Hydrogen Storage in Medium and High-Entropy Alloys via Neural Network Potentials

The study of medium- and high-entropy alloys (MEAs and HEAs) demonstrates tunable local chemical environments and lattice distortions, making them promising for energy storage and catalysis applications. However, their vast compositional space and complex atomic interactions make the discovery of fundamental physical principles and design rules challenging. In this work, neural network potentials (NNPs) are developed to investigate hydrogen (H₂) adsorption ranging from binary to six-element compositions and sizes scaled in multiples of six atoms (18-48 atoms), sampling ~400 adsorption sites per alloy cluster. The adsorption energetics and structural parameters predicted by the NNPs are validated against density functional theory (DFT), enabling screening of candidate structures. The analysis of the resulting adsorption energy distribution, local coordination, and elemental configurations identifies optimal clusters for H₂ adsorption exhibiting free energies within a narrow window (–0.2 to –0.1 eV) that balance stability and desorption kinetics in accordance with Sabatier principles. These optimal clusters define active-site motifs and compositional trends, guiding the data-driven design of entropy-stabilized alloy clusters with tailored hydrogen interaction strengths.