Hi, I believe this may have been indirectly answered, but I just wanted to clarify again. For the intensive property prediction from cluster expansion, should we be using per site or per energy? For eg. if I have A2B2X6 perovskites, and say I vary the X site, should I use energy / 6 or energy / 10? Is it the same for MC simulations?
Many thanks!
Hello,
I would suggest energy/10.
However, you can try both /6 and /10 to see which gives a more accurate CE fit.
For the cluster expansion itself it doesnt matter, it can be per atom, per site , per formula unit etc.
But for the Monte Carlo its very crucial to get this correct because there you need to go from the CE predicted energy to a total energy of the system to be able to do thermodynamics, and then you need to remember what normalization you used for the cluster expansion.
See the documentation and the parameter scaling Calculators — icet documentation
By default this will be number of total atoms in the supercell, hence assuming the energy predicted by the CE is per atom.
So its a good idea to always use per atom, or per total number of sites if you have vacancies, and then things will work as expected.
Thank you both for the replies! Great to see that I have been using the correct units for now at least.