GNEP

Copyright (2025) Hongfu Huang. This is the GPUMD potential extension package GNEP (Gradient-optimized Neuroevolution Potential). This software is distributed under the GNU General Public License (GPL) version 3.

Prerequisites

• You need to have a GPU card with compute capability no less than 3.5, a CUDA toolkit no older than CUDA 9.0 and A compiler that supports C++14.
• Works for both Linux (with GCC) and Windows (with MSVC) operating systems.

Compile

• Go to the src directory and type make.
• When the compilation finishes, two executables, gnep, nep and gpumd, will be generated in the src directory.

Run

• Go to the directory of an example and type one of the following commands:
  • path/to/gnep
• Only 1 GPU is supported for now.

How to use?

input file gnep.in

prediction          0
output_descriptor   2
type                2 Ge Se
cutoff              7 5
n_max               10 8
l_max               4     # only 3-body：max 8
basis_size          8 8
neuron              70
#energy_shift       1
#lambda_e           1.0          
#lambda_f           2.0        
#lambda_v           0.1
#weight_decay       0.0001  # Applicable to AdamW
#start_lr           0.001
#stop_lr            0.0000001
#lr_cos_restart     1 1 10 2.0 0.8   # minimal: lr_cos_restart 1
batch               1
epoch               400

GPUMD & NEP Manual

Some similar parameter settings and explanations can be found in the GPUMD manual:

• Latest released version: https://gpumd.org/

References

[Huang2026] Hongfu Huang, Junhao Peng, Kaiqi Li, Jian Zhou, Zhimei Sun, Efficient GPU-accelerated training of a neuroevolution potential with analytical gradients, Computer Physics Communications 320, 109994 (2026).