Questions about using CE in multicomponent ionic materials

icet
1

Dear developers,

ICET is a super helpful and powerful package performing CE. I can’t say how grateful I am for what you’ve done. However, I’m recently bothered by a couple questions about using CE in ionic compounds in which the elements may exibit multiple oxidation state.

Unlike metallic systems, ionic materials require maintaining charge neutrality and addressing long-range interactions.

  1. How can the CE method in ICET be adapted to handle the unique challenges in modeling multicomponent ionic materials? For example, in TiN system, Ti will have different oxidation states when N content fluctuates. Ideally the cluster expansion should treat Ti+, Ti2+ and Ti3+ differently. So how is this addressed in ICET?

  2. The ground state is largely related to the stable oxidation states of metal, especially when an ordered compound can be formed rather than solid solutions. Therefore, how can we justify that the structures obtained from ICET are indeed the ground state structures?

Here are a few related articles, just for your reference:
https://journals.aps.org/prb/abstract/10.1103/PhysRevB.106.144202
https://onlinelibrary.wiley.com/doi/abs/10.1002/aenm.201903240
https://www.sciencedirect.com/science/article/pii/S2451929419304322

Thank you so much. Look forward to your reply!

Best regards,
Calvin

2

Hi,
Thanks for the kind words.

Icet doesnt have any built in support for this (if I understand the question correctly).

If you want the CE to treat Ti+, Ti2+ and Ti3+ differently, then maybe you can model these as different atomic species and then make a normal cluster expansion?

Regarding the long range interaction, one could maybe subtract this contribution from the DFT energies before training a CE.
Then the CE would only describe the short range interaction.
But then you would also need to add the long ranged contribution back when e.g. running MC simulations which would likely be very slow.

I dont think I understans your second question.

3

Hi,

Thanks for your reply!

This sounds a bit complicated. Since in my system, i.e., TiN binary system, when various nitrogen vacancies form, different Ti atoms may exhibit diffirent oxidation states due to their surrounding environments, for example, some Ti ions are in Ti4+ and some are in Ti3+, I think it’s hard to incorporate all of them at the same time in the same system.

For this, I basically want to know if we cannot address the multiple oxidation states of the transition metal atoms in the system ,. eg., TiN system, when nitrogen vacancies are present, do you think the predictions of the ground state structures we obtained at the end are accurate?

Thank you. Look forward to your reply!

4

This sounds a bit complicated.

Yea I agree.

For this, I basically want to know if we cannot address the multiple oxidation states of the transition metal atoms in the system ,. eg., TiN system, when nitrogen vacancies are present, do you think the predictions of the ground state structures we obtained at the end are accurate?

I think its very hard to give an answer to how much you can trust a GS prediction from CE in general.
In general if you model achieves low train and test errors I think you can trust it (up to those errors), so if the CE you train for your TiN dataset achieves small errors then it maybe learned something about the different environments leading to different interactions.
If you’re looking for specific ground-states you could use the CE to generate a few possible candidate low energy structures and then calculate the energies of these with DFT and confirm the GS.