[CP2K-user] [CP2K:19855] Re: Mostafa Abedi e-mail + help with neural network potential

[Muhammad Saleh]

Hi.

My suggestion is to try to improve your dataset if this happens at the very 
beginning of your MD. For instance, introduce more configurational space. 
If, say, this unphysical configuration happens after several ps or steps, 
then you can take a snapshot of the configuration prior to the crash (10 or 
15 steps) and perform a DFT calculation on this structure, then use the 
result to update your dataset. Well, I can't tell much since I don't know 
much about your forces and energy RMSE as well as the correlation of the 
test set. What kind of descriptor are you using, etc.? You see, the input 
attached is input for the RuNNer, and what I can tell is that this is mode 
2, which is the training process. You might want to check what kind of 
descriptor or NN architecture you are using on the data set production 
(i.e., mode 1). I hope this will help, best of luck 

Best
MuS

On Monday, January 29, 2024 at 2:37:32 AM UTC+1 Marco Gallo wrote:

> Hi I want to ask Mosfata Abedi, 
>
> Questions about the paper that he published titled "J. Chem. Theory 
> Comput. 2023, 19, 7825−7832", 
> But somehow his e-mail given in the JCTC is not working.
>
> Email: mostaf... at brown.edu
>
> I am trying to create neural network potential for an aqueous -peptide 
> system, and I trained the model using 1320 structures obtained from 
> cp2+metadynamics, 
>
> But the trajectory that I obtain using the nnp potential is very unstable.
> Any help is appreciated.
>

-- 
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/15069dec-549c-491c-9296-4699aa48bd21n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20240130/441dda87/attachment.htm>