LDA+U with U correction on p-orbitals

Total energy, geometry optimization, DFT+U, spin....

Moderator: bguster

Locked
nirama
Posts: 1
Joined: Fri Jul 29, 2016 9:35 pm

LDA+U with U correction on p-orbitals

Post by nirama » Fri Jul 29, 2016 9:52 pm

Hi ABINITers,

Looking at the tutorial on DFT+U it is stated that: "You cannot treat s and p orbitals with LDA+U in the present version of ABINIT."
However, I tried applying lpawu = 1 on the Oxygen atoms in and Li2O2 structure. To my surprise it actually runs and spits out something different than just running plain LDA. (The output file also indicates that it is actually doing some +U correction).

So, if ABINIT it doesn't treat s and p orbitals with a +U correction what is it doing when I set lpawu = 1?

I'm running v. 7.10.2.

best,
Nicolai

amadon
Posts: 46
Joined: Mon Aug 24, 2009 10:58 am

Re: LDA+U with U correction on p-orbitals

Post by amadon » Mon Aug 22, 2016 3:13 pm

Dear Nicolai,

Indeed, DFT+U works also for p orbitals (l=1) in ABINIT. However, as the norm of p orbitals
inside the PAW sphere can be small, I recommend to use
dmatpuopt=4 which partially compensates for this and gives more physical
occupations (Note however that the default value for dmatputopt is 2).
Nevertheless dmaputopt=4 is more tricky to use
than dmapuopt=2: Indeed, it can be used only with PAW atomic data
generated with the "vanderbilt" scheme. Fortunately, it is the case of the atomic data available
in the JTH table (available on the ABINIT web site). You can still use dmatpuopt=2, however, larger values of U will
be necessary.

Best regards
Bernard Amadon
CEA
France

PS: Please precise your affiliation, Thanks.
Bernard Amadon
CEA
France

sheng
Posts: 64
Joined: Fri Apr 11, 2014 3:44 pm

Re: LDA+U with U correction on p-orbitals

Post by sheng » Tue Sep 13, 2016 12:09 pm

Dear Amadon

Thank you for your clarification regarding the application of DFT+U on the p orbitals.
According to your explanation, the PAW atomic data must be generated with the "vanderbilt" scheme if dmaputopt=4. Is it a strict condition?
Many of the JTH v1.0 PAW datasets are generated using the parameter "custom rrkj".
Are they applicable in this case?

Thank you.

Regards
Sheng

Locked