Large unit cells: ATAT upper limit?

Alloy Theoretic Automated Toolkit (ATAT)
1

The short quesion is:

What is, roughly, the (practical) upper limit for the size (i.e. number of sites) of the unit cell for a binary compound that can be treated with ATAT (maps, mmaps)?

The long version:

We want to use ATAT to make a CE of the total energy, but we get stuck in the generation of new configurations (when running maps or mmaps). We think that the reason for this is the big size of our unit cell (46 sites, not reducible to a smaller parent lattice), because the number of configurations is combinatorially explosive (~2^46) and, to the best of our knowledge, ATAT resorts to full enumerations of substitutional configurations (is that correct?).
If, e.g., we try genstr (with n=46 so it doesn’t try supercells) it seems to be pursuing an exhaustive generation.

One of the compounds we are studying is a binary A_x B_(46-x). Somehow we can not get ATAT to work: after proposing the first two pure extremes (all A, all B) and one dilute x=1 structure (additionally another x=2 str if using mmaps), maps (mmaps) get stuck in the "Finding best structure…" part. We left it running for about 24 hours without luck. We are compiling with the -DSLOWENUMALGO removed, as explained in the forum (if not, we can not even obtain the x=1 structure).

Below is an excerpt or the lat.in:


10.5147998310000 0.00000000000000 0.00000000000000
0.00000000000000 10.5147998310000 0.00000000000000
0.00000000000000 0.00000000000000 10.5147998310000
1.00000000000000 0.00000000000000 0.00000000000000
0.00000000000000 1.00000000000000 0.00000000000000
0.00000000000000 0.00000000000000 1.00000000000000 
0.00000000000000 0.30363237396369 0.11601266785864 A, B
0.00000000000000 0.69636762603631 0.11601266785864 A, B
0.00000000000000 0.30363237396369 0.88398733214136 A, B
0.00000000000000 0.69636762603631 0.88398733214136 A, B
0.11601266785864 0.00000000000000 0.30363237396369 A, B
0.11601266785864 0.00000000000000 0.69636762603631 A, B
0.88398733214136 0.00000000000000 0.30363237396369 A, B
0.88398733214136 0.00000000000000 0.69636762603631 A, B
.........................(36 more lines with different sites)....
0.00000000000000 0.50000000000000 0.25000000000000 A, B
0.00000000000000 0.50000000000000 0.75000000000000 A, B
2

I’m sorry but this is a very difficult problem to solve elegantly.
Two possibilities:
Generate some interaction clusters
corrdump -clus -2=[nearest neighbor distance]
If this gives you a very long list of distinct nearest neighbor pairs (>30) then I have no practical solution.
Otherwise, you might either:

  1. approach the problem using SQS (see mcsqs). That would at least give you a regular solution model (under a random solution approximation)
  2. generate your own randomly chosen configurations (with the help of cellcvrt to create a supercell) and fit your cluster expansion "manually" with clusterexpand.