Energies of core levels

[eleonora_ascrizzi]

Hi everyone,

I have a question about how can I get the energies of core levels with VASP. I have to say that I am familiar with codes that use localised wave fuctions and here I can check the eigenvalues of every orbital. To do the same in VASP I am using ICORELEVEL=1 or 2 and I am reading eigenenergies in the OUTCAR. Is this the corrrect methodology? I am not sure as I don't feel confident with the number I see.

Thanks a lot,

Eleonora

[martin.schlipf]

The wiki description of the tag ICORELEVEL explains the method in VASP quite well. You need to be aware of the limitations of VASP and need to think what physical quantity you want to compute. On the one hand, VASP does not relax the core density with ICORELEVEL=1 so that your results will be always a little bit different compared to a method where all states are optimized. On the other hand, VASP offers ICORELEVEL=2, which allows you to compare to results of core spectroscopy by explicitly exciting a core electron into the valence band. The absolute energy of the core states is usually meaningless because the self-interaction in local DFT yields to too high energies.

[eleonora_ascrizzi]

Thank you Martin for your kind answer.

May I also ask you if I have to run a single point? Or I just start from an existing wave function? I did not understand from the wiki description.
I would like to compare myself to XPS experimental results. Would you think that this is the correct methodology to use ICORELEVEL=2? Maybe knowing it is higher than experiment because of the self-interaction?

Thank you

Eleonora

[eleonora_ascrizzi]

I mean, could you tell me how to compare my data with experimental XPS? You said that the absolute energy is usually meaningless so which quantity should I use?

Thank you

Eleonora

[martin.schlipf]

I'm not an expert on XPS but I had a quick look at this description. Where DFT can help you is for things like the right part of this figure

So the absolute binding energy of the C 1s peak would be incorrect but you should still get the splitting of the peak according to the chemical environment. I believe for this case even ICORELEVEL=1 should be enough. Essentially, if you run an experiment and get an XPS spectrum where you are not sure what bonding occurs in the crystal, you could compute all candidate materials and see in which case the splitting matches best with the observations.

Regarding how to run the calculation: For ICORELEVEL=1, you can just add this to your SCF calculation or if you want to save CPU time converge the calculation and then run a single calculation to obtain the core eigenvalues. As mentioned earlier, there is not self-consistency of the core density so you do not need to iterate. ICORELEVEL=2 is a bit more complicated; there is a tutorial how to use it in the wiki.

[eleonora_ascrizzi]

Thank you Martin for your answers.

I tried to compute the energy level of 1s of an oxygen with the the initial and final state approximation getting the outup I show below. I am a little concerned about the big difference between the two methodologies: does it mean I am doing something wrong?

Thank you,
Eleonora

Input:
ICORELEVEL=2
CLNT=1
CLN=1
CLL=0
CLZ=1

Output:
the core state eigenenergies are
1- 1s -571.9235

Input:
LVHAR=.T.
ICORELEVEL=1
CLNT=1
CLN=1
CLL=0
CLZ=1

Output:
the core state eigenenergies are
1- 1s -505.7398

[martin.schlipf]

Did you create a supercell for the ICORELEVEL = 2 calculation?

[eleonora_ascrizzi]

No, I did not. Should have I?

I have 6 oxygens in my cell and I 'isolated' the first one giving it a 'different' type in the POSCAR and I repeated two times his pseudo potential in the POTCAR.
I hope I have made it clear, I report below the POSCAR

Thank you very much

Au+FeO
1.000000000000000
5.8799999999999999 0.0000000000000000 0.0000000000000000
0.0000000000000000 10.1799999999999997 0.0000000000000000
0.0000000000000000 0.0000000000000000 22.0000000000000000
O O Fe Au
1 5 6 24
Selective dynamics
Direct
here I put the coordinates

[martin.schlipf]

Well your cell is already quite long for the y and z axis, so perhaps it is already big enough. You should definitely try to see if the result change when you double the cell along the x axis. In general, the cell size is a convergence parameter for ICORELEVEL = 2. You need to increase it until the results do not change within the precision you desire.

Note that this is the important difference between the two different methods. Since ICORELEVEL = 2 actually excites electrons from the core into the valence band, there will be an interaction of this modified atom and its periodic replica. In reality, the excited atoms will be very far apart, so they do not interact. Hence, the need to increase the cell size until this interaction if vanishingly small.

[eleonora_ascrizzi]

Thank you martin you are very kind. I will try to double it in the x direction.

Have a nice day