Whether the element order in each file needs to be consistent.

[LJH-ST]

Hello,

I am currently working with DeepMD-kit v3.0.2 and dpgen v0.13.1, and I have a question regarding the order of elements in the files. Specifically, does the order of elements in the type_map in the param.json file need to be consistent with the order of elements in type_map.raw from the training data, as well as in the POSCAR file used for LAMMPS simulations?

Any clarification or guidance on this would be greatly appreciated!

[wanghan-iapcm]

No, the deepmd-kit will automatically handle the type mapping between those encoded in the model and those provided by the training data.

[LJH-ST]

Thank you very much for your answer. However, in my attempts, I observed a different phenomenon. When I made the element order in param.json, lammps.data (the LAMMPS structure file), and training_data consistent, the molecular dynamics simulation proceeded smoothly. However, in my initial attempt, when the element order in param.json, lammps.data (the LAMMPS structure file), and training_data was not consistent with each other, the simulation experienced a temperature explosion and failed right from the start (without any errors related to deepmd-kit). I am not sure if this is a critical issue, but I hope this can serve as a reference for other users in similar situations.

[njzjz]

lammps.data does not contain the type mapping information (for example, it does not store the information for which type is hydrogen), so the type in lammps.data must be consistent with the input file. This is an exception.