LAMMPS interface

[Jimmy Chapman]

Hello,

I would like to know if an interface to LAMMPS has been considered? As you are probably more than aware, ASE is nice for MD, but limited. Interfacing with something like LAMMPS would allow for users to tackle much larger systems and reach longer timescales.

[Nong Artrith]

Hi James.

It would be great to have an aenet-LAMMPS interface, but to my knowledge nobody is working on an implementation at the moment.  Personally, I do not know the LAMMPS source code very well.  However, if you want to work on an interface yourself, we can assist with the aenet side.

Best,
Nong

[Efrem Braun]

FYI, there is an existing Behler-Parrinello neural net interface to LAMMPS here: https://github.com/biklooost/PROPhet. I wrote an add-on script to Amp (a different neural net code) that writes out the Amp-trained neural net in a format that PROPhet/LAMMPS can take as an input (see the save_to_prophet() function here: https://bitbucket.org/andrewpeterson/amp/src/a80158ba4871ce3dd51f643a5255463d0921a3ce/amp/convert.py?at=master&fileviewer=file-view-default). It would be possible to write a similar extension for aenet.

The downside of this approach is that updates to aenet would need similar updates to the PROPhet code. For example, Chebyshev polynomials are not implemented in PROPhet. It would probably be more preferable for aenet to have its own extension. But if you're looking for something quick, writing a converter from aenet to PROPhet could do the trick.

Efrem Braun

--
You received this message because you are subscribed to the Google Groups "aenet" group.
To unsubscribe from this group and stop receiving emails from it, send an email to aenet+unsubscribe@googlegroups.com.
To post to this group, send email to ae...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Hasan Babaei]

Hello All,

I also need to do MD using aenet generated potential and I am thinking of converting aenet potential to prophet in order to use its lammps interface. 

I was wondering if you could help me with the structure of the aenet generated potential. Its a binary file. It would be great if you could let me know the easiest way to get access to the structure and the parameters in the potential file.

Best,

Hasan