Ab Initio-Based Bond Order Potential for Arsenene Polymorphs Developed via Hierarchical Reinforcement Learning

Area of Science:

• Materials Science
• Computational Physics
• Nanotechnology

Background:

• Arsenene, a 2D material, shows promise for electronics and quantum systems.
• Existing models struggle to capture the diverse properties of arsenene polymorphs.
• Accurate modeling is crucial for understanding arsenene's potential.

Purpose of the Study:

• To develop a novel computational model for arsenene.
• To accurately predict the mechanical and thermal properties of arsenene polymorphs and nanostructures.
• To advance the understanding of arsenene for technological applications.

Main Methods:

• Developed a bond-order potential model, ML-Tersoff, using multireward hierarchical reinforcement learning (RL).
• Trained the model on an ab initio dataset covering mechanical and thermal properties of arsenene polymorphs.
• Applied decision trees and hierarchical reward strategies for accelerated convergence in high-dimensional spaces.

Main Results:

• The ML-Tersoff model effectively captures properties of both buckled and puckered arsenene polymorphs without separate formalisms.
• An inverse relationship between critical strain and temperature in arsenene was observed.
• Thermal conductivity calculations for arsenene nanosheets align with ab initio data, showing decreased conductivity at higher temperatures due to anharmonic effects.

Conclusions:

• ML-Tersoff provides a unified and efficient approach to modeling arsenene's properties.
• The model's accuracy in predicting thermal conductivity supports its use in understanding heat transport in arsenene nanostructures.
• This work facilitates further research and development of arsenene-based devices.