matgl icon indicating copy to clipboard operation
matgl copied to clipboard

LAMMPS interface

Open shyuep opened this issue 1 year ago • 4 comments

An interface to LAMMPS for MatGL is needed. This can be done in two ways:

  1. Using LAMMPS's experimental python plugin infrastructure. See https://github.com/lanl/hippynn/blob/development/hippynn/interfaces/lammps_interface/mliap_interface.py for an example. This would not be very efficient code but it would be the easiest to finish quickly and allow users to leverage on LAMMPS workflows.
  2. Write a C interface to LAMMPS. This was done for the TF version of M3GNet and we can perhaps adapt that implementation.

shyuep avatar Jul 20 '23 21:07 shyuep

What about the LAMMPS interface by advancesoftcrop https://github.com/advancesoftcorp/lammps/tree/based-on-lammps_2Jun2022/src/ML-M3GNET ? As far as I can see it is compatible to matgl but internally it still calls python, so the performance is presumably similar to the performance of the ASE interface.

jan-janssen avatar Dec 19 '23 10:12 jan-janssen

I am also looking forward an interface to LAMMPS. Especially, I need to train multiple models and use the model-devision feature to better sample. I have no better way for instance.

SmallBearC avatar May 28 '24 13:05 SmallBearC

Yes, the AdvancedSoft interface works. But we are hoping to develop a more universal solution for MatGL itself.

shyuep avatar May 28 '24 13:05 shyuep

Yes, the AdvancedSoft interface works. But we are hoping to develop a more universal solution for MatGL itself.

As far as I understand it works with LAMMPS input files, but it does not work with LAMMPS as a Python library, right?

jan-janssen avatar May 28 '24 14:05 jan-janssen

Hello, this web page can't open https://github.com/advancesoftcorp/lammps/tree/based-on-lammps_2Jun2022/src/ML-M3GNET. Can you provide other downloading methods

What about the LAMMPS interface by advancesoftcrop https://github.com/advancesoftcorp/lammps/tree/based-on-lammps_2Jun2022/src/ML-M3GNET ? As far as I can see it is compatible to matgl but internally it still calls python, so the performance is presumably similar to the performance of the ASE interface.

duanshaodong123 avatar Feb 21 '25 02:02 duanshaodong123