Dear All,
I am generating sqs by using different correlations (pair, triple) and cutoff distances. However, I am not sure how to pick up the best sqs among all the structures generated. One possible figure of metric (fom) was proposed in this tutorial (https://grandcentral.apam.columbia.edu:5 … index.html), which can be summarized as follows.
- Run corrdump to compute the correlations of the candidate structures as well as the random alloy (using exactly the same correlation parameters for all the generated structures)
- Calculate the square root of the sums of the squares of the correlation function differences [candidate (in tcorr_final.out) compared to random (in tcorr_finalRND.out)].
- Pick the one with the smallest fom.
My concern with this definition for fom is that it treats all the correlation function differences with the same weight. However, as we know, usually the correlation with smaller distance is more important, e.g., the correlation of the pair between nearest neighbour sites should be more important than the next nearest sites. The Objective_function defined by mcsqs has taken this into consideration by using a term -M-OM-^IL. Thus, it sometimes happened that an sqs has a lower fom but higher Objective_function. I am struggling to define a more reasonable figure of merit. Any suggestion?
Many thanks.
Best regards,
Joe