[CP2K-user] [CP2K:21570] Re: Implicit solvent DFT calculation - charge runaway using SCCS

[Andreas Döll]

Hello,

I tried using SCCS with a smaller system which is well converged without it 
and there I had no problem, so it was the system which led to problems. 
Thank you for your input!
For the convergence issues, if they continue to persist, I will open a new 
thread. I found out that the &CELL section was a bit problematic. Before it 
looked like this:

&CELL
      ABC 36.20 33.51 3.85
      ALPHA_BETA_GAMMA 90.00 90.00 120.00
      PERIODIC XYZ
    &END CELL

With the following section my geometry looks reasonable, although the 
energy jumps quite a bit around, going from reasonable energies which I 
found previously for my system (around -558) to insane values and then 
dropping back down
&CELL
        A 36.20 0 0
        B -19.35 33.51 0
        C 0 0 15.0
      PERIODIC XYZ
    &END CELL

So it seems my issues with SCCS where of different origin.

Cheers, 
Andreas


Andreas Döll schrieb am Donnerstag, 12. Juni 2025 um 10:49:54 UTC+2:

> Hello Marcella,
>
> these covalent organic framework structures I want to study in the end via 
> MD in lammps are porous structures (need partial RESP charges as input from 
> DFT), so although solvent between layers is probably not possible,
> the solvent can be in the pore of the structure. I want to study if the 
> effect of the solvent affects the charge distribution is a major way, that 
> is why I was interested in implicit solvent in the first place.
> I have included kpoints Monkhorst 2 2 10 and removed SCCS for now, as well 
> as OT as it clashes with Kpoints.
>
> Now I can report back to You:
> I have problems getting the the geometric optimization to work out, even 
> when removing SCCS! I head that implicit solvent helps out with 
> convergence, that is why I thought maybe SCCS will help out to solve 2 
> problems at the same time. Do You have any advice how to properly converge 
> the structure? I did a preoptimization for the initial structure with a 
> generic forcefield in avogadro, but still my SCF does not converge, wether 
> I try with OT or kpoints or mixing.. I am a bit lost on how to proceed.
> Any help is welcome.
>
> Sincerely, 
> Andreas Doell
>
> Marcella Iannuzzi schrieb am Freitag, 6. Juni 2025 um 10:24:43 UTC+2:
>
>> Hi Andreas, 
>>
>> I see that the cell  is very small along z. Where is supposed to be the 
>> solvent?
>> Anyway, such small cells are in general problematic without kponts.
>> Have you tried without SCCS?
>> Regards
>> Marcella
>>
>> On Wednesday, June 4, 2025 at 6:15:47 PM UTC+2 andreas... at gmail.com 
>> wrote:
>>
>>> Hello,thank you for your very fast response. I immediately reran my 
>>> simulations!
>>> You are right. My cutoff was quite low, because I played around with the 
>>> parameters when debugging. Same holds true for OT. I reran my calculations 
>>> with
>>> - cutoff 400 Ry
>>> -No OT to get mixing, with Max_scf=500 still and outer_scf=50, 
>>> eps_outer=1.0E-5, eps_eigval 1.E-5Are there any other settings which are 
>>> not too system specific which need to be adjusted, maybe is the convergence 
>>> criterion too stringent? I am also currently struggling to converge 
>>> another, larger system (2 layers of the COF structure attached in my first 
>>> post, 264 atoms in total) where I have done pre-geometric optimization with 
>>> known forcefields from MD simulations to obtain a decent initial structure.
>>>
>>> I attached the updated files and still obtain the problem mentioned 
>>> above, my charge runs away (I had to restart my simulation, because I did 
>>> not achieve convergence but the charge runaway happens way before that)
>>>
>>> I hope for any discussion on this matter. 
>>> Thank you very much in advance.
>>>
>>> Sincerely, 
>>> Andreas Doell
>>>
>>> Marcella Iannuzzi schrieb am Montag, 2. Juni 2025 um 15:08:20 UTC+2:
>>>
>>>> Hello Andreas, 
>>>>
>>>> Are you sure that the problem is SCCS? 
>>>> I ask because the DFT settings seem not to be optimal.
>>>> In particular, the PW cutoff is very low, which could explain the 
>>>> incorrect calculation of the electron density and as a consequence of the 
>>>> total charge.
>>>>
>>>> As side comments, the mixing of the density  is not applied when OT is 
>>>> active.
>>>>  With OT, it helps running fewer iterations in the inner loop, such 
>>>> that the preconditioner is updated.
>>>>
>>>> Regards
>>>> Marcella
>>>>
>>>>
>>>>
>>>> On Monday, June 2, 2025 at 2:00:46 PM UTC+2 andreas... at gmail.com wrote:
>>>>
>>>>> My original message was too large, which is why I attach the output in 
>>>>> a seperate file. Due to its size it seems I have to cut it into pieces, so 
>>>>> I present the first and last slice of the output where you can observe the 
>>>>> charge going from 0.000002 to 73, as well as the "Cholesky decomposition 
>>>>> failed. Matrix ill conditioned ?" after which the run fails.
>>>>>
>>>>>
>>>>> Thanks You all in advance!
>>>>>
>>>>> Andreas Döll schrieb am Montag, 2. Juni 2025 um 13:53:44 UTC+2:
>>>>>
>>>>>> Hello cp2k community, 
>>>>>>
>>>>>> I want to calculate the RESP charges of a structure respecting 
>>>>>> periodic boundaries in a solvent of epsilon=8. For that I want to do a 
>>>>>> geometric optimization using the implicit solvent model. However, after 
>>>>>> some outer SCF iterations I get the following:
>>>>>>
>>>>>> *** WARNING in qs_sccs.F:458 :: The system (solute) has a 
>>>>>> non-negligible ***
>>>>>>  *** charge of +.000002. It is recommended to use non-periodic 
>>>>>> boundary   ***
>>>>>>  *** conditions (PERIODIC none) combined with an appropriate Poisson 
>>>>>>      ***
>>>>>>  *** solver (POISSON_SOLVER MT or analytic)
>>>>>>  
>>>>>> This charge deviation propagates until the end where I have a charge 
>>>>>> of 73! 
>>>>>> I attached input and output file. I do not understand why suddenly I 
>>>>>> have a leftover charge, which then goes wild. Is it not possible to use 
>>>>>> SCCS in periodic systems?
>>>>>>
>>>>>> Cheers, 
>>>>>> Andreas
>>>>>>
>>>>>>
>>>>>>

-- 
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/08901909-2cb4-4ad1-9382-0dd016a470dcn%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20250624/453271be/attachment-0001.htm>