Ground state convergance issues

[Ava Rajh]

Hi all,

I am trying to compare some ground state XAS and TDDFT XAS for a few systems and it was going great on other systems until I started on a relatively simple graphite cell.

I performed cell optimization and used the resulting structure of 64 C atoms for XAS calculation but i am having issues with ground state calculation of a system which would be compatible with further TDDFT,. SCF does not seem to converge and either diverges wildly or oscillates around 0.

I have included a sample input file, I have tried also with the DIIS OT, denationalization, and with a PBE functional but the issue persist. I am hoping someone might have an idea if I missed something obvious. It seems that the problem might be in the HF part, as the convergence is slightly better if I change the HF fraction, but it ultimately always diverges.

I would be grateful f anyone has any comments on the issue I am facing and I apologize if I missed something obvious.

kind regards, Ava

[Augustin Bussy]

Dear Ava,

you are correct, your issue most likely comes from your Hartree-Fock input. You are using a cutoff radius of 1.5 Ang for the truncated Coulomb potential. This is very small and might  lead to inaccuracies in the electronic structure. Typically, calculations are run with a truncation radius of 5-6 Ang. Note that the sphere with the same radius should fit in your simulation cell: you might need to increase the size of the system to fit such a large truncation radius.

In terms of performance, you are using an extremely tight EPS_PGF_ORB value. This will lead to very slow calculations. The default of 1.0E-5 is usually enough, and 1.0E-6 is already considered very accurate. On the other hand, EPS_SCHWARZ = 1.0E-6 is fairly loose. Using 1.0E-8 might help with numerical stability. If you are going to run a larger system, you will probably need to tweak the HF%MEMORY%MAX_MEMORY value for optimal runtime.

You can also try to use the more modern ADMM basis set defined in the BASIS_ADMM_UZH data file.

Best,

Augustin

---

From: cp...@googlegroups.com <cp...@googlegroups.com> on behalf of Ava Rajh <ava....@gmail.com>
Sent: Friday, January 12, 2024 11:34 AM
To: cp2k <cp...@googlegroups.com>
Subject: [CP2K:19733] Ground state convergance issues

 

--
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+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/0906e169-784f-4a37-ac57-2f15cca9a67fn%40googlegroups.com.

[Ava Rajh]

Dear  Augustin, 

Thank you very much for your input. My cell is quite small in z direction, that's why I had to set the  cutoff radius to 2 (or 1.5). I will try to increase the vertical size of the system and repeat the calculation. 

I will also try other performance optimizations you suggested, and change the basis set. 

Thynk you for the suggestions, 

kind regards, Ava

V V pet., 12. jan. 2024 ob 17:33 je oseba Augustin Bussy <augusti...@chem.uzh.ch> napisala:

You received this message because you are subscribed to a topic in the Google Groups "cp2k" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/cp2k/_CZw56CpD9E/unsubscribe.
To unsubscribe from this group and all its topics, send an email to cp2k+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/ZRAP278MB06748614D3FFD615AC508065D36F2%40ZRAP278MB0674.CHEP278.PROD.OUTLOOK.COM.

[Ava Rajh]

Dear Augustin and others,

I have followed your suggestions and the graphite convergence went wonderfully, and I obtained a good XAS spectra in the end. The calculations and further testing took a while so I am only now getting back to you. I have however stumbled on the same issue when trying to repeat the procedure with a similar system with one additional sodium atom between the graphitic layer. The structure was geometry optimized . well converged, and I am also starting from the well pre-converged GGA wavefunction file, but the SCF with any amout of HFX still won't converge and starts diverging wildly after only a few steps.  I have tried calculating with  up to EPS_SCHWARZ = 1.0E-10, but it does not improve the situation. I have also tried increasing the cell size, to set the cutoff value up to 6A, but the issue persisted.

If you or anyone else has any ideas what else I can try to get the wave function to converge I will be happy for any suggestions you can offer.

thank you for the time and all the help,

Ava