Machine learning-guided design of high-performance Mg-based thermoelectrics: insights into thermal expansion effects

Mg-based thermoelectric materials are becoming ideal candidates for thermoelectric applications, owing to their eco-friendliness and abundant availability. To overcome the limitations of conventional experimental methods and accelerate the development of high-performance thermoelectric materials, this study leverages high-throughput computing and machine learning to perform a comprehensive and systematic evaluation of a vast array of Mg-based thermoelectric materials. Our findings highlight the pivotal role of thermal expansion in modulating the thermoelectric figure of merit (ZT) in Mg-based systems. Specifically, thermal expansion alters the interatomic interaction potential, enhancing material anharmonicity and significantly reducing lattice thermal conductivity. Furthermore, thermal expansion reduces energy band dispersion, leading to a more concentrated density of states near the Fermi level. This effect increases effective mass, thereby potentially boosting the Seebeck coefficient. These insights not only deepen the understanding of the physical mechanisms by which thermal expansion influences thermoelectric performance but also establish a universal theoretical framework for optimizing high-performance thermoelectric materials. To accelerate the discovery and application of Mg-based thermoelectric materials, we have developed an XGBoost model with high predictive accuracy and robust generalization performance. This model enables the precise prediction of thermoelectric properties, providing a tool for rapid screening and optimization of Mg-based thermoelectric materials.