[CP2K-user] [CP2K:20651] Re: System Exploded During MD Simulation with NequIP Potential

[tkuehne]

Dear Kam-Tung Chan, 

this is to say that your initial structure, at least according to the 
employed NequIP potential, 
has a potential energy that is larger than the minimum energy + 300 K. 
Hence, the released 
potential energy is converted into kinetic energy. 
This is typically due to an insufficiently relaxed/equilibrated initial 
structure, which you can 
easily remedy by conducting a quick geometry optimization, which yields you 
the 0 K structure.  
However, in that case in the beginning of your MD simulation kinetic energy 
will be converted 
into potential energy, but in that case you only have to equilibrate an 
energy difference of just 
300 K. Alternatively, you can also extensively equilibrate your system 
using MASSIVE thermo- 
statting and potentially TEMP_TOL ... 

Best, 
Thomas Kühne

P.S. Obviously, this may also be a consequence that solely according to 
NequIP the structure is 
not equilibrated though it actually is employing another potential. 
On Sunday, September 1, 2024 at 8:03:12 AM UTC+2 Kam-Tung Chan (Tonybill) 
wrote:

> Dear CP2K users,
>
> I ran an MD simulation with a trained machine learning potential, NequIP, 
> using CP2K. However, the system was not stable, and starting the first step 
> of the MD run, the temperature increased from 300 K to 534 K. 
>
> I was following the documentation and the tutorial on Google Colab to 
> construct my input file:
> https://manual.cp2k.org/trunk/methods/machine_learning/nequip.html
>
> https://colab.research.google.com/github/gabriele16/cp2k/blob/nequip-cp2k-colab/colab/allegro-cp2k-tutorial.ipynb
> I am unsure if it is due to the problem of units, as my trained NequIP 
> model has the units in angstrom, eV, and eV/angstrom.
>
> I really appreciate any help you can provide!
>
> Best,
> Tonybill
>

-- 
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/e988661f-9049-4e1f-b556-a50861868c81n%40googlegroups.com.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20240902/bb73c0cd/attachment.htm>