To account volume relaxation and phase changes

Alloy Theoretic Automated Toolkit (ATAT)
1

Dear all,

I have some questions, more technical ones, related to ATAT and your paper here: https://dx.doi.org/10.1103/PhysRevB.86.134117

This is my first encounter with cluster expansion and ATAT and Cluster Expansion. I hope that you can help me with some basic techniques on performing cluster expansion calculations with ATAT code. I tried to read the manual as well as scan the ATAT forum, but still could not really solve my own problem.

First of all, I use Quantum Espresso.

So, my system is an oxide with vacancies (SrFeO3-x). As the vacancy concentrations increase, the phase of the structures changes from cubic, to tetrahedra, to orthorhombic, and finally to brownmillerite, according to the experiments. My questions are:

  1. Should I do DFT-CE calculations with volume and shape relaxations? After relaxations, should I do static calculations to get the energy? I did relaxed volume and shape, but the results says true ground-states differ from the predicted ground-states. Also, some of the predicted ground-states crosses the convex-hull of the calculated ground-states. I read in this forum that I should constraint maps by defining a smaller window of concentrations with -c0 and -c1. But when I did this, maps often stop generating new structures (stuck at `Finding best structures’ but never generating any new structures). I suspect that the concentration window is to small, though I am not really sure about it.
  2. If I have to fix the volume and shape, then how do I compensate the volume and shape relaxations in contributing to the total energy? In https://dx.doi.org/10.1103/PhysRevB.86.134117, it was mentioned that the volume relaxation was accounted by multiple constant volume calculations for each structure. So, how exactly can we do this with ATAT?
  3. If I have to calculate for each phase, should I make different directories for each phase? How to combine them in the end so that I can get the ECI which covers all the changes? how to define the concentration range (-c0 and -c1) in this case?
  4. When I got this message: `true ground-states differ from the predicted ground-states’ but my cv-score is low enough, should I ignore the message? Can I still use the ECI coming from this calculations?
  5. Sometimes during DFT calculations on the generated structures by maps, it fails to reach convergence. In quantum espresso, the calculations will be terminated immediately and took the latest recorded total energy. What should I do about it? Should I run DFT to this specific directory and write down the converged total energy to `energy’ files by hand?

Thank you very much before.

Best,
Musa

2

I also got similar situation. Hope somebody will be able to reply.

3

  1. Should I do DFT-CE calculations with volume and shape relaxations? After relaxations, should I do static calculations to get the energy?
    > YES
    I did relaxed volume and shape, but the results says true ground-states differ from the predicted ground-states. Also, some of the predicted ground-states crosses the convex-hull of the calculated ground-states. I read in this forum that I should constraint maps by defining a smaller window of concentrations with -c0 and -c1. But when I did this, maps often stop generating new structures (stuck at `Finding best structures’ but never generating any new structures). I suspect that the concentration window is to small, though I am not really sure about it.
    > you could also use checkrelax to see if some structures are relaxing by a lot (say, >0.08). This would mean you are trying to fit to structures that are just too different. Either you delete the offenders or use a method to deal with the instability, such as described in https://dx.doi.org/10.1016/j.calphad.2017.05.005 section 3.2

  2. If I have to fix the volume and shape, then how do I compensate the volume and shape relaxations in contributing to the total energy? In https://dx.doi.org/10.1103/PhysRevB.86.134117, it was mentioned that the volume relaxation was accounted by multiple constant volume calculations for each structure. So, how exactly can we do this with ATAT?
    > don’t fix volume and shape. Maybe you could fix only shape as in https://dx.doi.org/10.1016/j.calphad.2016.02.005 . Doing multiple constant volume relaxation and taking the minimum is fine but not necessary.

  3. If I have to calculate for each phase, should I make different directories for each phase? How to combine them in the end so that I can get the ECI which covers all the changes? how to define the concentration range (-c0 and -c1) in this case?
    > Yes, different directories are needed. See phb -h for how to trace phase boundaries between different crystal structures.

  4. When I got this message: `true ground-states differ from the predicted ground-states’ but my cv-score is low enough, should I ignore the message? Can I still use the ECI coming from this calculations?
    > It depends, are the violation in concentration range of interest? Then be careful. Otherwise it’s probably fine.

  5. Sometimes during DFT calculations on the generated structures by maps, it fails to reach convergence. In quantum espresso, the calculations will be terminated immediately and took the latest recorded total energy. What should I do about it? Should I run DFT to this specific directory and write down the converged total energy to `energy’ files by hand?
    > Yes, you should either flag the misbehaving runs with "touch error" or, even better, re-run them manually and creating the energy file yourself.