[CP2K-user] [CP2K:17022] Re: SCF Run Not Converged for CELL_OPT of Cu2O

[Geng Sun]

Hi,
I also experienced a few jobs that SCF can’t get converged in many iterations ( ~300 SCF iterations). All the jobs  involve magnetic moments located in transition metal cations.
My experience is that if I knew the ground-state magnetic moments (for example, I knew the magnetic moments of  CuO from literature),  I can reasonably set up the initial magnetic moments and the SCF usually converged smoothly in this case.
However, If I did not know the ground-state magnetic moments, such as copper oxide cluster with mixing Cu(I) and Cu(II) for which I can not guess a reasonable magnetic moments for each Cu cation, the SCF convergence was usually pretty difficult.

For the latter case, the OT method is usually trapped in a very strange electronic structure (like negative band gap) during structure optimization.
Only DIAG will work, and normally, I will need a very small ALPHA (0.02) and small BETA (0.0001) in BROYDEN_MIXING. The maximum iteration number is expected to be very large (~300).


To achieve SCF convergence for magnetic system, I noticed that the skills are very different respecting to different packages.  VASP seemly encourages users to set up a large magnetic moments for each atom. However, CP2K seems to calls for a reasonable initial guess.

Best,
Geng


Sent from Mail<https://go.microsoft.com/fwlink/?LinkId=550986> for Windows

From: Matt Watkins<mailto:mattwatkinsuk at gmail.com>
Sent: Monday, May 23, 2022 7:02 PM
To: cp2k<mailto:cp2k at googlegroups.com>
Subject: [CP2K:17017] Re: SCF Run Not Converged for CELL_OPT of Cu2O

Hello,
- wrt non-periodic systems it makes no sense to do a cell optimization on a non-periodic direction.You are only changing how much vacuum is there and that should make no difference if things are OK numerically - this also means the cell opt is not well defined (no optimal cell parameters in non-periodic directions).
- for the CuO system you need to understand what the electronic structure is expected to be - If it is metallic use smearing and diagonalisation. If you use +U to get a Mott insulator solution then you need to check the occupied states and see that they make sense. Check the SCF loop for warning messages about not converging or spurious Mulliken charges.
Otherwise, check your pw cutoff is large enough to get a sensible stress tensor.
Matt
On Friday, 20 May 2022 at 09:56:58 UTC+1 wave... at gmail.com wrote:
Hi Chelsea

I'm not sure about metal oxides, but the manual does mention that OT converges poorly with metals. I also had to avoid OT for pure Au.
Do you or anyone else know whether CELL_OPT is appropriate for non-periodic systems (e.g., molecules, nanoparticles).

Kind Regards

Sam
On Thursday, May 19, 2022 at 2:37:22 PM UTC+2 uri... at gmail.com wrote:
Dear  Marcella

Thank you very much for your help and the scf-coverge problem for cel-opt has been solved. The results are listed as follows:
-DIAG would perfectly solve the problem for cel-opt yet the following geo-opt still fluctuated without converge under same parameter.
- tighter EPS_DEFAULT didn't work with OT but would benefit for converge with DIAG method.
- higher PW-cutoff and changing the U_MINUS_J value both failed to converge the scf cycle with OT method in my cel-opt case

Thank you again for your help!

Regards
Chelsea


在2022年5月19日星期四 UTC+8 20:20:08<Faye Guo> 写道：

Dear all

The problem has been solved, the Diagonalisation rather than OT would converge the scf cycle and CEL-OPT.

However, a new one occurred in the following GEO-OPT which could hardly converge with an increasing total force and DFT+U energy. I've attached the input and output and I would really appreciate the suggestions.

Thanks in advance.
Bests

Chelsea
在2022年5月10日星期二 UTC+8 10:23:59<Faye Guo> 写道：
Dear  Marcella

Thank you very much for your help. I've watched some of your lectures on youtube which are of great help to me as a very beginner in cp2k and computaional method.

As to my case:
1. Smaller U-J value from 5-6.8 still failed to converge the scf cycle
2. I'm confused about whether DFT+U method is right for Cu2O optimization as the magnetic moment is only about 0.012uB( data from material project). And I do read some papers discussing Cu2O modeling without +U information
3. I've tried the DIIS minimizer method which did converge the scf cycle but brought about an abnormal GEO-optimization result. I guess this might related to that DIIS method is likely to give an unstable solution
I'll try the other options to see if there's anyone could make it.

I've uploaded the input and output files for both DIIS and CG methods and I really appreciated your help.

Bests
Chelsea
在2022年5月9日星期一 UTC+8 16:38:08<Marcella Iannuzzi> 写道：
Dear Chelsea

Some things that you can try are:
-use diagonalisation instead of OT
- use tighter EPS_DEFAULT
- higher PW-cutoff
- change the U_MINUS_J value

Without having the output it is hard to guess what is going on.

Regards
Marcella

On Sunday, May 8, 2022 at 5:16:12 PM UTC+2 uri... at gmail.com wrote:
Dear All,

I've been trying to perform a CELL_OPT for  Cu2O  but the calculation sadly cannot converge the SCF cycles.

 I tried all possible OT preconditioners with CG as minimizer by cp2k 6.1 edition which all failed and spin contamination even occurred under full_all condition.

I also tried the increase of max_scf from 100 to 400 as well as the energy gap from 1*10-4 to 1*10-2, yet neither of them worked.

So I'm wondering if any else had these issues and found ways to overcome them? And the input file is attached as follows.


Any help would be much appreciated. Thanks in advance.
Bests

Chelsea
--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com<mailto:cp2k+unsubscribe at googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/3981d4c1-a387-4f26-8103-530b558f61d3n%40googlegroups.com<https://groups.google.com/d/msgid/cp2k/3981d4c1-a387-4f26-8103-530b558f61d3n%40googlegroups.com?utm_medium=email&utm_source=footer>.

-- 
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+unsubscribe at googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/SA0PR07MB764439B98E1423ACFCBC6379C9D79%40SA0PR07MB7644.namprd07.prod.outlook.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20220524/d5aa4835/attachment-0001.htm>