Description
This model is a CHGNet universal potential for the
PyTorch Geometric (PyG) backend of
MatGL. The weights were
directly transferred
from the DGL checkpoint
materialyze/CHGNet-PES-MatPES-PBE-2025.2.10
— no retraining was performed.
The architecture is a faithful PyG port of the original DGL CHGNet implementation.
The DGL implementation has a slight modification from the original PyTorch implementation
by adding directed edge updates; this PyG port preserves that modification.
Training dataset
MatPES-PBE-2024.11: Materials Energy Surface dataset that contains off-equilibrium PBE static calculations.
- Train-Val-Test splitting with mp-id: 0.9 - 0.5 - 0.5
- Train set size: 391241
- Validation set size: 21736
- Test set size: 21735
Performance metrics
Training and validation errors
Identical to the source DGL checkpoint (weights are the same):
| partition | Energy (meV/atom) | Force (meV/Å) | stress (GPa) | magmom (μB) |
|---|
| Train | 26.54 | 81.4 | 0.375 | 0.066 |
| Validation | 32.02 | 123.8 | 0.617 | 0.067 |
| Test | 30.70 | 136.0 | 0.642 | 0.066 |
PyG vs DGL prediction parity
Evaluated on 10 structures spanning diverse chemistries (MoS, Fe, Mo, Al, NaCl, BaTiO₃,
Li₂O, MgO, and their perturbed variants) with both the PyG and DGL models running
independently.
| Quantity | Max |ΔPyG − ΔDGL| |
|---|---|
| Energy/atom (eV) | 0 (exact) |
| Forces (eV/Å) | 1.5 × 10⁻⁷ |
| Stress (GPa) | 4.1 × 10⁻⁶ |
| Magnetic moment (μB) | 0 (exact) |
Non-zero force/stress differences arise from floating-point summation order differences
between DGL and PyG message-passing kernels, not from any model divergence.
All differences are more than two orders of magnitude below the atol = 1e-5 threshold
of the automated parity test.
Usage
1import matgl
2from matgl.ext.pymatgen import Structure2Graph
3import torch
4
5pot = matgl.load_model("BowenD-UCB/CHGNet-PyG-MatPES-PBE-2025.2.10")
6pot.eval()
References
1Deng, B. et al. CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling.
2Nat. Mach. Intell. 1–11 (2023) doi:10.1038/s42256-023-00716-3.
Date: 2025.2.10
Author: Bowen Deng
Metadata
1{
2 "tags": [
3 "matgl",
4 "materials-science",
5 "graph-neural-network"
6 ],
7 "license": "BSD-3-Clause",
8 "author": "Bowen Deng"
9}
