Machine learning a bond order potential model to study thermal transport in WSe₂ nanostructures

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Materials Science

Background:

• Transition metal di-chalcogenides (TMDCs) possess unique thermal, chemical, and electronic properties crucial for applications like thermoelectrics and nanoelectronics.
• Accurate modeling of thermal transport in TMDC nanostructures is challenging due to limitations in first-principles calculations and classical molecular models.
• Existing models struggle with multi-layered TMDCs and dynamic processes like nucleation and growth.

Purpose of the Study:

• To overcome limitations in modeling TMDC nanostructures by developing a machine learning-based bond order potential (ML-BOP).
• To accurately capture the structure, dynamics, and thermal transport properties of TMDCs, using WSe2 as a model system.
• To enable advanced simulations of nucleation, growth, and transport phenomena in low-dimensional TMDC structures.

Main Methods:

• Developed a bond order potential (BOP) trained against first-principles data using a hierarchical objective genetic algorithm workflow.
• Employed molecular dynamics simulations with the ML-BOP model to investigate WSe2 nanotubes and nanoribbons.
• Analyzed temperature-dependent thermal conductivity and structural properties across different chiralities.

Main Results:

• The ML-BOP model successfully captures the energetics, thermal, and mechanical properties of free-standing WSe2 sheets.
• Observed higher thermal conductivity in armchair WSe2 monolayers compared to zigzag, with the opposite trend in smaller diameter nanotubes.
• Attributed thermal conductivity differences to anisotropy and phonon-phonon scattering mechanisms.

Conclusions:

• The developed ML-BOP model provides a robust tool for simulating TMDC nanostructures, overcoming previous computational limitations.
• The findings offer insights into the anisotropic thermal transport behavior of WSe2 nanotubes and ribbons.
• This model facilitates future research on low-dimensional WSe2 structures for thermoelectric and thermal management applications.