Machine-Learning Modeling of Elemental Ferroelectric Bismuth Monolayer.

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Chemistry

Background:

• Bismuth monolayer is a novel 2D single-element ferroelectric system.
• Understanding its properties requires accurate computational models.

Purpose of the Study:

• To model the potential energy surface of a bismuth monolayer.
• To investigate temperature-dependent phase transitions and ferroelectric domains.
• To analyze the impact of substrates on monolayer properties.

Main Methods:

• Message-passing neural network (MPNN) for potential energy surface modeling.
• High-accuracy, large-scale atomistic simulations.
• Analysis of free-standing vs. substrate-constrained systems.

Main Results:

• MPNN model achieved <1.2 meV/atom accuracy.
• Simulations revealed temperature-dependent phase transitions.
• Ferroelectric domains and lattice thermal conductivity were observed.

Conclusions:

• Accurate ML model enables large-scale simulations of bismuth monolayers.
• Substrate interactions influence phase transitions.
• Insights into ferroelectric domain behavior and thermal properties were gained.