-
A standalone C++ implementation (a class called
NEP3
) of the neuroevolution potential (NEP) as introduced in the GPUMD package (https://github.com/brucefan1983/GPUMD). We stress that there is no external dependence. -
An interface of the
NEP3
class to the CPU version of LAMMPS (https://github.com/lammps/lammps). It can be run with MPI.
-
The
NEP3
C++ class is defined in the following two files:src/nep.h
src/nep.cpp
-
The following folder contains some testing code and results:
test/
-
The
NEP3
C++ class is used as an engine powering the following two Python packages:
-
step 1: Copy
src/nep.h
andsrc/nep.cpp
intointerface/lammps/USER-NEP/
such that you have the following 5 files ininterface/lammps/USER-NEP/
:nep.h
nep.cpp
pair_NEP.h
pair_NEP.cpp
install.sh
-
Step 2: Now you can copy the
USER-NEP/
folder intoYOUR_LAMMPS_PATH/src/
and start to compile LAMMPS in your favorite way. Good luck!
-
atom_style
can only beatomic
-
units
must bemetal
-
Specify the
pair_style
in the following way:pair_style nep YOUR_NEP_MODEL_FILE.txt # YOUR_NEP_MODEL_FILE.txt is your NEP model file (with path) pair_coeff * * # This format is fixed
-
For multi-element system, the atom types must be carefully set. Take a NEP model
NEP_PdCuNiP.txt
as an example. In this NEP model file, the first line isnep3 4 Pd Cu Ni P
. Then in your LAMMPS input file, you must set- Pb atoms to type 1
- Cu atoms to type 2
- Ni atoms to type 3
- P atoms to type 4
-
Some atom types can be missing in the simulated system. For example you can use the above NEP model to simulate a CuNi binary alloy. It is important to make sure to still set Cu atoms to type 2 and Ni atoms to type 3. In this case, atom types 1 and 4 are missing.
-
Currently NEP cannot be hybridized with other potentials in LAMMPS.
-
If you want to calculate the heat current correctly, use the following command to get the 9-component per-atom virial:
compute 1 all centroid/stress/atom NULL
-
If you directly or indirectly use the
NEP3
class here, you are suggested to cite the following paper:- Zheyong Fan, Yanzhou Wang, Penghua Ying, Keke Song, Junjie Wang, Yong Wang, Zezhu Zeng, Ke Xu, Eric Lindgren, J. Magnus Rahm, Alexander J. Gabourie, Jiahui Liu, Haikuan Dong, Jianyang Wu, Yue Chen, Zheng Zhong, Jian Sun, Paul Erhart, Yanjing Su, Tapio Ala-Nissila, GPUMD: A package for constructing accurate machine-learned potentials and performing highly efficient atomistic simulations, The Journal of Chemical Physics 157, 114801 (2022).
-
If you use the LAMMPS interface of the
NEP3
class, a proper citation for LAMMPS is also suggested.