LDA+U

Message

zazi · #1 Post by **zazi** » Fri Feb 03, 2012 4:34 pm

Hi every one,
I want to do LDA+U calculations
I have read this in vasp manual that: LDAUL, LDAUU, and LDAUJ must be specified for all atomic species!.
Can some body tell me how I can write this in Incar with an exemple please.
Thanks for your amswer.

yasin · #2 Post by **yasin** » Sat Feb 04, 2012 6:03 pm

See vasp tutorial. hands-on example part 3

http://cms.mpi.univie.ac.at/vasp-workshop/

zazi · #3 Post by **zazi** » Sun Feb 05, 2012 5:57 pm

Thanks a lot for your answer, but I dont found in it what I'm looking for. do you have any precise answer please. Thanks!

superyoyo · #4 Post by **superyoyo** » Fri Feb 10, 2012 5:09 am

It will be something like this:

LDAU = .TRUE.
LDAUTYPE = 1
LDAUL = 1 2 1
LDAUU = 0.00 4.896 2.850
LDAUJ = 0.0 0.0 0.0
LDAUPRINT = 0
LMAXMIX = 4

there are 3 atom species. The U is applied to p orbital of specie 1 and 3 and d orbital of specie 2.

zaz · #5 Post by **zaz** » Tue Feb 14, 2012 2:35 pm

Thanks a lot superyoyo for your answer. If I take your example and make the structure two times. I will have six atoms (two of each atom). in this case can I write this:
LDAUU = 2*0.00 2*4.896 2*2.850

Or have I two write them for each atom like 0.00 0.00 4.896 4.896 2.850 2.850. Does this make difference for vasp?
Thanks a lot for your answer

egillsk · #6 Post by **egillsk** » Wed Mar 07, 2012 4:33 pm

zaz, you do this for the kind of species, not for each atoms in your supercell, even though you double your cell, the lDAU tags are still the same.

zaz · #7 Post by **zaz** » Wed Mar 28, 2012 1:05 pm

Hi egillsk!
I'm sorry but your answer is wrong! I did both and the right answer is to give the U value for each atom, not only for each species.
Thanks any way!

kambiz · #8 Post by **kambiz** » Thu Mar 29, 2012 3:06 am

Hey zaz,

egillsk is right. you just need the U and J values for each type. For me it works very fine

Cheers

zaz · #9 Post by **zaz** » Fri Oct 05, 2012 6:17 pm

HI!

Are you sure?
this is funny!

zaz · #10 Post by **zaz** » Wed Nov 07, 2012 11:44 am

I would like to have the answer of admin about this LDA+U, I'm in trouble, because a work of more than a very long time can be incorrect. Please Admin, can I have your answer about what is the write way to write the U values for LDA+U method in can you have more than atom by species in your cell.
Thanks a lot in advance
zaz

icall · #11 Post by **icall** » Thu Nov 15, 2012 2:20 am

It is not defined the same way as you define the magnetic moments. You basically give U for each atom kind. If you have 2 Fe and 2 O atoms in your cell, you basically type LDAUL = 2 -1 and LDAUU = 4.0 0.0 unless you define any of your Fe atoms as a different species.

askhetan · #12 Post by **askhetan** » Mon Dec 03, 2012 3:25 pm

zaz - its is only specified for each species in the same order as they appear in your POSCAR file. Thats it. just think about how impractical it will become for people who simulate systems comprising of 100 atoms.

zaz · #13 Post by **zaz** » Wed Jun 05, 2013 4:19 pm

Hello every one and thanks for your answers,

Thanks a lot for your answers. I give you here and exemple of my Incar file for FeMgSiO3 compound,

for the U parameters part I have use:
LDAU = .TRUE.
LDAUTYPE = 2
LDAUU = 4.5 0.0 0.0 0.0

I have two questions:
1 question: Is this enough? is this right?
2 question: Is it necessary to relax the structure in some way (for exemple with GGA) before applaying this LDA+U method. by applaying this directely I have found that Al atom is more stable in Mg site than Fe!!!! this is very streng. I'm sure that some thing was wrong in my work

. B) %-6 %-6 %-6

WolverBean · #14 Post by **WolverBean** » Fri Jun 07, 2013 8:16 pm

egillsk and kambiz are correct: the number of entries for LDAUL and LDAUU must match the number of atom types, not the total number of atoms.
Actually, to be more precise, the number of entries must match the number of atom types defined in your POTCAR. So for example, if you're doing a calculation on Fe3O4 and you want to assign a different value of U to the tetrahedral Fe than to the octahedral Fe, you can create a POTCAR that includes Fe twice and O once. Your LDAUU and LDAUJ lines would then have three entries: once for each type of Fe, and once for O.

As to whether you need to do geometry relaxation with LDA+U turned on: in my experience, geometries with +U turned on are different than those without +U. How different they are depends on the system. For example, in MoO3, which has no occupied d electrons, changing the value of U changes the lattice constants by ~ 0.2%, which is small enough to ignore. For MoO2, the length of the Mo-Mo bond is quite sensitive to the value of U, so using the DFT geometry for a DFT+U calculation will give substantial errors. The importance of optimizing geometry with +U turned on is system dependent, and you will unfortunately have to check for yourself whether it matters in your system or not.

Another important thing to watch out for when optimizing geometries: if your system has any magnetic properties, make sure you do your geometry optimization in the correct magnetic state. As a simple example, the optimal lattice constants for NiO in the ferromagnetic and antiferromagnetic states differ by ~ 10%: a much larger error than you'd see by changing the value of U on Ni within a single magnetic configuration.

<span class='smallblacktext'>[ Edited Fri Jun 07 2013, 08:27PM ]</span>