Slab calculation doesn't converge when Environ is added to the job

[Reinaldo Pis Diez]

Dear folks

First of all, apologies if you read this mail on the PWScf list.

I'm trying to simulate a large Iron oxide nanoparticle as a slab to calculate the energy associated to some atom replacements. Experimental data is obtained in aqueous solution, so, it seems reasonable to include solvent effects into my calculations.

I've started optimizing the bulk structure of the Iron oxide using DoJo PPs both for Fe and O. Then, I added some vacuum above and below, and some OH^{-} groups above because those structures are easily hydrated.

Slab jobs (relax only) converge well without solvent, both neutral (unreal) and with 6 negative charges. But when I add solvent effects through Environ (scf only), important oscillations happen for total energy, for electrostatic embedding and for the one-el term when environ contributions are added to the local potential and no SCF is achieved. I've tried both for the negatively charged system and for the neutral (unreal) one.

I've tested almost all the possible combinations in the environ.in file (attached is the last try) without success. I'm using QE 6.5 and Environ 1.1, and as I said above, DoJo PPs from http://www.pseudo-dojo.org/. Attached are the PWScf and Environ input file and the (interrupted) output. I'd appreciate any hints to go forward with these calculations.

Regards,

Reinaldo Pis Diez
Center of Inorganic Chemistry
Department of Chemistry
Natl Univ of La Plata
Argentina

[andr...@gmail.com]

Hi Reinaldo,

Thanks for posting your questions here as well, I have probably missed them in the PWSCF mailing list. There are a couple of things that can be attempted to fix the issue. The first one is related to environ_thr. I see that you have already tried changing it to a larger value, but the value you used may be still too small. The purpose of environ_thr is to avoid that Environ contribution are computed too early during the SCF run. Since these contributions depend strongly on the electronic density, and the starting guess of pw is a random one, we don't want to have random Environ contributions. However, my current understanding is that it is better to add Environ as soon as possible to the pw calculation, usually within 1-3 scf steps. I see that your calculation takes a few iterations to go in the right path, and with a threshold of 5, it need to wait until iteration 16. But after this, the SCF accuracy goes up again to 15, so Environ is again put on hold for a while. This causes an oscillatory behavior in the SCF accuracy, which leads to no convergence. I would start by using a much larger environ_thr, maybe 20 or even 50. Unfortunately this threshold (being based on scf accuracy) is system-size dependent, so systems with many electrons like yours will need larger thresholds.

A second possibility for the poor convergence is the fact that your system has a somewhat open structure and solvent may get too much inside your system. To fix this, you may want to try the solvent-aware correction, i.e. setting solvent_radius to something different from 0 (2.6 for water seems ok) and playing with filling_threshold, starting from the default of 0.825 and going down, if it helps. Note that values lower than 0.7 probably point to something unphysical in the interface and would suggest to use some explicit solvent molecules. 

Let us know if these things help or not.

Best,

Oliviero

Oliviero Andreussi
--
Assistant Professor
Department of Physics
University of North Texas
Email: oliviero....@unt.edu 

Phone: +1-(940)-369-5316
Skype: olivieroandreussi
Web: https://www.materialab.org

[Reinaldo Pis Diez]

Dear Oliviero,

Thank you very much for your answer and your hints for trying to converge my system.

I've made several calculations, testing your three suggestions, that is, increasing environ_thr, a solvent_radius different from cero, and decreasing the filling_threshold. I've also decreased the mixing_beta up to 0.05 in some cases. I've tried both for the neutral system and for the negatively charged one, as this last one is my goal.

Unfortunately, I couldn't achieve selfconsistency in any case. In those cases where mixing_beta was sufficiently small, the scf accuracy went down up to 0.02 or so, but then, it began to oscillate and no convergence was achieved. Moreover, for the charged system I got the message "Warning: Polarization charge not converged" all the time.

I've attached the environ.in and the last output I've obtained for the system with 6 additional electrons.

Any hint will be welcome.

Regards,

Reinaldo

[andr...@gmail.com]

I am sorry that these fixes did not work, we would need to run these tests to check where the problem is, but it may take some time. In the meantime, you could try to increase the verbosity of the environ output (verbose = 2) and look at the cube files that are produced. Sometimes these files can show some clear problems with the system. 

Best,

Oliviero

[Reinaldo Pis Diez]

Thank you Oliviero, I'll try your suggestions and will see if I can learn more about my system.

Regards,

Reinaldo