How to increase fitting accuracy of cluster expansion

Alloy Theoretic Automated Toolkit (ATAT)
1

I tried to do perovskite alloy system CsPb(IBr)3, using VASP and ATAT. I found that maps stopped with bad fitting accuracy, although it can not find any new ground state. The crossvalidation score is 6.5 meV. However, the formation enthalpies of alloys are mostly lower than 15 meV. So the errors are relatively large. I checked the cluster.out. They are only 4 pair figure as interaction parameters. I think the accuracy could be improved if more figures are considered.

My question is: Can I tune the parameters, such as number of figures and crossvalidation score, so that I can increase the stop accuracy for maps?

2

Maps does not stop based on internal accuracy criterion.
When larger structures are needed, sometimes it takes time to generate new ones.
Be patient or use some of the speeding up tricks posted on this forum.