[QE-users] My slab system does not converge during optimization.Can you help me?

[Jayfe Anthony Abrea]

Dear Dr. Thomas,

For the long time tweaking and troubleshooting, I was able to provide a way
for my calculations involving Mn to converge. I want to thank you for your
inputs and suggestions.

What I did is to follow your suggestion of changing the Mn pseudopotential
from 1.0.0 to the older 0.3.1 version. I'm not sure why it converges on the
older version in contrast to the newer one. It was the least expected
change that I wanted to do to my setup but somehow it works. I redo the
convergence test using 0.3.1 version and I found the convergence at around
65 Ry cutoff, same as the one suggested by the materialscloud.org site.
There's only a difference of 0.0015 Ry on calculated energies between using
65 Ry cutoff and 120 Ry cutoff. Using beyond 100 Ry cutoff has driven my
computer crazy thus I would opt to use the least cutoff viable for me. I'd
be using the dual of 12 to be safe. I can also revert back to a simpler
setup (e.g. removing eamp, esm; using 'david') for my further calculations.

I am grateful for your time looking at my predicament Dr. Thomas. Such a
pleasure having a correspondence with you. Hoping to keep in touch with you
in the future.

Thank you once again.

Cheers,

Jayfe Anthony Abrea
Graduate Student
University of San Carlos

On Wed, Apr 21, 2021 at 2:48 PM Thomas Brumme <tbrumme at msx.tu-dresden.de>
wrote:

> Dear Jayfe,
>
> having a closer look at your input I'm a bit confused about some things
> but lets try to stick first to your specific
> problem - you mentioned in your first email, that the calculation is
> running with elements other than Mn and
> that you thus think that Mn is the problem.
>
> Now, did you try replacing the pseudopotential?
>
> Looking at the graph of the SSSP:
> https://www.materialscloud.org/discover/sssp/plot/precision/Mn
> I see that the pseudopotential you're using might have some problem - even
> with a dual of 16 it is not really
> looking good until a cutoff of about 120 Ry. Sure, phonons and related
> properties are different from standard
> scf convergence, but it is weird. Maybe try using the older 0.3.1 version
> of the pslibrary.
>
> Did you try to increase the cutoff?
>
> Considering the very same graph of the SSSP, I think that a cutoff of 45
> Ry is just not enough for Mn.
> Try increasing in reasonable steps (i.e., 50, 60, 70) with a dual of 8 or
> 16 for the 0.3.1 or the 1.0.0 version of
> the pslibrary, respectively. If it converges (only one scf, no relaxation,
> but printing the forces) compare results
> to see what might be a converged value. How big is the change in the total
> energy? How do the forces
> converge? What about magnetization? And so on and so forth... The standard
> way to get a reasonable set of
> ecutwfc and ecutrho for your system...
>
> So, concerning your input:
>
> I know that you created the input with some tool, but it does not make
> sense. You should check this...
> First of all, delete everything which you don't use, just to be sure that
> there is no problem. Why do you set
> esm parameters, if you don't use the method? Why is eamp still set to a
> finite value? Before trying some
> different diagonalization methods try the standard 'david'. startingpot
> and startingwfc usually don't need to
> be set.
>
> In summary, I would try to change the pseudo and increase the cutoff. And
> I would clean the input. Then you
> need to make convergence checks.
>
> Regards
>
> Thomas
>
>
> On 4/20/21 4:55 PM, Jayfe Anthony Abrea wrote:
>
> Hello Dr. Thomas,
>
>
>
> First of all, I would like to thank you for giving your inputs on my case
> last week. I do hope you can still extend a helping hand on this
> predicament of mine.
>
>
>
> I would like to give you an update on my predicament on optimization
> process. Attached in this email is my input file in which the eamp was set
> to zero. During calculations, the results are still oscillating for each
> iteration. I didn’t stopped the calculation but I can see that it’s not
> going better. It even gave a positive value for calculated scf energy per
> iteration as seen in this pic:
>
>
>
>
> https://cviscpshs-my.sharepoint.com/:i:/g/personal/jaabrea_cvisc_pshs_edu_ph/Eat1rLSbCO9Mth7ra0er6PwBfrWsXtUj9s7f_mOPkckzgA?e=lQWb4L
>
>
>
> The accuracies for each iteration are very low for each iteration as seen
> in this pic:
>
>
>
>
> https://cviscpshs-my.sharepoint.com/:i:/g/personal/jaabrea_cvisc_pshs_edu_ph/Ebnkk0mfJjlKiFkMffu63HkBYml45BOtNl3q8JjB_JkTJQ?e=ZRpyqc
>
>
>
> Then let me share to you the ongoing result for the magnetization for each
> iteration in this pic:
>
>
>
>
> https://cviscpshs-my.sharepoint.com/:i:/g/personal/jaabrea_cvisc_pshs_edu_ph/Eat1rLSbCO9Mth7ra0er6PwBrm1Ki8UqakTaleeJ8Nleng?e=4oIMv9
>
>
>
> I hope you can shed light to this matter Dr. Thomas. I’m not sure now as
> to why the Mn cannot undergo proper optimization unlike the Fe and Co.
>
>
>
> Best regards,
>
> Jayfe
>
>
>
> Sent from Mail <https://go.microsoft.com/fwlink/?LinkId=550986> for
> Windows 10
>
>
>
> --
> Dr. rer. nat. Thomas Brumme
> Theoretical chemistry
> TU Dresden - BAR / II49
> Helmholtzstr. 18
> 01069 Dresden
>
> Tel:  +49 (0)351 463 40844
>
> email: thomas.brumme at tu-dresden.de
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20210507/94b949c7/attachment.html>