#1 Stagnation in the SCF loop after several relaxation steps by Adriano Di Pietro 27.05.2021 14:13

Dear Fleur community,

I'm currently performing structural relaxation in a film setup composed of the 2 materials Fe/HfO_2. I'm comparing the stability of configurations having the oxygen on top of Fe and in interstitial sites. In the Interstitial configuration I encounter stagnation of the SCF loop (i.e the density does not converge) after several relaxation steps. I imagine this is a pretty common occurrence as structural relaxation on new interfaces can prove quite challenging: what are the typical strategies to fix the problem?

I attached the inp.xml file and the input.txt file for reproducibility

Thanks in advance and keep up the great work!

#2 RE: Stagnation in the SCF loop after several relaxation steps by Gregor 27.05.2021 15:32

There are different degrees of "stagnation" and also different causes. Could you paste the distances from a few of the last iterations and also attach the out.xml file?

#3 RE: Stagnation in the SCF loop after several relaxation steps by Adriano Di Pietro 27.05.2021 17:53

Unfortunately the out.xml file cannot be uploaded as it's too heavy (even if compressed it 2.6 MB which is higher than the allowed 2 MB) . If there is any other way I can send you the out.xml file let me know.
The other files are here:

#4 RE: Stagnation in the SCF loop after several relaxation steps by Gregor 27.05.2021 18:22

Ok, that does not look like a nightmare. Maybe working a little with the mixing history already helps. You could try the following.

1. If there are any, remove all files from the working directory that have names like "mixing_history" or similar, maybe with a suffix.
2. In the inp.xml file set calculationSetup/scfLoop/@maxIterBroyd to 10.
3. Run Fleur for another 50 iterations.

Don't forget to make a backup of the working directory before you do these steps.

The density mixer uses data from previous density mixing steps to improve the convergence behavior. But sometimes this has the opposite effect. We might have such a situation here. Then it is a good idea to remove this mixing history from time to time. Whenever fleur stops in a regular way without a fully converged density it writes out these mixing_history files that are read in when you restart Fleur. This is why we delete these files. The maxIterBroyd tells Fleur after how many iterations the mixing history is reset. So with the changes after every 10 iterations we start with a fresh mixing history. This effect will not become visible in only a few iterations. Therefore I suggest to perform 50 iterations to see whether these were beneficial changes.

Another question: After how many force relaxation steps did this occur? Was it after the 2nd step?

#5 RE: Stagnation in the SCF loop after several relaxation steps by Adriano Di Pietro 27.05.2021 19:45

Thanks a lot for the suggestions, I'll play around and let you know if it helped.
In response to your question: no it was not specifically the second force step. It performed at least 5 or 6 before displaying this strange behavior.

#6 RE: Stagnation in the SCF loop after several relaxation steps by Adriano Di Pietro 28.05.2021 18:49

Indeed, the solution actually helped: after having performed a couple cycles wit MaxIterBroyd set at 10 and converging with some difficulty (ca. 150 Iterations) the system recovered some stability and now has a much more benign convergence behavior.

Thanks a lot!

Xobor Einfach ein eigenes Xobor Forum erstellen
Datenschutz