[CP2K-user] [CP2K:21272] Re: Energy Instabilities in SCF Convergence of High Impact Energy Collisions

[Quentin Pessemesse]

Hello, 
Your wavefunction seems close to convergence but is trailing when 
convergence fails. I would suggest using a inner SCF loop (20 to 50 iter) 
and an outer scf loop (5 to 20 iter) 
https://manual.cp2k.org/trunk/CP2K_INPUT/FORCE_EVAL/DFT/SCF/OUTER_SCF.html
At the end of each inner loop, the preconditionner will be used again to 
provide a wavefunction that should converge more smoothly. However, 
convergence with diffuse functions can be complicated. You could try 
removing them from the lightest atoms (H), it could tremendously improve 
convergence if it proves too tricky, it should not generally impact your 
results much.
You could also use extrapolation which could help improve MD stability 
(ASPC for instance: 
https://manual.cp2k.org/trunk/CP2K_INPUT/FORCE_EVAL/DFT/QS.html#CP2K_INPUT.FORCE_EVAL.DFT.QS.EXTRAPOLATION).
Last piece of advice would be to reduce the cell size as larger cells 
require larger plane-waves basis sets, which can prove costly while not 
improving the quality of your results.
Do keep in mind that DFT sometimes fail when dealing with complicated 
charge-separation situations where multireference character becomes 
important.
I hope this will help, take care, 
Quentin
Le vendredi 7 mars 2025 à 01:21:04 UTC+1, Nicholas Laws a écrit :

> Hi CP2K Community,
>
> I hope this message finds you well. I am trying to perform a high energy 
> collision of a fast incident neutral (EMI-BF4), which is composed of a 
> cation (EMI+) and an anion (BF4-), and a surface. The "surface" is modeled 
> as a harmonic potential set 10 Å away from the EMI-BF4 neutral. The EMI-BF4 
> neutral travels into the harmonic potential at 70 eV, but begins to exhibit 
> energy instabilities approximately 150 MD steps following collision. As a 
> result, the fragmented species do not continue to propagate and equilibrate 
> as expected. In my setup, I use the wB97X-D functional, an aug-cc-pVTZ 
> basis set (chosen specifically for its diffuse functions to better describe 
> the negative charge distribution and long-range interactions of BF4-), and 
> all-electron potentials. To address the observed convergence issues, I have 
> tried:
>
>    1. Switching to DIAGONALIZATION in the SCF
>    2. Increasing the number of added MOs (ADDED_MOS)
>    3. Experimented with MIXING (Broyden) parameters
>
> yet the SCF still struggles to converge reliably beyond the collision 
> event. My goal is to obtain precise atomic partial charges on the 
> post-collision fragments, ideally showing near-integer formal charges that 
> reveal how the original cation and anion charges are redistributed.
>
> I was curious if the community could suggest additional strategies or 
> settings within CP2K that might stabilize the SCF convergence under these 
> conditions? I’ve attached my input, output, and trajectory files in hopes 
> that they clarify the nature of the instabilities.
>
> Thank you very much for any guidance you can provide, and please let me 
> know if there is any further information I can supply.
>
> All my best,
> Nick
>

-- 
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 visit https://groups.google.com/d/msgid/cp2k/738f62c5-b1bf-4dea-b025-4e9e8310079dn%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20250307/3fae7640/attachment.htm>