DIRECT_P_MIXING Approach

[ma455...@gmail.com]

Hi all,

I'm a newbie of using cp2k. I'm currently trying to perform all-electron basis sets calculations with the consideration of relativistic effect for heavy elements. I have tried all mixing approaches and found that only DIRECT_P_MIXING works. However, the convergence is slow. It seems the "P_Mix/Diag." SCF steps do not work well. When it changes to "DIIS/Diag." SCF, the convergence is much faster.

I'm wondering are there any keywords that can make the DIIS/Diag. steps starting earlier?

Additionally, I'm new in relativistic effect calculation, I would really appreciate if there are any suggestions associated with that.

Thanks & Regards,

Hongyang

[Krack Matthias (PSI)]

Hi Hongyang

 

DIIS/Diag can provide indeed fast (often the fastest) convergence depending on the system type, but it requires an already pre-converged wavefunction and a filled DIIS buffer. The onset for DIIS/Diag is controlled by the input parameter EPS_DIIS (default: 0.1) and the DIIS buffer size by MAX_DIIS (default: 4). The default values are well chosen for most cases and their variation might not results in a significant gain. If you start with DIIS/Diag too early, the convergence can become unstable.

 

HTH

 

Matthias

 

--
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/d3c52742-e241-4e5a-98d0-06c18cd96114n%40googlegroups.com.

[ma455...@gmail.com]

Hi Matthias,

Thanks you very much for this information! I will give it a try on these two keywords to see if there are any improvements.

Thanks & Regards,

Hongyang