Dear Axel,
I’d like to use the command ‘clusterexpand’ to expand other properties of alloys. Then I have to set up some criterion for converge, while the command ‘cv’ seems to work only for the energy to expand. I am wondering is there any method to make the ‘cv’ work for other properties?
Thank you very much!
cv is used to match the total energy only and get the proper set of clusters, which is crucial because it draws the characteristics (genes) of the system.
After that, with clusterexpand, you use the already existent set of clusters to find out the eci for that specific physical quantity. It’s a least square fit. Why do you need a convergence criterion?
Thank you for your reply! I think the least square fit only ensures the proper description of already-known structures, but I don’t know how the predict power of the ECIs is. Thus I’d like to take advantage of the cross validation.
Well, I have a look at the "clusterexpand" script and there is a ‘-cv’ option to print the cross validation.
Good answer: use the -cv option! I’ve now added this to the help file (sorry it was missing).
The best clusters for energy may not be the best for the prediction of another quantity, so it is useful to check. But the best clusters will typically not be very very different either.