Machine learning-based high-throughput screening of Mg-containing alloys for hydrogen storage and energy conversion applications

Abstract

The development of novel materials for hydrogen storage and conversion applications is expected to facilitate the transition to clean energy. In particular, near-ambient hydrogen storage, thermal energy storage, and lithium conversion electrodes are selected in this study as the applications for which the development of novel Mg-containing materials is of great importance. We utilize a machine learning model, based on the graph neural network, developed for predicting hydride formation enthalpy in intermetallic compounds, to perform high-throughput screening based on the atomic composition and crystal structure of the starting intermetallic compounds. Trends and structure-property relations are discussed, as well as the possibilities for tailoring the stability of Mg-containing hydrides by alloying. For 636 compounds identified as stable by DFT calculations, we predict hydride formation enthalpy and equilibrium potential of metal hydride conversion electrode for Li-ion batteries. Based on the predicted enthalpy of hydride formation, 32 intermetallics are identified as suitable for near-ambient hydrogen storage applications. Among them, MgBe13, seen as a promising material to achieve a high gravimetric density of hydrogen, is additionally studied using DFT. Further investigation of the Na-Mg-Al alloys is proposed as a good route in the search for new thermal energy storage materials. Binary Mg-containing intermetallics are discussed as conversion-type negative electrodes in Li-ion batteries.