Convergence issues of paramagnetic stainless steel

[francesco.maresca]

Hello everyone!

I am curently working on calculating the stacking fault energy of paramagnetic stainless steels (FeNiCr) using SQS supercells. I have a few questions regarding the procedure in order to find a more robust way on performing ionic relaxations on this kind of systems.

So my systems are created using ATAT software in their bulk structure (fcc) and then I use the tilted box aproach to alter the periodicity of the box and create a stacking fault.

Regarding simulating paramagnetism, I had a very hard time converging the systems but eventyally I found a set of mixing parameters that work, but are very slow.

Code: Select all

AMIX     = 0.01
BMIX     = 0.00001
AMIX_MAG = 0.8
BMIX_MAG = 0.00001
NUPDOWN  = 0

With this parameters I perform ionic relaxation first with IBRION = 2, and then when it fails I restart the calculation from CONTCAR with IBRION = 1. The volume of the box is obtained by the minimum volume of the Energy - Volume curve. During the restart I copy the final magnetic moment distribution of the IBRION = 2 to the new run as an input with MAGMOM = ...

Is this methodology correct?

Also I have a very hard time converging the faulted geometries. They oscilate a lot using the same parameters as the bulk ones. I attach one such calculation that has not yet converged. DO you have any comments on how to improve my INCAT in order to be faster and stop oscilating?

Thank you!

[michael_wolloch]

Dear Francseco Maresca,

Welcome to the VASP Forum.

First, I will address your issues with electronic convergence:

• Using a larger smearing size is usually beneficial for convergence in metals. Make sure that your entropy contributions are not too large, but SIGMA=0.2 is usually OK.

• Changing the Mixing tags is usually the last step to take. Before that, I would switch to ALGO=All, which is more robust than the blocked Davidson.

• If you have access to VASP 6.5.1 or higher, you should also use ISEARCH=1, which makes the algorithm even more stable.

• If that still causes issues, I would suggest starting from a non-magnetic CHGCAR file, as suggested in the wiki article on Troubleshooting electronic convergence.

For the ionic relaxation, please consult the Structure optimization page on the wiki if you have not already done so.

• IBRION=2 is probably a good place to start, but you might want to lower POTIM a bit, e.g., to 0.3 to make smaller steps.

• In that case, setting MAXMIX to a negative value (e.g., -30 for magnetic systems) can help to reduce the number of SCF steps for each ionic iteration.

• If that is still not giving you a stable relaxation, consider switching to damped molecular dynamics via IBRION=3. Experiment with SMASS and POTIM there until you get good results. However, take care that you do not leave the local minimum of your stacking fault and relax back into the ground state.

• I don't think that you need to switch the ionic minimization algorithm to IBRION=1. Your convergence criterion on the forces is pretty tight.

I hope this helps. Feel free to post again if you have further questions, and if any of the tips worked, also let me know, so that others can learn what worked and I can lock the topic.

Cheers, Michael
[/list]