to fit U and to make sure the band structure remains correct.

[BlackBarrel]

Thank for all answers above! I'm not familiar with mGGA and probably have used it in a wrong way! My electrode is indeed metallic. However, all electrodes should be metallic according to my knowledge. Does it mean that mGGA is not suitable in Two Probe calculation？May I use other xc, say LDA, in electrode calculation and use mGGA in the central region SCF?


Special thanks to Nordland, your information is very useful!

[Anders Blom]

See, that would be the ideal way - to use different functionals in different regions. But, it's not really possible in the way DFT works currently; maybe new algorithms can be developed later on that make it possible.