My Community

Dear Admin

I am trying to do DFT calculations for UO2 surface.
Some recent papers say that i should apply a Uramping method to find the right U value for the system.

It is written that:
"begin U=0 with a diagonal occupation matrix and ramp U by 0.1eV at a time, iteratively applying the occupation matrix from the previous calculation to the next value of U."

Does this means:

1- relax the geometry with U=0, ICHAG=2, IBRION=2 or 1 and NSW=...
2- use the CHGCAR of U=0 calculation as (ICHARG=11) in the U=0.1 calculation?

I will be very pleased if you can help me.
Thank you

I assume you're referring to the work of Meredig et al. in PRB 82, 195218 (2010). I'm not sure they used VASP for their work -- you might contact them directly to ask how they did what they did -- but it seems to me that what you want to do is:
1) converge the wavefunction with U=0. Save the CHGCAR and WAVECAR files.
2) perform a subsequent calculation with ISTART=1, ICHARG=1, and LDAU=0.1. Again, save the CHGCAR and WAVECAR files.
3) perform a subsequent calculation with ISTART=1, ICHARG=1, and LDAU=0.2. Again, save the CHGCAR and WAVECAR files.
... and so forth.
While that seems slow and tedious, I'm not sure there's a way to vary U on the fly or extract the occupation matrix without hacking the code itself.

You may want to wait to do any geometry relaxation until after you've converged a wavefunction at your desired value of U, as the optimal geometry in DFT+U may differ from the optimal geometry in DFT.

Also, if UO2 is antiferromagnetic, you may have all kinds of fun getting VASP to pick, and stay in, the right antiferromagnetic state, particularly during geometry optimization. In geometry optimization on NiO, which has the same issues as UO2, I've had VASP switch from antiferromagnetic to nonmagnetic during a geometry optimization, and therefore find the wrong optimal geometry. Keep an eye on the magnetization values on each uranium atom! If they go to zero, you'll have to stop, run a non-spin-polarized calculation, then read that in to a spin-polarized calculation with MAGMOM explicitly set in order to re-capture the desired AFM state. Good luck!

Thank you for the answer but there are somethings that i dont understand from your reply. For example, in 111 surface. i have 6 U atoms, lets say magnetic moment for U is 2.0. So if i have an AFM surface at the end of the OSZICAR file the total magnetic moment will be zero. If the surface is FM, it has to be 12. Isnt it? I dont understand what you meant in last sentence.

Many thanks.

Hi

What WolverBean means is that with the LDA+U, you may have some difficulties to stay in the correct AFM state. The convergence is often difficult, especially for low symmetry systems, and because of that, vasp sometimes kicks uranium atoms out of their magnetic state and into some other funny state.

So you need to keep an eye on the electronic occupancies of uranium atoms just to check if your system has not jumped into a funny metastable state.

Many thanks dear boris.
It is difficult to understand really. Can you please tell me that whether my understanding is right?

1- submit the job with ISTART=0, ICHARG=0 and U=0 (also i dont change J value), ISPIN=2 MAGMOM=...
2- take the CHGCAR and WAVECAR files from firs step for the new calculation. also copy CONTCAR to POSCAR. Then change ISTART=1, ICHARG=1 and U=0.1. (also keep J value unchanged)

3- change U value untill i see the integer occupancies in the OUTCAR file. For example for both spin components they should be 1.000 in OUTCAR.

Is it right dear boris?

sorry i mean in the firs step ICHARG=2

Hi

You probably don't need to keep the wavecar file. The chgcar is enough. Otherwise your steps are correct: icharg = 2 for the first run, then keep the chgcar, set icharg = 1, and slightly increase U to 4 eV.

However, do not stop increasing U as soon as you get integer occupancies, because it will happen rather quickly, eg when U = 2 eV. You should increase U until you reach the commonly accepted value for UO2, ie U = 4 eV.

Also, if I were you, I would use the Liechtenstein approach and not the Dudarev approach.

Good luck!

Dear VolverBean and Boris

Some of my calculations failed and i dont know why. Can I want to benefit from your experiences on +U caculations please?

What do you think about my INCAR file below?

SYSTEM = UO2
ISTART = 0
ICHARG = 2
ISMEAR = 0
SIGMA = 0.05
NSW = 250
ISIF = 2
IBRION = 1
ALGO = fast
LREAL = auto
ENCUT = 400
NBANDS = 300
EDIFFG = -0.01
ISPIN = 2
MAGMOM =18*2.0 36*0.0
LDAU = .TRUE.
LDAUTYPE = 2
LDAUL = 3 -1
LDAUU = 0.00 0
LDAUJ = 0.50 0
LMAXMIX = 6
LASPH = .TRUE.

I want to optimize a FM UO2 surface with this INCAR. This is my first INCAR. After this first run i changed ISTART=1 ICHARG=1 and LDAUU=0.1 and so...

Some calculations with more number of atoms have failed. May be it is related with my incar parameters. I can not be sure about convergence and smearing parameters as well as ISIF and IBRION tags. Can you tell me where my mistake is please?

Thanks

ccccc,

Your INCAR file looks fine to me. If your calculation is not converging, you can try switching to ALGO=N (normal). I have found that sometimes, ALGO=F (fast) does not converge spin polarized calculations, but I've never had a problem converging when ALGO=N.

To clarify the remarks at the end of my last post:
In bulk UO2 with antiferromagnetic ordering, the total magnetic moment of the unit cell is zero. However, the magnetic moments on the individual U atoms should not be zero: they should be +2 and -2 (or near to that). There also exist solutions to the Kohn Sham equations in which in addition to the overall magnetic moment being zero, the magnetic moments on each U atom are also zero. These nonmagnetic solutions are higher in energy than the antiferromagnetic solution (probably by several eV). However, they ARE solutions, and for certain initial guesses these are the solutions VASP will find. If you set ISPIN=2 but do NOT use the MAGMOM tag, for example, VASP will probably NOT find the antiferromagnetic solution; it will probably find the higher energy non-magnetic solution instead. What's worse, sometimes when doing geometry optimization, I have found that even if my starting spin configuration had one U at +2 and the other U at -2, at some point during the geometry optimization, the magnetic moments on both U atoms went to zero. Once they go to zero, they tend to stay that way, because both magnetic moments being zero is a more symmetric configuration - even though it is also higher in energy. Unfortunately, if all you see is the overall magnetic moment after each geometry step, you cannot tell that this has happened, because the overall magnetic moment on the unit cell was always zero. So, you must keep an eye on the occupancy matrix in the DFT+U calculation, or else use the LORBIT tag to make sure the individual magnetic moments on each atom are printed to the OUTCAR file.
Ferromagnetic solutions, luckily, do not have this problem. If the total spin state starts at +12 for example (+2 on each of 6 atoms), it will stay that way.
I agree, all this is quite confusing. Write back if you're still having trouble.

Thank you for the explanation dear WolverBean.
It is very helpful for me. Actually i solved the problem with bulk material. Now i am trying to optimize UO2 surfaces like 110 111 100. I use this above INCAR for surface calculations. I generated 3 layers of 110 surface for (1x1) (2x2) and (3x3) periodicity. (1x1) and (2x2) calculations are going on well but (3x3) calculation fails every time. Also i generated 4 and 5 layers and had the same error. What do you think about this?

Also in all papers it is written that for 110 surface all bands would get fully occupied when U=1. But in all of my calculations with different number of atoms and layers, i had fully occupied energy bands for very different U values. Sometimes less than 1, sometimes bigger than 2 or between 1 and 2. Do you think it is a problem?

Many thanks..