[CP2K-user] Energy Fluctuations in Geometry Optimization

fa...@gmail.com fabia... at gmail.com
Mon Mar 1 10:59:51 UTC 2021

Hi Xiao,

As Lenard pointed out your SCF does not converge, consequently the forces 
are not converged and BFGS cannot optimize the structure.
You should adjust a few parameters:
MAX_SCF 200 ! change if needed to make sure SCF is really converged
ELECTRONIC_TEMPERATURE     2000 ! large smearing helps convergence. If 
needed you can reduce this once the optimization is converged and restart
ALPHA 0.01 !  you can play with this but it probably needs to be << 0.1

To speed up the first few steps of geometry optimization you could consider 
reducing the k-point grid to something like


Once the the geometry is more reasonable you can switch back the the more 
accurate k-point sampling.

On Sunday, 28 February 2021 at 20:36:45 UTC+1 ASSIDUO Network wrote:

> If I'm not mistaken, I think your structure is too far away from the 
> optimized geometry. Your best bet would probably be to get the best 
> structure possible before starting the geometry optimization.
> This is true because I see the following in your output file:
> At the 50th cycle your convergence criterion is around 0.9, nowhere near 
> 1.0E-07, and you also get a message of "SCF run NOT converged".
> I don't really see any issues with the flags you chose, so I think your 
> issue is probably with the structure. The same thing happened to me the 
> other day.
> On Saturday, February 27, 2021 at 7:15:49 PM UTC+2 ti... at hotmail.com 
> wrote:
>> Dear CP2K Users,
>> I'm trying to run a geometry optimization for CoF2 (110) slabs, but the 
>> energy does not seem to decrease. Instead, it fluctuates a lot and 
>> sometimes increases significantly. I already used BFGS as the optimizer, 
>> and the eps_scf was set to 10^(-7). I'm wondering if anyone knows how to 
>> deal with such a problem. My input file, the generated xyz file, and the 
>> output file are attached below.
>> Thanks in advance!
>> Best regards,
>> Xiao
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210301/0541e013/attachment.htm>

More information about the CP2K-user mailing list