Electronic Entropy Training and Reference Data for Machine Learning Interatomic Potentials

A 303.3 MB dataset in XYZ format organized into three subsets for developing electronic-entropy-embedded Machine Learning Interatomic Potentials (MLIPs). It was created by Steen Lysgaard and last updated in June 2026. The data is derived from polyanionic sodium cathode materials and includes structures for training, benchmarking, and analyzing electronic entropy contributions.

Use Cases

• Training electronic-entropy-embedded MLIPs based on the provided 'Training_dataset'.
• Benchmarking MLIP performance on structures from MACE-driven Genetic Algorithm simulations using the 'Reference_dataset'.
• Quantifying electronic entropy contributions in DFT and MLIP structural optimizations using the 'Electronic_entropy_reference_dataset'.
• Analyzing energy distributions associated with charge ordering in Na2/3FePO4 phases.
• Evaluating MLIPs' ability to reproduce correct energy ordering and structural stability of olivine NaFePO4 phases.

Strengths

• Dataset is structured into three distinct, purpose-built subsets (Training, Reference, Electronic Entropy Reference) for clear workflow stages.
• Includes both pre-optimized and DFT-optimized structures for direct comparison.
• Training data is derived from a cited, publicly available dataset of polyanionic sodium cathode materials.
• Reference dataset contains specific configurations (three lowest- and three highest-energy) selected from MLIP-driven GA simulations.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data is focused exclusively on iron-containing polyanionic sodium cathode materials, limiting generalizability.

Provenance

Source

figshare

Collection Method

Derived from the 'polyanionic sodium cathode materials dataset'; structures obtained from MACE-driven Genetic Algorithm simulations and DFT optimizations.

Freshness

Last updated 2026-06-02 11:26:34; freshness should be verified.