High cv score for ternary alloys

Dear Axel & ATAT users,

I am trying to investigate a ternary iron alloy with light elements at high pressures.
The cv score is as high as over 0.3 from CE fitting the enthalpies of about 130 structures.
If the total energies are used, the cv score is close to 0.1. I think the enthalpy should be used, but it is hard to minimise the cv score because there is dramatic volume change for some structures.
I tried to exclude some structures with significant relaxations, but the cv score doesn’t improve.
Do you have suggestions for my case?

Thank you very much!
Oliver

run

checkrelax

to see if some structures have relax too much (say, more than 0.07).
If the structure are literally not on the same lattice anymore, a cluster expansion can be very difficult.
Then you can
(i) exclude those structures that relax too much (do this only if there are a few).
(ii) if you see that the relaxation are excessive only for some composition range, then try to narrow the composition range.
(iii) you can try the recently proposed scheme in https://dx.doi.org/10.1038/ncomms8559 to handle mechanical instabilities.

If the structures did not relax too much, then look for poorly converged vasp runs (start with those structures that can’t be fitted well).