error of NNP simulation

[Kordia Bulrat]

Dear CP2K users,

I tried to run a molecular dynamics simulation using neural network potential (NNP).

I have installed the most recent development version of CP2K and prepared necessary files using n2p2 programs. (https://compphysvienna.github.io/n2p2/index.html)

But the run terminted with an error message "NNP| Symmetry function type not supported"

It seems that symmetry functions format in the input.nn file is not correct.

I serached the cp2k manual, but I could not find relevant format information in cp2k manual.

I attached input, output, and input.nn files.

If anyone has a similar experience,  your suggestion will be of great help.

Thanks in advance, 

Kordia

[Mostafa Abedi]

Hi.

Did you do the NNP training for this system? Your symmetry function (SF) set is really big, especially the angular part. Did you optimize the number of SFs before training? 

I haven't tried using NNP in CP2K, but you can have a look at the water example:

https://github.com/cp2k/cp2k/tree/master/tests/NNP/regtest-1

https://github.com/cp2k/cp2k/tree/master/data/NNP/bulkH2O-jcp2020-cnnp

I use the LAMMPS interface. You can also try that.

Mostafa

--
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+uns...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/1642fee0-b5ad-4100-b2fd-86ca07b3a93cn%40googlegroups.com.

[Kordia Bulrat]

Dear Mostafa,

Thank you for your kind suggestion.

Yes, I trained the potential for the present system. Reference data were obtained from first-principles MD simulation carried out using VASP but smaller cell size.

I reduced the number of symmetry functions using nnp-prune program in n2p2 

OK. I will try LAMMPS

Kordia

2023年4月10日(月) 21:19 Mostafa Abedi <abedimo...@gmail.com>:

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/CAHQ2YDxbeTRNwd-N8oTe-gXEee3NZ3tQ8PJtgh%2BcE%2Bu7gkv2dQ%40mail.gmail.com.

[Vincente Della Balda]

Hi Kordia,

The error message exactly states why the NNP MD does not run: The symmetry functions you would like to use are not supported in CP2K (see here at line 486: https://github.com/cp2k/cp2k/blob/master/src/nnp_environment.F).

If you would like to run the simulation with CP2K you need to stick to the radial symmetry function of type 2 and angular one of type 3. Regarding cutoff functions, cosine (type 1) and hyperbolic tangent (type 2) are supported.

Best,

Vincente

[Kordia Bulrat]

Hi,  Vincente

Thank you for answering my question.

I will check details of symmetry functions as you suggested.

Best,

Kordia

2023年4月11日(火) 17:56 Vincente Della Balda <vinc.de...@gmail.com>:

To view this discussion on the web visit https://groups.google.com/d/msgid/cp2k/8925bd42-3b04-492f-aaa0-31b41e051c39n%40googlegroups.com.