Cvs of 3.40282e+38 while using maps

Alloy Theoretic Automated Toolkit (ATAT)
1

When using maps, why does it often give a cross validation score of 3.40282e+38 for some choices of clusters? I understand that this is the maximal float. Under what conditions does it decide to report the csv as this number? Is it actually that large?

Thank you for you time.

2

I also noted this. This is what I got in my maps.log file:
"Among structures of known energy, true and predicted ground states agree
No other ground states of 12 atoms/unit cell or less exist.
Crossvalidation score: 8.50706e+37"

Could someone comment if this is an issue? Also I notice that the files gs_str.out and gs.out are empty. Why is that?

Regards,
Aakash

3

A CV of 3.40282e+38 means that the CV score could not be calculated for that choice of clusters because colinear correlations appear when one structure is removed from the fit during the CV calculations.

This usually happens when there are not enough structural energies. The problem is more persistent when many clusters are needed (e.g. if there are many competing ground states with similar energies).

Possible solutions:

  1. just let the program run longer to generate more structures
  2. Check if the predicted energies and ab initio energies are close enough for your purpose and ignore the problem.
  3. Exclude some composition region from the ground state search (e.g. if those regions are masked anyway by other ground states on different lattices). This makes it easier to find cluster expansion with few clusters and minimize the e+38 problem.
  4. Force your own choice of clusters ( via nbclusters.in file, see mmaps -h ).
  5. Please note that it is NEVER a concern if the e+38 appear on the screen during the calculations but do not show up in the log file. The bad CV scores are just ignored in this case on the program does the right, optimal, thing.