How to choose values of LDAUU and LDAUJ in DFT+U calculations in VASP?

[rmz4ed]

Hi,

I am running LDA+U calculations for the first time in VASP and am wondering how the LDAUU and LDAUJ parameters are chosen. Based on the VASP wiki example of NiO (wiki/index.php/NiO_GGA%2BU), one value is inputted for each atom type (LDAUU = 5 and 0 and LDAUJ = 0 and 0 for Ni and O in the example), but I am unsure how to choose these values.

Thanks!

[marie-therese.huebsch]

Hi rmz4ed,

There are different strategies to choose the on-site Coulomb repulsion LDAUU and on-site exchange interaction LDAUJ.

(i) If experimental data is available, you can choose LDAUU and LDAUJ to reproduce, e.g., the optical gap or the binary formation enthalpies. This is not truly following the first-principles philosophy, but it can help find a computational model to describe other aspects of your system.

(ii) Another option is to obtain the values from GW calculations. This takes more computational effort compared to DFT+U calculation, but it follows the first-principles philosophy.

(iii) For some materials, you might find good suggestions in literature based on (i) or (ii).

(iv) You can also treat these as tunable parameters and test multiple values. In this case, you may try to make an educated guess to scan appropriate values based on similar compounds for which suitable values are known and that feature the same element as the system you are interested in.

Maybe there are more strategies, but these are the ones I can think of.

Best regards,
Marie-Therese

[alireza_rezvani]

Hi,

Apologize if my question sounds simple, I am very new to DFT calculation. Are there any steps how to obtain LDAUU from GW calculations?

Thanks,
Ali

[martin.schlipf]

I think the idea is to use a high-level method (could be GW, hybrid functionals, ...) to get an accurate band structure and then tune the U and J to reproduce that band structure. This strategy works best if you have a comparatively simple reference system where you can afford this higher level method and then want to simulate are more complex system (surfaces, molecular dynamics, defects, ...) and reuse the same U and J parameters.

However, note that in general U is system depended. So your reference system needs to be as close as possible to the real one. Furthermore, be cautious when transferring U from a paper to your work. Since the U is only applied within a sphere around the atom, the same U value may produce different results based on the size of the sphere.

Finally, there is one more method documented in the wiki: You can determine U using linear response. This will work well if DFT (without U) yields at least a somewhat reasonable description of the system. It can also be costly because it requires supercell calculations.

[martin.schlipf]

I talked a bit about this and there is another method based on GW to get access to the U's. You can use cRPA. This is often used for metals where some of the other techniques do not work.