Theoretical hydrogen storage properties of high entropy alloys: A combined DFT and machine learning approach

Abstract

High entropy alloys (HEAs) have gained attention for solid-state hydrogen storage due to their unique properties, including lattice distortion and the cocktail effect. This study applied machine learning techniques in conjunction with density functional theory (DFT) study to analyse and predict the hydrogen storage properties of HEAs from a dataset comprising 304 entries (2006 – 2024). DFT methods through Dmol 3 module was used to mainly compare and highlight any hydrogenation differences between traditional alloys and I entropy alloys, which included AB, AB 2 and AB 5 classes of alloys. In these arrangements, the A site is usually occupied by elements that easily donate electrons and form hydride-stabilizing sites while the B site elements bond with A elements to form stable intermetallic structures and often contribute to hydrogen dissociation and electronic conductivity. DFT study revealed that while the base alloys exhibit metallic characteristics conducive to hydrogen uptake, hydrogen absorption in base alloys leads to reduced density of states at the Fermi level, indicative of electron localization and the formation of pseudogaps or narrow band gaps. On the other hand, high entropy alloys exhibit more prominent electronic changes and stronger hydrogen binding, as evidenced by larger band gaps and more negative hydrogenation enthalpies, despite almost similar unit cell volumes compared to their traditional counterparts. This indicates that hydrogen storage behaviour is governed not just by lattice size or structure type, but also by the local bonding environment and electronic configuration induced by multi-elemental interactions. Interestingly, AB alloys transform from cubic structure to orthorhombic structure upon hydrogenation, suggesting that the hydrogen-induced strain is not uniform along all crystallographic axes, resulting in lattice distortion that breaks cubic symmetry. The hexagonal structures of both AB 2 and AB 5 alloys remain intact after hydrogen uptake, indicating better hydrogen storage capabilities than AB alloys. For machine learning analysis, three models linear regression, decision tree, and random forest were developed to predict hydrogen storage capacity ( H/M ratio), Gibbs free energy of hydrogen absorption ( ∆G H ) at 300°C, and the equilibrium temperature ( T H ) . Among these, the random forest model demonstrated superior accuracy, with R² values of 0.927, 0.908, and 0.950 for H/M ratio, ∆ G H , and T H , respectively. Validation with experimental data confirmed the model's reliability. Feature importance analysis using SHAP local explanation summary tool revealed that the valence electron concentration (VEC) is a key factor influencing hydrogen storage properties. This study highlights the potential of combining DFT study and machine learning models, particularly the random forest approach, in efficiently predicting and optimizing the hydrogen storage properties of HEAs. It uniquely combines DFT and machine learning to predict 3 key parameters; i.e. H/M ratio, equilibrium temperature and Gibbs free energy of hydrogen absorption. These findings provide a valuable tool for guiding the design and selection of novel HEA-based materials for hydrogen storage applications.

Topics & Concepts