Thanks for your response.

I also checked Abinit but couldn't find any useful documentation of the use of hybrid functionals there. There is a youtube video that's not really self-explanatory. Comments in the Abinit mailing list from 2016 suggest that you shouldn't use hybrid functionals at that time. It seems to be still a very experimental features.

Furthermore, I found nothing regarding the question if SOC is supported in calculations using hybrid functionals. At least, in the example in the youtube video, the variable nsppol is set to 2. For SOC, it would have to be set to 1.

So, let me formulate my question more precisely:

Has anyone here ever successfully made a band structure calculation using both, hybrid functionals _and_ SOC, in one of the free DFT codes?

For quantum ESPRESSO I think the answer is in www.quantum-espresso.org/faq/pseudopotentials/ point 2.7 (or ctrl+f -> HSE).

I know you are looking for HSE or similar, but did you test your system with PP for QE which you can find on www.quantum-espresso.org/pseudopotentials/# (Fully Relativistic ones). There is only few PP of this kind for only a few atoms. Additionally If you choose Ultrasoft (US) PP you should check the ecutwfc and ecutrho for your system, because sometimes values proposed inside of PP text-file are not suitable for some systems (for example: rational and proper ecutwfc might be two times bigger and ecutrho = 8 or 12*ecutwfc (default is ecutrho=4*ecutwfc and its too small)). Furthermore you should check keywords for this US/PAW-FR-PP:

----> noncolin=.true.

----> lspinorb=.true.

www.quantum-espresso.org/wp-content/uploads/Doc/INPUT_PW.html#idm528

Kind regards

Thanks.

Using noncolin=.true. and lspinorb=.true. indeed works perfectly for LDA and GGA funtionals, together with the right pseudopotentials. In particular the PAW pseudopotentials generated by Andrea Dal Corso work perfectly.

I tried using a meta-GGA functional from libxc, e.g. the TB09 functional. It seems that these functionals are implemented only for norm-conserving pseudopotentials, which rules out the PAW pseudopotentials.

I tried the SG15 ONCV pseudopotentials, because there are also relativistic versions available.

The problem is that band structure is not smooth but rather messy. I could be wrong but I have the feeling that the exchange functionals are only well implemented for k=0. But it might also be the pseudopotential. I'm not sure.

I cannot really find other pseudopotentials that are both, norm-conserving and fully relativistic and most of the pseudopotential generation tools I've looked at don't support fully relativistic calculations.

Right now, I try to use ELK. Because it is FP-LAWP, pseudopotentials are not a probem.

Regards,

Moritz

Thanks for your response.

I have already found these links and I've checked that hybrid functionals and PAW indeed work together in abinit.

But the problem is that hybrids don't seem to work with spin-orbit coupling.

Using abinit version 8.6.3, I get the error message

"Hartree-Fock option can be used only with option nspinor=1."

The presentation that you've linked to is from last year, so it seems that Fock-exchange together with PAW is rather newly implemented. It is explicitly written that

"the spin polarized case has not been extensively tested".

Nothing is written about non-collinear magnetism required for spin-orbit coupling.

The site:

https://www.abinit.org/sites/default/files/last/topics/generated_files/topic_Hybrids.html

indeed lists a few examples of how to use hybrid functionals:

tests/libxc/Input: t67.in, t68.in, t69.in

tests/paral/Input: t93.in, t94.in

tests/v7/Input: t65.in, t66.in

but none of them includes spin-orbit coupling (although some of them include spin-polarized calculations).

From this I have concluded that using hybrid functionals together with spin-orbit coupling is not yet implemented in Abinit.

If, as you wrote, there is an example for hybrids+spin-orbit, I would be very happy if you could give me the link.

Moritz

I also thought about different ways of combining different results from different calculations. I am still a bit hesitant, since such procedures seem to be a bit arbitrary, because it is not clear to me whether or not different ways of combining such calculations will actually lead to comparable results.

Thanks for mentioning JDFTx. I did not know this code and I will have a look at it.