Ni-Al convex hull

Alloy Theoretic Automated Toolkit (ATAT)
1

Dear Dr. Axel and ATAT users,

I wanted to know if anyone has been successful in getting the Ni-Al convex hull. At best, I am able to get 3 ground states besides elemental Al and Ni. However, a quick check on the web shows a few more stable phases. Could anyone comment on how they got the Ni-Al convex hull?

Regards,
Aakash

2

[attachment deleted]

After repeated attempts, I noted that only 4 ground states in the Ni-Al phase diagram show up on the convex hull by cluster expansion. I tried direct runs and also forcing certain clusters using nbclusters.in but still don’t get the Ni-Al convex-hull.

  1. Can someone provide some guidance if this is a case with cluster expansion? I notice that only the highly cubic phases show up (i.e. NiAl and Ni3Al)

  2. Also if I am interested in Ni3Al doped with some elements such as Co, should I be confident in keeping the same vasp.wrap settings? The convex-hull in my point of view was to get an idea of my vasp settings and also see if long-range interactions really matter.

Dr. Axel, please provide some thoughts.

This is the hull and eci plot that I get.

Regards,
Aakash

3

The other phases are not fcc superstructures. You would need to generate multiple cluster expansions, each on different lattices. (For instance, the phase at 50% is bcc-based.)
For some, you may reduce it to a single structure if the phase remains essentially stoichiometric. I am not sure if anyone has previously pursued this to the full extent in this system.

4

Thanks a lot for the suggestion. Just to follow up on this, I am not quite sure why I get the 50% NiAl (CsCl structure) which is bcc. I notice I always has this as one of the ground states. Could you please comment why this happens even if it is not related to the starting fcc structure I use (fcc Ni).

Thanks,
Aakash

5

I think what you are getting is the L10 structure, which looks like a distorted CsCl structure (c/a ratio is different). But if you perturbed the c/a ratio away from the fcc-like symmetry, your ab inito code would relax that structure to CsCl.
This issue of relaxing to a different structure can be addressed by excluding structures that relax too much ( you can find those with checkrelax ). There is a new, more rigorous, way to proceed, see https://dx.doi.org/10.1038/ncomms8559