Machine learning based prediction of lattice thermal conductivity for half-Heusler compounds using atomic information

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Materials Science

Background:

• Half-Heusler compounds are promising for thermoelectric energy conversion and spintronics.
• High lattice thermal conductivity in these materials, due to their crystal symmetry, poses a challenge for device thermal management.
• Accurate prediction of lattice thermal conductivity is crucial for effective thermal management.

Purpose of the Study:

• To investigate the feasibility of predicting lattice thermal conductivity in half-Heusler compounds using machine learning.
• To develop a model based solely on atomic information for predicting thermal conductivity.
• To enable rapid assessment of new half-Heusler materials for potential applications.

Main Methods:

• Utilized density functional theory to calculate lattice thermal conductivity for various half-Heusler compounds.
• Developed a machine learning model trained on atomic properties (atomic radius, atomic mass) of constituent elements.
• Validated the model's predictive accuracy using computational data.

Main Results:

• A machine learning model was successfully constructed to predict lattice thermal conductivity with high accuracy.
• The model relies exclusively on the atomic radius and atomic mass of elements at each site within the crystal structure.
• The prediction is applicable to both known and unknown half-Heusler compounds.

Conclusions:

• Machine learning provides a highly accurate method for predicting lattice thermal conductivity in half-Heusler compounds.
• The developed model simplifies the prediction process by using readily available atomic information.
• This approach facilitates low-cost, rapid development of novel functional materials for energy and electronics applications.