AI-assisted discovery of the ‘self-optimizing’ mechanism in magnesium-based thermoelectric materials

Magnesium-based thermoelectric materials, prized for their environmental friendliness and abundant resources, have long been a popular choice for waste heat recovery and solid-state cooling. However, exploring the vast chemical structure space to find high-performance materials has traditionally relied on trial-and-error, which is slow and inefficient. Researchers at Beihang University have now developed a systematic solution combining high-throughput calculations with machine learning.

The key discovery is that thermal expansion plays a crucial role in enhancing thermoelectric performance. As crystals heat up and expand, atomic distances increase, enhancing lattice anharmonicity and reducing lattice thermal conductivity. At the same time, the electronic band structure becomes more concentrated, increasing the effective mass of charge carriers and boosting the Seebeck coefficient. Together, these effects improve the thermoelectric performance metric, ZT.

The team selected magnesium-based crystal structures from the open quantum materials database, screened them for stability, and performed density functional theory calculations to build a large-scale dataset. They then evaluated five machine learning models, ultimately adopting the XGBoost model for high-accuracy predictions and rapid screening, providing a powerful tool for designing next-generation magnesium-based thermoelectric materials.

This study not only reveals the general physical mechanism by which thermal expansion regulates thermoelectric performance but also offers a quantitative strategy for material optimization. The work was published in Science Bulletin and provides new insights for magnesium-based thermoelectrics as well as broader thermoelectric systems.

---

---