学术文献库

The predictive performance screening of novel compounds can significantly promote the discovery of efficient, cheap, and non-toxic thermoelectric materials. Large efforts to implement machine-learning techniques coupled to materials databases are currently being undertaken, but the adopted computational methods can dramatically affect the outcome. With regards to electronic transport and power factor calculations, the most widely adopted and computationally efficient method, is the constant relaxation time approximation (CRT). This work goes beyond the CRT and adopts the proper, full energy and momentum dependencies of electron-phonon and ionized impurity scattering, to compute the electronic transport and perform power factor optimization for a group of half-Heusler alloys. Then the material parameters that determine the optimal power factor based on this more advanced treatment are identified. This enables the development of a set of significantly improved descriptors that can be used in materials screening studies, and which offer deeper insights into the underlying nature of high performance thermoelectric materials. We have identified $n_v$$ε_r$ / $D_o^2m_{cond}$ as the most useful and generic descriptor, a combination of the number of valleys, the dielectric constant, the conductivity effective mass, and the deformation potential for the dominant electron-phonon process. The proposed descriptors can accelerate the discovery of new efficient and environment friendly thermoelectric materials in a much more accurate and reliable manner, and some predictions for very high performance materials are presented.