chgnet
Pretrained Universal Neural Network Potential for Charge-informed Atomistic Modeling
Decision gist · record as of 2026-08-14
Yes, with conditions. CHGNet is actively maintained, has no known vulnerabilities, and offers a practical alternative to DFT for structure prediction and relaxation. Install it if you need fast, charge-informed atomistic modeling and can manage the dependencies (torch, pymatgen, ase). Medium install friction is manageable given the prebuilt wheels. Not suitable if you require absolute DFT accuracy or work exclusively with systems outside the Materials Project's coverage.AI-flagged interpretation of the facts on this page — verify before relying
Before you install
- Requires torch and pymatgen; GPU support optional but recommended for large-scale simulations.
- Python ≥3.10 required.
- Medium install friction due to dependencies on torch, pymatgen, and ase.
License · maintenance · safety
Modified BSD (permissive) — Licensed under Modified BSD (permissive), allowing use in commercial and private projects with minimal restrictions beyond attribution and liability disclaimers.
last release 2025-09-22 (326 days) · last repo commit 2026-02-19 · 398 stars
0 known vulnerabilities (OSV.dev, 2026-08-14) · 143,005 downloads/mo, #11,190 on PyPI
Alternatives
Verify before relying
pip install chgnet
from chgnet.model.model import CHGNet
from pymatgen.core import Structure
chgnet = CHGNet.load()
structure = Structure.from_file('structure.cif')
prediction = chgnet.predict_structure(structure)
print(prediction['e']) # energy per atom- Specific accuracy metrics or benchmarks compared to DFT for different material classes beyond the Matbench Discovery reference.
- Memory requirements and computational cost for typical structure relaxation or MD workflows.
- Whether fine-tuning on custom datasets is production-ready or still experimental.
What it is and what it does
CHGNet is a pretrained universal neural network potential designed for atomistic modeling of crystal structures. It predicts energy, forces, stress, and magnetic moments by learning from over 1.5 million structures from the Materials Project, using charge information (inferred from DFT magnetic moments) to capture electron interactions and local ionic environments. The model achieves near-DFT accuracy while being orders of magnitude faster, making it suitable for high-throughput materials discovery and structure optimization.
The package integrates with pymatgen for structure handling and ase for molecular dynamics simulations. It provides multiple pretrained checkpoints (including versions trained on different datasets like MPtrj and R2SCAN) that can be loaded directly or fine-tuned for specific systems. Users can perform direct inference on static structures, run structure relaxations, or conduct charge-informed molecular dynamics without needing to train a model from scratch.
Use it for
- Predict stability and properties of crystal structures for high-throughput materials screening without expensive DFT calculations.
- Optimize atomic positions and cell parameters for new materials or hypothetical structures using CHGNet-based relaxation.
- Run molecular dynamics simulations with charge-informed forces to study thermal properties and phase behavior.
- Fine-tune the pretrained model on a custom dataset to improve accuracy for a specific class of materials or chemical system.
- Calculate elastic properties (bulk modulus, shear modulus) and phonon properties using CHGNet with atomate2 workflows.
Worth the install?
AI-flagged interpretation of the facts on this page. Verify before relying on it.
Yes, with conditions.
CHGNet is actively maintained, has no known vulnerabilities, and offers a practical alternative to DFT for structure prediction and relaxation. Install it if you need fast, charge-informed atomistic modeling and can manage the dependencies (torch, pymatgen, ase). Medium install friction is manageable given the prebuilt wheels. Not suitable if you require absolute DFT accuracy or work exclusively with systems outside the Materials Project's coverage.
Install
chgnet on PyPI
Before you install
Medium install friction due to dependencies on torch, pymatgen, and ase. The package is actively maintained with recent commits and provides prebuilt wheels for Python 3.10–3.12 across major platforms (macOS, Linux, Windows), reducing compilation burden.
Requires torch and pymatgen; GPU support optional but recommended for large-scale simulations. Python ≥3.10 required.
License in practice
Licensed under Modified BSD (permissive), allowing use in commercial and private projects with minimal restrictions beyond attribution and liability disclaimers.
Quickstart
pip install chgnet
from chgnet.model.model import CHGNet
from pymatgen.core import Structure
chgnet = CHGNet.load()
structure = Structure.from_file('structure.cif')
prediction = chgnet.predict_structure(structure)
print(prediction['e']) # energy per atomVerify before relying
- Specific accuracy metrics or benchmarks compared to DFT for different material classes beyond the Matbench Discovery reference.
