Discovery of new clathrates by computational screening, and directed synthesis

Charge-balanced inorganic clathrates are semiconducting materials with unique properties, such as ultra-low thermal conductivity, superconductivity, and high-density ion storage. They consist of covalent frames of nanometer-size polyhedral cages encapsulating metallic guest atoms. While the frame of conventional clathrates mostly consists of group IV elements (Si, Ge, Sn), we have recently discovered new stable intermetallic clathrates with III-V frames and enticing transport properties.

In this work, we combine Zintl compositional predictions, high-throughput density functional theory calculations, and machine learning to screen several thousands of compositions of III-V and II-VI type-I clathrate frames, mixed with transition metals, to identify stability trends and families of potentially stable compounds. In particular, a novel workflow is developed, which uses transfer learning to refine convolutional graph neural networks for predicting the site ordering of clathrate superstructures, as well as their equilibrium density and decomposition energy.

Theoretical predictions are tested by ex-situ synthesis and crystal growth, which have revealed new clathrates and clathrate-like compounds. Single-phase synthesis and first-principles calculations enable the joint theoretical and experimental exploration of their transport properties, revealing promising properties for thermoelectric energy conversion.