Promising Multicomponent Cubic Alloys for Hydrogen Storage: Insights from First-Principles Calculations and Machine Learning

Abstract

An extensive exploration of the chemical space was conducted to design and identify promising multicomponent cubic alloys with appropriate enthalpy of reaction for hydrogen storage applications. We aim to identify alloys with suitable hydrogen absorption conditions for ambient conditions, according to favorable thermodynamic criteria, while addressing computational challenges in modeling large-scale systems. 18 elements were selected, leading to the systematic investigation of over 8000 quinary alloy systems across four distinct crystal phases (within solid–solution alloys, mono- and dihydrides). This effort resulted in a comprehensive data set of more than 34,000 equimolar MH_x structures, where M represents a combination of 5 elements chosen among the 18 selected atoms. To handle the computational demands of density functional theory (DFT) calculations on such a large scale of disordered supercells designed by the special quasirandom structure (SQS) method, a machine learning (ML) approach was introduced to accurately predict the enthalpy of hydride formation. By training the ML model on a strategically chosen subset of the data, high predictive accuracy was achieved while significantly reducing computational costs. By applying filtering parameters constrained by thermodynamic considerations, such as the value of plateau pressure or the presence of a single plateau, the integrated DFT-SQS-ML framework successfully identified 568 quinary alloy systems as ideal candidates for hydrogen storage. The findings establish a solid foundation for experimental validation and further advancements in the field of hydrogen storage materials.