- Memory requirements and computational cost for typical structure relaxation or MD workflows.
- Whether fine-tuning on custom datasets is production-ready or still experimental.
Package facts
| License | Modified BSD permissive |
| Python support | Supports the current Python release >=3.10 |
| Install friction | Medium. Platform-specific wheel |
| Runtime dependencies | 7 packagesasecythonnumpynvidia-ml-py3pymatgentorchtyping-extensions |
| Maintenance | Actively maintained 326 days since the last release |
| Last repo commit | |
| First released | |
| Downloads | 143,005 / month, #11,190 on PyPI 30-day window, as of 2026-08-14 |
| Known vulnerabilities | None known OSV.dev, checked 2026-08-14 |
| Classifiers | Intended Audience :: Science/ResearchLicense :: OSI Approved :: BSD LicenseOperating System :: OS IndependentProgramming Language :: Python :: 3.10Programming Language :: Python :: 3.11Programming Language :: Python :: 3.12Topic :: Scientific/Engineering :: Artificial IntelligenceTopic :: Scientific/Engineering :: ChemistryTopic :: Scientific/Engineering :: Physics |
Evidence: chgnet-0.4.2-cp310-cp310-macosx_10_9_universal2.whl; chgnet-0.4.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl; chgnet-0.4.2-cp310-cp310-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl; chgnet-0.4.2-cp310-cp310-musllinux_1_2_i686.whl; chgnet-0.4.2-cp310-cp310-musllinux_1_2_x86_64.whl; chgnet-0.4.2-cp310-cp310-win32.whl; chgnet-0.4.2-cp310-cp310-win_amd64.whl; chgnet-0.4.2-cp311-cp311-macosx_10_9_universal2.whl; chgnet-0.4.2-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl; chgnet-0.4.2-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl; chgnet-0.4.2-cp311-cp311-musllinux_1_2_i686.whl; chgnet-0.4.2-cp311-cp311-musllinux_1_2_x86_64.whl; chgnet-0.4.2-cp311-cp311-win32.whl; chgnet-0.4.2-cp311-cp311-win_amd64.whl; chgnet-0.4.2-cp312-cp312-macosx_10_9_universal2.whl; chgnet-0.4.2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl; chgnet-0.4.2-cp312-cp312-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl; chgnet-0.4.2-cp312-cp312-musllinux_1_2_i686.whl; chgnet-0.4.2-cp312-cp312-musllinux_1_2_x86_64.whl; chgnet-0.4.2-cp312-cp312-win32.whl
Tags
Let your AI agent find packages like this
Example. Real query, live index.
You found this page by searching. An agent finds it by wishing: SkillFed indexes 14,416 PyPI packages by what they can do, searchable in plain language.
wish › “neural network potential for materials”
- chgnetCHGNet is a pretrained graph neural network that predicts energy,…
- mace-torchMACE trains and evaluates machine learning interatomic potentials…
- lion-pytorchLion is a PyTorch optimizer that implements an evolved sign momentum…
Give your agent the search over MCP, or paste the wish link into any chat.
More Artificial Intelligence packages
LiteLLM provides a unified Python interface to call 100+ LLM providers (OpenAI, Anthropic, Gemini, Bedrock, Azure, and others) using OpenAI-compatible API format, available as both a Python SDK and a self-hosted AI Gateway proxy server.
Install it if you need to work with multiple LLM providers or want to centralize LLM routing in your organization.
Client library and CLI tool for downloading, uploading, and managing models, datasets, and repositories on the Hugging Face Hub platform.
Install it if you work with Hugging Face Hub models or datasets.
LangChain provides a framework for building agents and LLM-powered applications by composing language models, tools, and memory through a unified API that abstracts over multiple model providers.
hf-xet provides chunk-based deduplication and efficient file transfer for the Hugging Face Hub, enabling faster uploads and downloads of large files with local disk caching.
Tokenizers converts raw text into token sequences for NLP models, with support for training custom vocabularies and using pre-built tokenizers (BPE, WordPiece) optimized for speed via Rust.
Transformers provides a unified framework for loading, fine-tuning, and running state-of-the-art pretrained models across text, vision, audio, video, and multimodal tasks using PyTorch, JAX, or TensorFlow.
Install it if you need to run or train any transformer-based model for NLP, vision, audio, or multimodal tasks.
See also chemprop · fairchem-core · pyocse · symfc · matscipy · phonopy · vesin-torch · anastruct · ViennaRNA · pyxtal