For an ordered crystal, we generally converge the k-mesh resolution for a primitive cell or other smaller supercells. We then use this resolution for any other size of the supercell for the same material. Can we do the same with special quasi-random structures, given that they are disordered?

I've noticed that on doing spin-polarised relaxation for various SQS of different sizes of a magnetic material, they each might turn out to have different magnetic moments. That is inconsistent, though I don't know if this is the case with any magnetic material.

I thought since SQS are disordered, the local atomic arrangement/ environment shall be different for different SQS, both, of the same and different sizes. This might lead to different magnetic moments. What is your understanding of the topic?

Also, can we establish a standard method on the convergence of k-mesh resolution for SQS?


1 Answer 1


Very good question.

I think that when you select one of the many different disordered structures as been a representative one, that structure is not disordered anymore, it is an structure as any other. Following this, I think that the convergence test can be done in a similar way that for any other primitive cell.

The problem of course, is when you decide to calculate for several disordered structures. In this case, one convergence study may not work for all the structures.

The big problem is how to select the best structure among all the possible disordered.

  • $\begingroup$ Thank you. SQS is inherently configurationally disordered. Even when we choose one of many SQS, it still remains disordered. I use mcsqs (from Alloy Theoretic Automated Toolkit) which uses a stochastic means to generate SQS, This means an SQS generated for the same material with the same # of atoms will still be a bit different every time I run the code to generate one. Besides, SQS don't have primitive cells. $\endgroup$ Commented Apr 30, 2020 at 16:23
  • $\begingroup$ When it comes to deciding how good an SQS is, I think it depends on the end-use of that SQS. I use it to study phonon dispersion and configurational entropy for which I qualify my SQS based on their configurational entropy calcuated using the Cluster Variation Method. $\endgroup$ Commented Apr 30, 2020 at 16:25
  • $\begingroup$ Well, I just used SQS with ATAT few times, and yes, each time you use mcsqs, you will end with a "final" SQS that can be different fro the previous run. If you select one of the generated structures as the "best" (using any physical chemistry property, like entropy for example) and use it to generate the whole crystal, it is the "primitive" cell, isn't? $\endgroup$
    – Camps
    Commented May 1, 2020 at 0:41
  • $\begingroup$ ATAT implemented two paths in order to achieve SQS structures, mcsqs (based on Monte Carlo stochastic procedure), and by enumeration. MCQS tends to be less expensive computationally but will generate different structures at each run (I guess providing the same seed between calculations could generate reproducibility). We can also use the MAPS (from ATAT) together with a specific DFT engine, to provide cluster expansion. $\endgroup$ Commented May 13, 2020 at 13:17

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .