A High Throughput DFT Study of Half Heusler Solid Solution Mixtures

Half heuslers exhibit properties well-suited for thermoelectrics. The best performing half heusler thermoelectrics such as TiNiSn, NbFeSb, and ZrCoBi-based systems are solid-solution alloys with disorder scattering which reduces thermal conductivity. However, only a handful of such alloy systems have been reported. Furthermore, these systems are studied with mixing on just one of the three sub-lattices at a time. Hence, a comprehensive study of solubility in mixed alloy systems can guide future choices of half heusler thermoelectrics. Here, we use DFT to calculate mixing energies and relative stabilities of around 1000 binary solid solutions using restrictions of isovalent substitution and single sublattice mixing. Half of these have mixing energies suitable for forming single phase alloys, when |E_mix| < 0.05 eV/atom (kBT @ 600K). We then extend our search to higher component mixtures and the unexplored concept of multi-sublattice mixing. Continuing with DFT is too slow given the combinatorically expanding composition space, so we develop a machine learning model to help classify mixing behavior. With trends identified from the model and DFT, we hope to better guide future experiments toward improved half heusler thermoelectrics.