EdenGNN

This repository hosts pretrained EdenGNN checkpoints and reference configuration files associated with the paper Efficient equivariant framework for universal charge density prediction. EdenGNN (Equivariant Density Graph Neural Network) predicts charge densities from atomic structures and can be coupled with downstream DFT workflows for electronic-structure calculations.

Important note on configuration files

The config.yaml files stored in this repository are for reference only. They are not plug-and-play configuration files and must be modified based on the user's own environment before use. In particular, users should update paths, software-specific templates, dataset locations, checkpoint paths, output directories, and runtime settings to match their local setup.

For full installation instructions, configuration details, and workflow usage, please refer to the main EdenGNN code repository:

• https://github.com/rubenlee11/EdenGNN

Repository structure

mp/

Pretrained checkpoints and reference configs for the universal models trained on Materials Project non-magnetic materials database.

mp/vasp/

VASP-based universal checkpoints for PAW charge-density prediction:

• EdenGNN-Uni_vasp_pseudo.ckpt: pseudo-charge-density model.
• EdenGNN-Uni_vasp_aug.ckpt: augmentation-occupancy model.
• config_vasp_density.yaml: reference configuration for the pseudo-density model.
• config_vasp_aug.yaml: reference configuration for the augmentation model.

These models correspond to the universal VASP workflow described in the paper.

mp/abacus/

ABACUS-based universal checkpoint and reference config:

• EdenGNN-Uni_abacus.ckpt: universal ABACUS model checkpoint.
• config_abacus.yaml: reference configuration for the ABACUS workflow.

This checkpoint corresponds to the ABACUS dataset used for training the general model with the sg15_oncv pseudopotential family and the associated numerical atomic orbital basis.

md/

System-specific checkpoints for the eight molecular-dynamics benchmarks described in the paper:

• si.ckpt
• gaas.ckpt
• insb.ckpt
• inas.ckpt
• gan.ckpt
• lif.ckpt
• al2o3.ckpt
• al.ckpt
• config.yaml: reference configuration for the MD benchmarks.

mos2/

Checkpoint and reference config for the twisted bilayer MoS2 benchmark:

• mos2.ckpt
• config.yaml

This model is associated with the training/validation data generated from randomly perturbed twisted bilayer MoS2 structures.

Model scope

The checkpoints in this repository are intended for the EdenGNN workflows described in the paper and in the main code repository. Their behavior depends on using compatible data preparation and DFT settings.

For example, the following should remain consistent with the original workflow whenever applicable:

• DFT software backend (VASP, ABACUS as configured in the main project).
• Pseudopotential choices and, for ABACUS, basis-set choices.
• Charge-density grid settings and related cutoff parameters.
• Input file templates and directory layout expected by the EdenGNN pipeline.
• Task type (density, augmentation, or both).

For VASP-based PAW models in particular, reproducibility depends on using pseudopotentials consistent with those used during dataset generation and training.

How to use these checkpoints

Use these checkpoints together with the main EdenGNN repository rather than as standalone files.

A typical workflow is:

1. Clone and install the main code repository.
2. Prepare your DFT environment and input templates.
3. Start from the reference config.yaml in this repository.
4. Modify the configuration to match your local environment.
5. Set the run.checkpoint field to the desired checkpoint path.
6. Run training fine-tuning or prediction from the main EdenGNN codebase.

Please consult the upstream project for command-line usage, expected filelists, prediction workflow, and software patch requirements:

• https://github.com/rubenlee11/EdenGNN

Related data repository

The dataset split files, pseudopotential metadata, and selected reproduction examples associated with these checkpoints are stored separately in the companion dataset repository:

• TrueSavage/EdenGNN-Data

Citation

If you use these checkpoints, please cite the associated EdenGNN paper.