Dealing with large unit cell

icet
1

Dear all,

I am working on a material system with a large unit cell (84 atoms, over 2000 ų in volume). The material is chemically disordered, and I’m trying to find the ground state configuration.

To make the structure as close to cubic as possible, I have constructed a 1x1x2 supercell. This results in over 6e8 possible configurations.

Even with relatively small cut-off radii of [8.0, 6.0] for the cluster space, I still end up with 1450 parameters in the expansion. Unfortunately, due to limited computational resources, I can only afford to run DFT on ~200 structures, but the annealing procedure suggests that this is too few for a reliable cluster expansion model.

Is there any strategies that can solve this situation?

Best regards,
Zane

2

Hi,
If you can only afford 200 DFT calculations, then I’d suggest to reduce the cutoffs (maybe discard triplets?) such that you end up with maximum around 100 parameters.

Then its always a good idea to try to converge the choice of cutoff wrt to e.g. the cross-validation error.

3

Dear erikfransson,

Thank you very much for your suggestions.

I will give it a try.

Best regards,
Zane