Ductile Mg-Te-Pb thermoelectric materials with ultralow lattice thermal conductivity predicted by a deep learning potential model

Inorganic thermoelectric materials with excellent thermoelectric performance are often brittle. In this work, we introduce functional units into deep learning molecular dynamics-accelerated crystal structure prediction to develop novel functional materials exhibiting both outstanding thermoelectric and mechanical properties. We selected the elements Mg, Te, Pb, and Bi, which are prone to form materials with superior thermoelectric and mechanical performance. A deep learning potential was iteratively trained to accelerate the energy evaluation during structure prediction. Subsequently, the [XY6] functional units were introduced to guide the discovery of novel crystals. Following stability screening, we identified three novel materials that exhibit both outstanding thermoelectric and mechanical performance: I-Mg2Te3Pb, II-Mg2Te3Pb, and MgTe2Pb. I-Mg2Te3Pb possesses an ultralow lattice thermal conductivity of 0.065 W/(m·K), while II-Mg2Te3Pb achieves a remarkably high Seebeck coefficient of 767 μV/K. All three materials can withstand shear strains exceeding 60% without significant structural failure. The enhancement in thermoelectric performance is attributed to the [PbTe6] units significantly suppressing lattice thermal conductivity, while the improvement in mechanical properties results from the ordered arrangement of [PbTe6] and [MgTe6] units and the “catch-bond” mechanism, which introduces additional potential slip systems. These findings propose a new pathway for developing materials that integrate excellent thermoelectric and mechanical properties.