Understanding Hydrogen Diffusion with AI: A Step Forward for Sustainable Energy

Hydrogen has long been considered a promising clean energy source, but one of the main challenges to its large-scale deployment is how to store it effectively. Among the various solutions under consideration, one of the most promising involves the use of magnesium to form magnesium hydride. However, many open questions remain about how hydrogen behaves within this material.

To address this issue, an international team of researchers has developed a machine learning-based system capable of accurately reconstructing the kinetic properties of hydrogen in magnesium across a range of temperatures. The results were published in the journalnpj Computational Materials.

“The model we developed can accurately predict how hydrogen behaves at different temperatures, allowing us to shed light on aspects of these dynamics that are still poorly understood,” explains Cesare Franchini, professor at the Department of Physics and Astronomy “Augusto Righi” of the University of Bologna, who coordinated the study. “This achievement also highlights the remarkable potential of machine learning systems in materials science, particularly when it comes to exploring complex dynamics that have traditionally been very difficult to study using experimental methods alone.”

Despite being a clean, abundant alternative to fossil fuels with zero CO₂ emissions, hydrogen still lacks a sustainable, efficient, and safe system for storage and transport. Magnesium could offer a viable solution: magnesium hydride is a highly stable compound that can be stored safely. The main obstacle lies in our limited understanding of how hydrogen behaves inside solid materials like magnesium, making it difficult to accurately predict its dynamics.

To overcome this limitation, the researchers developed an active learning system trained in real time using a series of simulations of magnesium hydride behavior under different conditions and temperatures. The outcome was a neural network-based model capable of striking an exceptional balance between computational efficiency and predictive accuracy.

“This represents a major step forward, not only in understanding hydrogen dynamics in magnesium, but also in paving the way for future storage solutions involving other types and combinations of materials,” adds Franchini. “Machine learning systems like this — capable of precisely modeling hydrogen behavior — can make a crucial contribution to the green transition and help shape a future powered by sustainable energy.”

The research was publish in npj Computational Materials, under the title “Hydrogen diffusion in magnesium using machine learning potentials: a comparative study”. Contributors from the University of Bologna include Cesare Franchini, Luca Leoni, and Luca Pasquini from the Department of Physics and Astronomy “Augusto Righi”.