[Feature Request] _Add new atom types_

[gullbrekken]

Summary

Enable addition of several atom types for each element in LAMMPS (pair_style deepmd).

Detailed Description

It would be nice to have the ability to add more atom types to the simulation in LAMMPS for various calculations. That is to have several atom types for each element. Currently, DeePMD does not support this, it seems.

Further Information, Files, and Links

No response

[wanghan-iapcm]

deepmd-kit does not assume that one element has one atom type. actually you may assign any type, which is a int, to any of the atoms in your system

[gullbrekken]

Thanks for the reply! I have a model trained on 4 atom types, and I tried to add a fifth atom type to the LAMMPS data file, but then I get these warnings, telling me the interactions with the fifth atom type will be ignored.

pair_style deepmd ../../../../../../final-potential/frozen_model_1_compressed.pb ../../../../../../final-potential/frozen_model_2_compressed.pb ../../../../../../final-potential/frozen_model_3_compressed.pb ../../../../../../final-potential/frozen_model_4_compressed.pb out_file md.out out_freq ${out_freq2}
pair_style deepmd ../../../../../../final-potential/frozen_model_1_compressed.pb ../../../../../../final-potential/frozen_model_2_compressed.pb ../../../../../../final-potential/frozen_model_3_compressed.pb ../../../../../../final-potential/frozen_model_4_compressed.pb out_file md.out out_freq 10000
Summary of lammps deepmd module ...

Info of deepmd-kit:
installed to: /cluster/home/oystegul/deepmd-kit/deepmd-kit-c
source: v2.2.11-11-gbab92582
source branch: r2
source commit: bab9258
source commit at: 2024-11-19 17:38:14 +0800
support model ver.: 1.1
build variant: cuda
build with tf inc: /cluster/home/oystegul/.virtualenvs/tf_test/lib/python3.12/site-packages/tensorflow/include;/cluster/home/oystegul/.virtualenvs/tf_test/lib/python3.12/site-packages/tensorflow/include
build with tf lib: /cluster/home/oystegul/.virtualenvs/tf_test/lib/python3.12/site-packages/tensorflow/libtensorflow_cc.so.2
set tf intra_op_parallelism_threads: 0
set tf inter_op_parallelism_threads: 0
Info of lammps module:
use deepmd-kit at: /cluster/home/oystegul/deepmd-kit
source:
source branch:
source commit:
source commit at:
build float prec: double
build with tf inc:
build with tf lib:
pair_coeff * *

WARNING: Interaction between types 1 and 5 is set with deepmd, but will be ignored.
Deepmd model has only 4 types, it only computes the mulitbody interaction of types: 1-4. (/cluster/home/oystegul/deepmd-kit/source/lmp/pair_deepmd.cpp:1282)
WARNING: Interaction between types 2 and 5 is set with deepmd, but will be ignored.
Deepmd model has only 4 types, it only computes the mulitbody interaction of types: 1-4. (/cluster/home/oystegul/deepmd-kit/source/lmp/pair_deepmd.cpp:1282)
WARNING: Interaction between types 3 and 5 is set with deepmd, but will be ignored.
Deepmd model has only 4 types, it only computes the mulitbody interaction of types: 1-4. (/cluster/home/oystegul/deepmd-kit/source/lmp/pair_deepmd.cpp:1282)
WARNING: Interaction between types 4 and 5 is set with deepmd, but will be ignored.
Deepmd model has only 4 types, it only computes the mulitbody interaction of types: 1-4. (/cluster/home/oystegul/deepmd-kit/source/lmp/pair_deepmd.cpp:1282)
WARNING: Interaction between types 5 and 5 is set with deepmd, but will be ignored.
Deepmd model has only 4 types, it only computes the mulitbody interaction of types: 1-4. (/cluster/home/oystegul/deepmd-kit/source/lmp/pair_deepmd.cpp:1282)

[wanghan-iapcm]

This outcome is expected because the model has never been trained on the fifth type of atom, and therefore lacks the necessary information to accurately model the potential energy associated with it.

[njzjz]

Did you set types in pair_coeff?

[gullbrekken]

I did not. I wasn't aware this was possible. In which case, the feature is already present and the case can be closed!