Learning Piezoelectric Tensors via Strain-Conditioned Polarization Clusters

A physics-informed cluster graph neural network (PCGNN) model for predicting piezoelectric tensors, developed by Chunlin Yu and last updated in May 2026. The model applies controlled strain perturbations to crystal structures and reconstructs macroscopic tensors through symmetry-consistent aggregation. On the Materials Project test set, PCGNN achieved a mean absolute error of 0.135 C/m², outperforming baseline models.

Use Cases

• Predicting piezoelectric tensor entries for new materials based on the described strain–polarization response pathway.
• Screening unlabeled materials for strong piezoelectric response as mentioned in the description.
• Benchmarking new tensor prediction models against the reported PCGNN performance on the Materials Project test set.
• Studying symmetry-imposed sparsity patterns in material tensors using the described capsule transformer method.

Strengths

• Model performance is quantified with a specific mean absolute error of 0.135 C/m² on the Materials Project test set.
• The dataset includes a direct comparison, showing the model outperformed EATGNN and CGCNN by 32.5% and 52.6% respectively.
• The associated file size is precisely documented as 365.4 KB.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is very small at 365.4 KB, indicating limited scope.

Provenance

Source

figshare, author Chunlin Yu.

Collection Method

Likely contains model outputs or training data related to the described physics-informed cluster graph neural network (PCGNN).

Freshness

Last updated 2026-05-08 15:36:17; freshness should be verified.