Finding best cluster expansion

Alloy Theoretic Automated Toolkit (ATAT)
1

Dear Axel,
I am not sure if this question comes under this category; this may be shifted wherever you feel best.
During cluster expansion once a few structures have been relaxed, we use
nohup maps -d &
nohup.out shows
Finding best cluster expansion…
1 2 0 0.0813829
1 2 1 0 0.0830814
1 2 1 1 0 0.0867071
1 3 0 0 0 0.0650377
1 3 1 0 0 0.0686184
1 3 1 1 0 0.0716511
1 3 4 0 0 0.0704228
1 3 4 1 0 0.073742
1 3 4 7 0 0.054389
1 4 0 0 0 0.0542315
1 4 1 0 0 0.0547202
1 4 1 1 0 0.056403
1 4 4 0 0 0.0547204
1 4 4 1 0 0.0551012
1 4 4 7 0 0.0543899
1 4 9 0 0 0.0565688
1 4 9 1 0 0.0573123
1 4 9 7 0 0.0589562
1 4 9 27 0 0.213344
1 5 0 0 0 0.0506445
1 5 1 0 0 0.0524066
1 5 1 1 0 0.0552766
1 5 4 0 0 0.0502338
1 5 4 1 0 0.0514132
1 5 4 7 0 0.0457699
1 5 9 0 0 0.0537437
1 5 9 1 0 0.0554606
1 5 9 7 0 0.0461418
1 5 9 27 0 0.115733
1 5 13 0 0 0.0556557
1 5 13 1 0 0.0570264
1 5 13 7 0 0.0467315
1 5 13 27 0 0.25891
1 6 0 0 0 0.0477022
1 6 1 0 0 0.0493498
1 6 1 1 0 0.0516343
1 6 4 0 0 0.0477661
1 6 4 1 0 0.0485921
1 6 4 7 0 0.0467741
1 6 9 0 0 0.0485482
1 6 9 1 0 0.0503729
1 6 9 7 0 0.0447946
1 6 9 27 0 0.119442
1 6 13 0 0 0.0475224
1 6 13 1 0 0.0498932
1 6 13 7 0 0.0461535
1 6 13 27 0 0.256064
1 6 17 0 0 0.0447904
1 6 17 1 0 0.0463251
1 6 17 7 0 0.0378449
1 6 17 27 0 3.40282e+38
1 7 0 0 0 0.049516
1 7 1 0 0 0.0522961
1 7 1 1 0 0.054571
1 7 4 0 0 0.0496254
1 7 4 1 0 0.05098
1 7 4 7 0 0.0483351
1 7 9 0 0 0.052833
1 7 9 1 0 0.0547243
1 7 9 7 0 0.0482971
1 7 9 27 0 0.141365
1 7 13 0 0 0.0510816
1 7 13 1 0 0.0533867
1 7 13 7 0 0.0491118
1 7 13 27 0 3.40282e+38
1 7 17 0 0 0.0486029
1 7 17 1 0 0.0500784
1 7 17 7 0 0.0364418
1 7 17 27 1 3.40282e+38
Can you please tell me what these numbers (in different columns) mean especially "3.40282e+38". This number "3.40282e+38" appears a couple of times even though ‘mapsrep’ shows a good CV score of 0.0161592…Should I really worry about this.
I am also attaching fit.out (renamed as mapsrep_fit.txt). Can you please suggest if the CV score can be further reduced…Or may I put it as "Should it be further reduced?"

Thanks

Suddhasattwa Ghosh

2

Each line contains one attempt to fit a cluster expansion:
[number of point clusters] [number of pairs] [number of triplets] [number of quadruplets] [was the ground state line ok] [cv score]

A cv score of 3.40282e+38 (MAXFLOAT) is just a sign that this particular attempt had colinear correlations while calculating the cv score (it is flagged with a large number so that it wouldn’t be chosen ever). It’s not a problem as long as one of the fits did give a low cv score.

Given that your formation energies are of the over of 0.2 eV, a cv score of 0.0161592 is pretty good!