Increasing Cluster diameter of pairs, triplets

Alloy Theoretic Automated Toolkit (ATAT)
1

[attachment deleted]

Dear Dr. Axel,

I want to increase my cluster diameter (or the distance of interaction) being considered for cluster expansion from 5 angstrom, which I think is the default, to a higher value. Maps does not have a command line option to do this. Could you please suggest a way to do this?

I notice that the distance is always up to 5 angstrom and not higher.

Regards,
Aakash

2

Thanks a lot for the reply. I agree we could supply our own clusters. I am trying to generate the convex hull for Ni-Al system. I know my VASP settings are good enough. I just noticed that in all the different cases, the cluster diameter is ~5 angstrom. Isn’t there any way we could increase the cluster diameter?

If not, could you please suggest how I could manually provide better clusters? Is this a hit and trial method? Thanks so much for your input.

Regards,
Aakash

3

You can only supply the number of clusters, rather than the distance.

I don’t think maps has a restriction on the cluster diameter. It might be in your case, maps just finds this cluster set preferable than larger sets.

4

Finding the optimal cluster set seems to be a problem of its own. There are different criteria, and different approaches to those criteria. ATAT uses CV score minimization. https://link.springer.com/article/10.100 … 4-3#page-1 has more detailed information.

As far as a user, I only know one way: nbclusters.in, and let maps decide how far the diameter goes.

5

Thanks terencelz,

I do not know how to provide my own clusters. Could you provide some guidelines on how you decide to choose the clusters in nbclusters.in?

Also, I noted that the command gce can be used to control the cluster diameter, but I am not sure how to incorporate it with my maps search for ground states. Could some provide any help?

Here is the man page for this command:
https://www.brown.edu/Departments/Engine … ode41.html

Regards,
Aakash

6

I believe gce is the generalized corrdump, seeing how they share similar calling signatures. These two commands don’t fit the structures to get coefficients, but use clusters and their coefficients to get the properties, like scaler energy, and tensor elastic constants.

As to maps, nbclusters.in just looks like

10 8 5

It means 10 pairs, 8 triplets and 5 quadruplets.

By now, I think you have to let maps decide what specific clusters it pleases to choose.

Why these numbers? I don’t know the strict criterion.

7

Thanks a lot again for your comment. So I resubmitted a job with nbclusters.in as 10 8 5 as suggested by you. I do notice that now the cluster diameter is greater than 5 for the pairs. So I guess it just depends on your choice of clusters. I hope this will help me get the correct convex-hull for Ni-Al. Although so far I have just found 2 ground states.

[attachment deleted]