Perspective: Web-based machine learning models for real-time screening of thermoelectric materials properties

• DOI: 10.1063/1.4952607
• Publication Date: 2016-05-27T19:01:00Z
• Authors: Michael W. Gaultois; Anton O. Oliynyk; Arthur Mar; Taylor D. Sparks; Gregory J. Mulhol...; Bryce Meredig
• Keywords: Characterization

The experimental search for new thermoelectric materials remains largely confined to a limited set of successful chemical and structural families, such as chalcogenides, skutterudites, Zintl phases. In principle, computational tools density functional theory (DFT) offer the possibility rationally guiding synthesis efforts toward very different chemistries. However, in practice, predicting properties from first principles challenging endeavor [J. Carrete et al., Phys. Rev. X 4, 011019 (2014)], researchers generally do not directly use computation drive their own efforts. To bridge this practical gap between needs tools, we report an open machine learning-based recommendation engine (http://thermoelectrics.citrination.com) that suggests promising compositions based on pre-screening about 25 000 known also evaluates feasibility user-designed compounds. We show can identify interesting chemistries thermoelectrics. Specifically, describe characterization one example compounds derived our engine, RE12Co5Bi (RE = Gd, Er), which exhibits surprising performance given its unprecedentedly high loading with metallic d f block elements warrants further investigation material platform. predicts family have low thermal electrical conductivities, but modest Seebeck coefficient, all are confirmed experimentally. note may simultaneously optimize three entering into zT; selected study due composition facile synthesis.