Predictions of Boron Phase Stability Using an Efficient Bayesian Machine Learning Interatomic Potential.

Area of Science:

• Materials Science
• Computational Chemistry
• Condensed Matter Physics

Background:

• Elemental boron exhibits diverse allotropes (α-B, β-B, γ-B) with distinct thermodynamic properties.
• Accurate prediction of phase stability requires reliable interatomic potentials for molecular dynamics (MD) simulations.
• Understanding boron allotrope stability is crucial for materials design and applications.

Purpose of the Study:

• To investigate the thermodynamic phase stability of α-B, β-B, and γ-B.
• To develop and validate a Bayesian interatomic potential for simulating boron allotropes.
• To explore the role of defects and entropy in boron phase transitions.

Main Methods:

• Utilized a Bayesian interatomic potential trained via sparse Gaussian process (SGP) machine learning.
• Employed SGP potentials within molecular dynamics (MD) simulations for property prediction.
• Generated simulation data using on-the-fly active learning for quantum mechanical accuracy.

Main Results:

• Simulated phase diagram (500-1400 K, 0-16 GPa) shows excellent agreement with experimental data.
• Identified the B13 defect as critical for stabilizing β-B below 700 K.
• Demonstrated that entropy dominates phase stability over α-B at higher temperatures.

Conclusions:

• SGP potentials provide quantum mechanical accuracy for predicting thermodynamic, structural, and vibrational properties of boron.
• The study successfully predicted defect-related phenomena using potentials trained on defect-free systems.
• This approach enables accurate investigation of crystal phase stability and transitions in solid-state materials.