Overview

PRISM is a graph neural network framework for crystal property prediction that explicitly encodes unit‐cell geometry and periodic boundary conditions through four specialized expert modules: an atomistic radius graph for short‐range interactions, a cell‐space graph with a superatom node to capture global lattice periodicity, a feature‐space similarity graph enforcing invariance across equivalent cell representations, and a multiscale bipartite graph linking atomic and global embeddings. These expert outputs are combined through learnable gating mechanisms that dynamically balance and integrate their complementary insights (mixing local chemistry, global structure, and feature-space correlations) into unified representations for accurate material property prediction.

🌐 Project | 📃 Paper

Table of Contents

• Installation
• Dependencies
• Dataset
• Training
• Results
• Known Issues
• License

Installation

Instructions to set up the environment:

# Create a Conda environment
conda env create -f environment.yml

# Activate the environment
conda activate PRISM

Datasets

Jarvis

For tasks derived from Jarvis dataset, we followed the methodology of Choudhary et al. in ALIGNN, utilizing the same training, validation, and test datasets. The dataset is automatically downloaded and processed by the code.

The Materials Project

For tasks derived from The Materials Project, we followed the methodology of Yan et al. in Matformer, utilizing the same training, validation, and test datasets. The dataset is automatically downloaded and processed by the code, except for the bulk and shear modulus that are publicly available at Figshare.

Matbench

For tasks derived from the Matbench dataset, we followed the proposed methodology from the Matbench library. The dataset is automatically downloaded and processed by the code.

Training

To recreate the experiments from the paper

Jarvis:

cd scripts/
bash train_cartnet_jarvis.sh

The Materials Project

cd scripts/
bash train_cartnet_megnet.sh

Evaluation

Instructions to evaluate the model:

python main.py --inference --checkpoint_path path/to/checkpoint.pth

Results

Jarvis Dataset

Results on Jarvis Dataset:

Method	Form. Energy (meV/atom) ↓	Band Gap (OPT) (meV) ↓	Total energy (meV/atom) ↓	Band Gap (MBJ) (meV) ↓	Ehull (meV) ↓
Matformer	32.5	137	35	300	64
PotNet	29.4	127	32	270	55
eComformer	28.4	124	32	280	44
iComformer	27.2	122	28.8	260	47
CartNet	27.05 ± 0.07	115.31 ± 3.36	26.58 ± 0.28	253.03 ± 5.20	43.90 ± 0.36
PRISM	25.87 ± 0.36	109.26 ± 2.546	26.34 ± 0.38	236.49 ± 5.56	23.07 ± 0.62

(best result in bold and second best in italic)

The Materials Project

Method	Form. Energy (meV/atom) ↓	Band Gap (meV) ↓	Bulk Moduli (log(GPa)) ↓	Shear Moduli (log(GPa)) ↓
Matformer	21	211	0.043	0.073
PotNet	18.8	204	0.040	0.065
eComformer	18.16	202	0.0417	0.0729
iComformer	18.26	193	0.038	0.0637
CartNet	17.47 ± 0.38	190.79 ± 3.14	0.033 ± 0.00094	0.0637 ± 0.0008
PRISM	16.59 ± 0.1	179.71 ± 1.58	0.033 ± 0.00094	0.0655 ± 0.0008

(best result in bold and second best in italic)

Matbench Dataset

Method	e_form MAE (meV) ↓	e_form RMSE (meV) ↓	jdft2d MAE (GPa) ↓	jdft2d RMSE (GPa) ↓
MODNet	44.8 ± 3.9	88.8 ± 7.5	33.2 ± 7.3	96.7 ± 40.4
ALIGNN	21.5 ± 0.5	55.4 ± 5.5	43.4 ± 8.9	117.4 ± 42.9
coGN	17.0 ± 0.3	48.3 ± 5.9	37.2 ± 13.7	101.2 ± 55.0
M3GNet	19.5 ± 0.2	–	50.1 ± 11.9	–
eComFormer	16.5 ± 0.3	45.4 ± 4.7	37.8 ± 9.0	102.2 ± 46.4
iComFormer	16.5 ± 0.3	43.8 ± 3.7	34.8 ± 9.9	96.1 ± 46.3
PRISM	15.20 ± 0.31	30.43 ± 1.38	38.41 ± 12.44	97.90 ± 38.25

(best result in bold and second best in italic)

Known Issues

Due to the presence of certain non-deterministic operations in PyTorch, as discussed here, some results may not be exactly reproducible and may exhibit slight variations. This variability can also arise when using different GPU models for training and testing the network.

License

This project is licensed under the MIT License - see the LICENSE file for details.