Gaussian Approximation Potentials for Iron from First-Principles Molecular Dynamics

Daniele Dragoni from École Polytechnique Fédérale de Lausanne created a Gaussian Approximation Potential model for the α-phase of iron. The model was trained on energies, stresses, and forces derived from first-principles molecular dynamics simulations of pristine and defected bulk systems, surfaces, and γ-surfaces. This dataset provides a machine-learned interatomic potential designed to describe complex iron systems more efficiently than direct first-principles calculations.

Use Cases

• Training machine learning models for interatomic potentials based on first-principles simulation data.
• Simulating the behavior of iron in bulk, surface, and defected configurations using a Gaussian-process-based potential.
• Benchmarking the transferability of Gaussian Approximation Potentials against traditional functional-form potentials.
• Studying material properties of the α-phase of iron without performing costly first-principles calculations for each simulation.

Strengths

• Model is based on first-principles molecular dynamics simulations, providing a high-fidelity training foundation.
• Training data includes energies, stresses, and forces for multiple system types (pristine bulk, defected bulk, surfaces, γ-surfaces).
• The Gaussian Approximation Potential method is a novel, non-parametric approach using Gaussian-process regression.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and dataset size are unknown, which may limit suitability assessment.
• The model is specifically for the α-phase of iron, potentially limiting direct application to other phases or materials.

Provenance

Source

Daniele Dragoni, École Polytechnique Fédérale de Lausanne

Collection Method

Generated from first-principles molecular dynamics simulations.