Dear Dr. Axel,
I was trying to generate A17B3 sqs using 60- 80-, and 100-atom super cell, respectively for hcp lattice. But the correlations of the obtained sqs so far didn’t match that of the random alloy even to the first pairs. Are these results still reasonable? I used mcsqs code. The details of my calculations are included in the attached file. Please check. Is there something wrong in my calculations? Thank you!
I think you are using the code correctly. It’s just just that when the denominator of the concentration (here 20) is large, the cell size to match correlations exactly grows a lot. To see why consider this: in your case the fraction of S-S pairs in the rnd state should be (3/20)*(3/20)=9/400 (a fraction that cannot be reduced). So you need a cell whose number of nn pair is an integer multiple of 400.
But each atom in your structure is linked to 6 pairs. The LCM of 400 and 6 is 1200. So you need 1200 pairs, or 200 atoms/cell.
BTW, the output for 100 atom is already really good: your correlations are only 1% off the true rnd state. I would go with that. Even the 60-atom are only 3% off at most. Not bad at all!
Dear Dr. Axel,
Thank you for your reply! Actually the calculations of sqs generation for A17B3 have been running for over one month before I got these results. It is very time-consuming. Anyhow I can use these results to perform further calculations. Thank you very much!
I am quite sure a fairly good solution would have come up before one month…
You can play with the annealing temperature if the code appear stuck in a local optimum.
Dear Alex, could you please show me how to use annealing technique to avoid local minimum when generating SQS using mcsqs code? Thank you!
Simulating annealing is the algorithm that mcsqs uses!
You can control it during run time (e.g. increasing or lowering temperature T) as explaining in mcsqs -h
OK. I will try it. Thank you very much!