[CP2K-user] [CP2K:20335] Re: Frames Convergent in MD but Not in Separate SCF Calculations

[Marcella Iannuzzi]

Dear Xie,

As you already suggested, the reason of the slow or no-convergence of the 
SCF  for configurations extracted from the MD trajectory, is the poorer 
initial guess, in comparison to what obtained from the extrapolation scheme 
along the MD.
This is very clear from the at least 2 order of magnitude difference in the 
initial value of the "Convergence" at the MD step, in comparison to what 
obtained in the single point calculation.
The extracted configurations are probably rather distorted, maybe with 
almost broken or very elongated bonds, tending to formation of radicals. 
This might require more robust optimisation algorithms. 

By the way, when the SCF becomes unstable and the charge is not anymore 
conserved, there is no point to continue the simulation, it will never 
recover.
This refers to the SCF for step 5200, where after some outer-scf cycles , 
the energy keeps jumping by several Hartree and the charge on the grid is 
not  zero.

Best
Marcella

    35 OT DIIS     0.15E+00    3.6     0.00067106     -1560.8442586833 
 5.22E-03

  Leaving inner SCF loop after reaching    35 steps.
 
 Electronic density on regular grids:       -724.0022698337       
-0.0022698337
  Core density on regular grids:              723.9999999995       
-0.0000000005
  Total charge density on r-space grids:       -0.0022698341
  Total charge density g-space grids:          -0.0022698341
.......


    35 OT SD       0.15E+00    3.6     0.00066429     -1541.1842316224 
-6.68E-02

  Leaving inner SCF loop after reaching    35 steps.


  Electronic density on regular grids:       -660.6939947665       
63.3060052335
  Core density on regular grids:              723.9999999995       
-0.0000000005
  Total charge density on r-space grids:       63.3060052330
  Total charge density g-space grids:          63.3060052330


On Sunday, June 16, 2024 at 6:35:23 AM UTC+2 Qiming Xie wrote:

> Dear CP2K experts,
>
> I hope this message finds you well. I am writing to seek your assistance 
> regarding an issue I encountered while using CP2K for molecular dynamics 
> (MD) simulations.
>
> After performing 20,000 steps of MD simulations, I extracted 50 frames and 
> carried out separate SCF calculations to check for convergence. However, I 
> found that not all frames converged as expected. I would appreciate any 
> insights you could provide regarding this issue.
>
>    1. 
>    
>    For frames that do converge, the convergence speed during the MD 
>    simulations often achieves convergence within one OUTER_SCF step. However, 
>    when performing separate SCF calculations, it often takes more than three 
>    OUTER_SCF steps to converge. According to the gudieline, the MD process can 
>    automatically utilize the wave functions from previous structures to 
>    extrapolate the initial guess wave function for the current step, which 
>    seems reasonable.
>    2. 
>    
>    However, I have encountered about 10 frames where the separate SCF 
>    calculations do not converge. Could this be related to temperature or 
>    configuration?
>    3. 
>    
>    Regarding parameter settings, I have tried to keep the SCF and MD 
>    calculations as consistent as possible:
>    - Theoretical method: PBE
>       - Basis set and pseudopotential: DZVP-MOLOPT-SR-GTH
>       - Using OT
>       - EPS_SCF for Inner and Outer SCF: 10E-5
>       - CUTOFF and REL_CUTOFF: 400, 50 respectively
>       - MD uses CSVR thermostat with temperature controlled at 300K
>    
> I have attached the input and partial output files for your reference.
>
> Thank you for your time and assistance.
>
> Best regards, 
> Xie
>

-- 
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/4275d483-3574-4826-82c0-952fae200149n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20240616/07401eb6/attachment.htm>