Chemistry Weekly Seminar - Dr. Arthur Mar, University of Alberta

Machine-Learning Predictions and Experimental Validation of Heusler Structures

Given an arbitrary combination of metals, what crystal structure will be adopted by the resulting compound?  This question has been posed since the early days of Pauling’s efforts to relate structure and bonding.  The large family of intermetallic compounds collectively known as Heusler compounds are important in applications such as thermoelectric materials, ferromagnets, magnetocaloric materials, and catalysts.  They span several types of idealized compositions and structures:  half-Heusler (MgAgAs-type), full- and inverse Heusler (Cu₂MnAl- and Li₂AgSb-type), and quaternary Heusler (LiMgPdSn-type).  In reality, disorder phenomena are common, and preparing solid solutions between half- and full-Heusler phases serves as a powerful tool to control physical properties.  Machine-learning models have been developed to predict new compounds that are likely to adopt full- and half-Heusler structures and to evaluate their site preferences.  These predictions were then tested by experimental methods to synthesize these compounds and determine their crystal structures.