Thermal conductivity of selenium crystals based on machine learning potentials.

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Chemistry

Background:

• Selenium's thermal transport properties are crucial for thermoelectric applications.
• Existing empirical potentials struggle to accurately model selenium's phonon transport.
• Understanding lattice dynamics is key to optimizing thermoelectric materials.

Purpose of the Study:

• Develop a high-precision machine learning potential (MLP) for selenium.
• Accurately simulate phonon transport and thermal properties.
• Investigate and compare thermoelectric potential of different selenium structures.

Main Methods:

• Generated training data using ab initio molecular dynamics simulations.
• Developed and validated a machine learning potential (MLP).
• Calculated phonon dispersion relations and thermal transport properties using the MLP.

Main Results:

• MLP accurately reproduced density functional theory results for phonon dispersion.
• Trigonal selenium (t-Se) showed higher thermal conductivity than monoclinic selenium (m-Se).
• t-Se's chain-like structure facilitates superior thermal transport compared to m-Se's ring-like structure.

Conclusions:

• t-Se exhibits superior thermoelectric potential due to a smaller bandgap and significantly higher electrical conductivity, despite higher lattice thermal conductivity.
• The developed MLP provides a reliable tool for studying selenium's thermal and electronic properties.
• This research advances the understanding of selenium for thermoelectric device applications.