DFT+U

External optional components of Abinit (BigDFT, Wannier90...) and codes using/providing data from/to Abinit (AtomPAW, ONCVPSP, EXC, DP, Yambo...)

Moderators: ebousquet, bxu

Locked
lakhal
Posts: 3
Joined: Mon May 01, 2017 9:26 pm

DFT+U

Post by lakhal » Mon Oct 02, 2017 12:42 pm

Dear abinit users
I have a question about DFT + U, I read in the abinit tutorial just the example of LDA + U, my question is: can I use GGA + U or not?
Thanks to all

fhssn1
Posts: 36
Joined: Mon Feb 26, 2018 7:52 pm

Re: DFT+U

Post by fhssn1 » Mon Mar 05, 2018 8:17 pm

I would be interested in an answer to the same question. They say that LDA+U is really DFT+U, but why not GGA+U?

I have seen a paper that mentions DFT+U and GGA as part of the computational setup, so it's likely that this is doable in general. But not sure whether it can be done in ABINIT. I haven't tried it myself.

Secondly, the whole point of such extensions is improved accuracy of the XC functional approximation. Then why do we have 'two tracks' of accuracy enhancement: 1) better functionals (e.g., HSE06) 2) addition of '+U'? In other words, what can '+U' do that HSE06 can't, and what can HSE06 do that '+U' can't? What are the pros and cons of each enhancement.

And is it possible to have HSE06+U?

ebousquet
Posts: 469
Joined: Tue Apr 19, 2011 11:13 am
Location: University of Liege, Belgium

Re: DFT+U

Post by ebousquet » Tue Mar 06, 2018 10:07 am

Dear fhssn1 (and lakhal),
GGA+U works too, no problem.
+U correction is made on a single orbital of a given atom, then only this orbital is corrected. Different values of U can be used, this is up to the user to test the effect of U on the physical properties.
In short, hybrid functionals works by using a given amount of exact exchange Hartree-Fock (HF) and compensate double counting. The amount of exact exchange is fixed for a given hybrid flavor (HSE06, B1WC, etc), thought the user can change it. This correction is applied to all orbitals of all the atoms.
Note that HF calculation in plane wave is time consuming w.r.t. DFT+U, such as hybrids can be from 100 to 200 times slower than DFT+U.
Which one to chose depends on what you want to calculate, if DFT+U is enough then use DFT+U.
I advise you to read the seminal papers of the different hybrid functional flavors for more details.
All the best,
Eric

Locked