SIPFENN: Formation Energy Predictions for Voronoi-Tessellated Materials

A neural network-based tool for predicting formation energies of materials based on elemental and structural features. The models were trained on data from the Open Quantum Materials Database (OQMD) and achieve mean average errors between 28 and 42 meV/atom on a test set of 21,800 compounds. The work was authored by Adam M. Krajewski of Pennsylvania State University.

Use Cases

• Predict formation energies for new materials based on Voronoi-tessellated structural features.
• Tune models for specific material systems by incorporating new data to improve generalization.
• Compare model performance for highest accuracy versus low computational cost in materials screening.

Strengths

• Three final models were developed, each optimized for a specific goal: highest test accuracy, performance on new materials discovery, and low computational cost.
• The models achieve a mean average error as low as 28 meV/atom on a test set of 21,800 compounds.
• The low-cost model reduces computational cost by a factor of 8 compared to other architectures.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Open Quantum Materials Database (OQMD)

Collection Method

Data likely gathered from computational materials databases and used to train neural network models.