Downloads · 30 days
0
CHGNet-PES-MatPES-PBE-1M-2026.9
BowenD-UCB/CHGNet-PES-MatPES-PBE-1M-2026.9
CHGNet-PES-MatPES-PBE-1M-2026.9 is a machine learning model from BowenD-UCB. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for matgl.
This model is a compact (~1M parameter) CHGNet universal machine learning interatomic potential for the PyTorch Geometric (PyG) backend of MatGL. It was trained from scratch with continuous three-body autograd on the…
Access
Public
Updated Sep 27, 2026
Repo size
8.8 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.pt4.4 MB · 100%
At a glance
- Library
- matgl
- Access
- Public
- Created
- Sep 21, 2026
- Updated
- Sep 27, 2026
- SHA
- a8a8ffe3
Model card
From the Hugging Face model README
Description
This model is a compact (~1M parameter) CHGNet universal machine learning interatomic potential for the PyTorch Geometric (PyG) backend of MatGL. It was trained from scratch with continuous three-body autograd on the Materials Project Potential Energy Surface dataset (MatPES) using the PBE functional.
This release incorporates the supercell line-graph periodic boundary invariance fix (#839), ensuring strict invariance across self-images and supercell representations.
Model Architecture
- Model Type: CHGNet (
matgl.models.CHGNet) - Backend: PyTorch Geometric (PyG)
- Parameters: 1,083,842 (~1M)
- Embedding Dim: 128
- Interaction Blocks: 4
- Conv Hidden Dims: [64]
- Radial Cutoff: 6.0 Å
- Three-body Cutoff: 3.0 Å
Training Dataset
- Dataset: MatPES PBE 2025.2 from the MatPES collection
- Paper: A Foundational Potential Energy Surface Dataset for Materials (arXiv:2503.04070)
- Official Splits:
- Train set size: 391,374 structures (
MatPES-PBE-2025.2-train.json) - Validation set size: 21,743 structures (
MatPES-PBE-2025.2-val.json) - Test set size: 21,660 structures (
MatPES-PBE-2025.2-test.json)
- Train set size: 391,374 structures (
Performance Metrics on MatPES 2025.2
| Split | Structures | Energy MAE | Force MAE | Stress MAE | Magmom MAE |
|---|---|---|---|---|---|
| Train | 391,374 | 22.58 meV/atom | 86.75 meV/Å | 0.4852 GPa | 0.0805 μ_B |
| Validation | 21,743 | 25.60 meV/atom | 111.00 meV/Å | 0.6060 GPa | 0.0840 μ_B |
| Test | 21,660 | 26.72 meV/atom | 110.53 meV/Å | 0.6010 GPa | 0.0877 μ_B |
Usage
import matgl
from pymatgen.core import Structure, Lattice
# 1. Set backend to PyG
matgl.set_backend("PyG")
# 2. Load model from Hugging Face Hub
potential = matgl.load_model("BowenD-UCB/CHGNet-PES-MatPES-PBE-1M-2026.9")
# 3. Create ASE calculator or predict directly
from matgl.ext.ase import PESCalculator
struct = Structure(Lattice.cubic(3.8), ["Cs", "Cl"], [[0, 0, 0], [0.5, 0.5, 0.5]])
calc = PESCalculator(potential=potential)
- Library
- matgl
- Created
- Sep 21, 2026
- Updated
- Sep 27, 2026