Alchemical pseudopotential

[Asier_Z]

Dear all,
I am using alchemical pseudopotentials to calculate the band structure of Ir(1-x)Pt(x)BiSe. The problem is that the pseudopotentials are not mixed properly: I have tried with different values of x and I always get the same result. I suspect that Abinit is calculating PtBiSe, but I don´t know what I am doing wrong.
I attach the input file of Ir(0.75)Pt(0.25)BiSe so that you can help me.

Ir(0.75)Pt(0.25)BiSe.in

(2.48 KiB) Downloaded 293 times

[ebousquet]

Dear Asier_Z
Did you test your system without spin-orbit coupling first? I've never used SOC with VCA so far and don't if anyone did it before.
Which pseudopotential files are you using?
All the best,
Eric

[Asier_Z]

Dear Eric,
I am using hgh pseudopotentials from the Abinit website, until I tried alchemical mixing the have given me good results. I did´t know there could be a problem when using SOC and alchemical pseudopotentials, I will calculate it again without SOC to see what happens. I will let you know if it works.
Regards,
Asier

[ebousquet]

I don't know if SOC will be problem or not with VCA, this is why it is better to check.
Regarding the pseudopotential, it might be more interesting to use the one of pseudodojo (reduced cutoff w.r.t. HGH). They are not made with SOC but I can do that for you if you are interested (I actually already have several of them with SOC like Pt and Bi).
Cheers,
Eric

[Asier_Z]

Dear Eric,
The calculation without SOC didn´t solve the problem, so I am still stuck with that. Could it be that the HGH pseudopotentials I am using can´t be used with VCA? Related to this, do you know any kind of pseudopotentials that work well with VCA? For now, it would be enough for me to calculate it without SOC, once I solve this I will try to add it if it is possible

[ebousquet]

Dear Azier,
I would advise you to use the pseudopotentials from Pseudodjo project:
http://www.pseudo-dojo.org/
You can see the related paper on ArXiv (the paper will come out in the next days):
https://arxiv.org/abs/1710.10138
I have used them with VCA and it worked quite well:
https://pubs.acs.org/doi/10.1021/acs.chemmater.5b03564
Can you first test these pseudos in your system without SOC and if this looks OK we can go for SOC.
Lets see how it goes.
All the best,
Eric

[gonze]

Dear Azier and Eric,
The SOC should not be a problem per se. I think Eric is right to suggest to switch to the pseudo-dojo and make a try.
Still, the problem might simply be that the VCA does not work in your material for this pair of atoms (Ir and Pt),
despite them being close in the periodic table (Z=77 and 78). The VCA is indeed not a very safe method (=a cooking recipe).
It is expected to work reasonably well in the isovalent case (but even this is not guaranteed), but when the
ion charge is not the same, VCA is more likely to fail.
Best,
Xavier

[Asier_Z]

Dear Eric and Xavier,
First of all, thank you very much for your help. I used pseudo-dojo pseudopotentials but the following warning appears every time I try to run the program:

"-- !WARNING
src_file: mkdenpos.F90
src_line: 176
message: |
Density went too small (lower than xc_denpos) at 106536 points
and was set to xc_denpos = 1.00E-14. Lowest was -0.85+251.
Likely due to too low boxcut or too low ecut for pseudopotential core charge."

I have increased the ecut but it doesn´t solve the problem. These kind of pseudopotentials work well on my computer (I tried them for IrBiSe), but in order to speed up the calculation I am sending the programs to a supercomputer, and there is where the problem arises.
Apart from that, if it is the case that VCA can´t be applied to Ir and Pt, what other method could I use to get the results I expect for Ir(1-x)Pt(x)BiSe?
Regards,
Asier

[ebousquet]

Dear Asier,
Regarding your Warning, does it prevent your calculation to continue? Sounds like your ecut could too small, how much is your ecut? However, according to what you say (it works locally but not on the supercomputer) it is possible that it comes from compilation flags you use there.
Regarding VCA not working, I'm afraid that the only solution is to build supercells (unless somebody else has another idea)...
Cheers,
Eric