Interpretation of the mcsqs results?

Hello Axel:

I hope to generate a 40-atom sqs containing 1 V, 9 Cr, 10 Mn, 10 Fe and 10 Ni atoms. These 40 atoms are assigned in a size 125 supercell composed of FCC unit cell (each FCC unit composed of 4 atoms; the dimension is 1 by 2 by 5, so 10 such FCC unit). For each atom site, the probability of the appearance of the element is according to its composition.

I only hope to control the nearest pair configuration, so I use corrdump to generate the corresponding cluster.out file and the output looks fine.

I then use mcsqs -n=40 to let the algorithm find the best SQS. The mcsqs.log keeps updating for some time and then stops. To make sure it indeed stop updating, I wait for 3 hrs and finally stop the script.

Below is what I get from bestcorr.out

2 2.538513 0.053863 0.054161 -0.000298
2 2.538513 0.011420 0.005533 0.005887
2 2.538513 0.046875 0.047656 -0.000781
2 2.538513 0.010842 0.003420 0.007422
2 2.538513 0.000000 0.000565 -0.000565
2 2.538513 0.004252 0.004869 -0.000617
2 2.538513 0.000000 0.000349 -0.000349
2 2.538513 0.039887 0.041933 -0.002045
2 2.538513 0.012535 0.003009 0.009526
2 2.538513 -0.000000 0.000216 -0.000216

As you can see, thee out of ten pair correlations are 0, which is perfect. However, I’m wondering if mcsqs can reduce the other seven pair correlations to 0 as well? (Or to put the question the other way, when the algorithm stops, has it sampled all possible configurations and gives the one with the smallest correlation of all?)

I’ve also tried to change the temperature setting (1, 5 and 10) by "mcsqs -T=XXX" to see if I can obtain better results, but the above cluster.out is the best I can get (and this is for -T = 1)

I’ve also put my rndstr.in file in the attachment if you need to know more detail of the cell setup. Please let me know your thoughts on this problem. Thanks in advance!

Congyi

The only way to know for sure if the solution is the best possible is to use an exhaustive enumeration algorithm (as is gensqs). But in your case, you have of the order of 4^40 configurations, so this is not feasible and I also doubt that a better solution exists Because you have a 1/40 composition for one of the species. On the other hand this solution is quite good already. If you want better, use a bigger cell