Perhaps you want to explain what SQS means.

Hello Jan-Michael Mewes, SQS stands for Special Quasi-random Structures. This is a technique/algorithm to generate randomized supercells out of perfectly ordered cells. It involves swapping atoms (by a monte carlo treatment most probably) and trying to mimic pair correlations, among atoms in various nearest neighbour clusters, of a completely random state. The cells generated by this algorithm are also called SQS.

So your question basically is if you need to converge k-points, energy-cutoffs and smearing width in calculations for disordered cells?

In that case, the answer is yes. You always want to converge those. How tight you want to converge them depends on what quantities you are interested in (relative energies, absolute energies, volumes).

Cheers

Jan

Jan-Michael Mewes, these disordered cells might have ~100 atoms and converging all of k points, energy cutoff, and smearing width can be cumbersome. Would it not be okay to just increase the values for an ordered cell by say 1.5 times and then use those for the bigger disordered cell? [for k, keeping the resolution of the mesh intact].

Energy-cutoff and smearing are mostly system dependend and you can use the values obtained for smaller cells for the larger ones. However, k-points do depend on the cell size, you should converge them again if you change your cell size. To work around this, you could converge the k-spacing instead, which automatically takes into account cell size. Also, if your disordered cells all have (roughly) the same size and number of atoms, you really just need to converge the k-points for one of them. Anyhow, its most important you use the same settings for all of the disordered cells since (I guess) you are interested in relative energies.

Jan-Michael Mewes, I was doubtful because I thought all three of them might depend on the chemical environment within the cell which would be different between the ordered and disordered cells of even the same size. Do you think energy cutoff and smearing width wouldn't be affected by the bonding/chemical environment within the cell?

I understand what you mean by k-spacing. Based on the same chemical arrangement argument, I'm unsure if even the k-spacing or resolution would remain the same among different cells, even two cells of the same size - one ordered, the other disordered.

You don't have to worry about that. Consider that people like me do MD simulations where everything changes all the time yet we keep cutoffs and k-points fixed at the previously converged settings.

Jan-Michael Mewes, Ah! That never occurred to me. But do you think taking values for a disordered cell that are 1.3 - 1.5 times that of the converged values for an ordered cell is a safe thing to do or am I just being silly?

If the cell is the same size and contains the same elements (same POTCARs), I would just use the same values. Increasing them will make the results less comparable. If you wanna be sure, just converge them again for one of the disordered cells. But I would expect them to be very similar.

Jan-Michael Mewes, I'm running calculations for convergence on one of the smaller disordered cells. Hopefully, things shall remain the same. Thank you for the discussion, it was helpful.