A Deep Neural Network Potential Study on the Mechanical and Thermophysical Properties of β

A neural network potential (NNP) predicts the mechanical and thermophysical properties of the high-energy-density material β-HMX with density functional theory-level accuracy. The model, developed by Jiyuan Wei and last updated in May 2026, achieves orders-of-magnitude computational efficiency improvements. Its predictions for bulk modulus, thermal expansion coefficient, and heat capacities show deviations from experiments within 13% and 1%, respectively.

Use Cases

• Predicting bulk modulus and mechanical properties of β-HMX under pressure based on the described neural network potential.
• Mapping the coupled pressure-temperature dependence of thermophysical properties like thermal expansion coefficient.
• Simulating constant-pressure and constant-volume heat capacities (Cp and Cv) for molecular crystals with DFT-level accuracy.
• Studying phonon mode suppression and stiffening in high-energy-density materials under extreme conditions.

Strengths

• Model predictions for volumetric thermal expansion coefficient and heat capacities show deviations below 1% from experiments.
• Bulk modulus prediction at zero pressure agrees with experimental measurements within 13%, a noted improvement over classical molecular dynamics.
• The active-learning framework automatically eliminates structurally redundant configurations during NNP development.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data is specific to the β-HMX molecular crystal; generalizability to other materials is not described.

Provenance

Source

Jiyuan Wei via figshare

Collection Method

Developed through an active-learning framework for neural network potentials, validated against density functional theory and experimental data.

Freshness

Last updated 2026-05-12 12:19:30; freshness should be verified.