Reproducing 2013 Example for mcsqs

Alloy Theoretic Automated Toolkit (ATAT)
1

I am new to ATAT and to SQS. I have been attempting to learn the mscqs command and its capabilities. As part of my starting off point, I tried to reproduce the example in "Efficient stochastic generation of special quasirandom structure" from 2013. I am aware that someone else has also posted on this same example, however it does not answer my questions.

When I run the same commands exactly as they were given in the example, after what appears to be 8 or 9 attempts, the mscqs.log file stops updating and stops with an objective function of 16.398240. Each time I rerun I get the same value, even after running for many days.

My first assumption is that with such a large cell it should not be done in seconds. My second assumption is that the objective function should be smaller, and there should be fewer mismatches. At what point do you assume it is good enough? Should it look like the example from some point of view when complete?

2

First, it possible for mcsqs to give a first answer or even a final answer in a few seconds (it is a random algorithm not a full enumeration).
I can’t tell just from the objective function if it is good or bad - the correlations’ closeness to the random ones is more informative. (For instance if you input a lot of correlations to match, objective function gets bigger, but the end answer can be just as good.)
If mcsqs seems to not give any further improvements, try to play with the wr or T parameters (using a mcparam.in file to do it at run time - see mcsqs -h ).
The runs in the paper did take at least a few hours.