Researchers at Tohoku University and the University of Tokyo launched the Digital Hydrogen Platform, a curated database with over 30,000 data entries from more than 4,000 publications, to map the physical trade-offs between hydrogen storage capacity and equilibrium pressure.
Clean energy development relies on finding ways to store and release hydrogen safely and efficiently as a promising alternative energy source. Yet, researchers face a persistent challenge in balancing how much hydrogen a material can hold against the practical operating conditions required to get it back out.
A research team led by Tohoku University, alongside collaborators at the University of Tokyo, developed the Digital Hydrogen Platform, known as DigHyd and publicly available online.
Building the DigHyd Repository Through AI Mining and Human Curation
Assembling a reliable open-access repository required more than automated web scraping. According to research details, the team combined AI-assisted literature mining with manual validation, checking extracted information against original publications.
For metal hydrides, which store hydrogen within their internal atomic structures, researchers manually curated 4,643 storage-capacity measurements. They added 652 pairs of enthalpy and entropy measurements. These thermodynamic quantities allow scientists to predict how storage behavior changes under varying temperatures, while direct links to the original publications enable thorough data tracing and verification.
To interpret this vast repository, the team deployed an artificial intelligence method called symbolic regression. They specifically used an algorithm named GoodRegressor.
In rigorous five-fold cross-validation tests, the method delivered predictive performance that matched leading black-box models. It also generated clear mathematical equations that scientists could readily read and analyze.
Check out the #AppliedPhysicsA cover figure used for the October cover: Digital hydrogen platform (DigHyd): a rigorously curated database for hydrogen storage materials empowered by AI-assisted literature mining by Seong-Hoon Jang, Di Zhang, et al. https://t.co/ehwP2CnPqG pic.twitter.com/9sXkoFZcUh
Understanding hydrogen storage requires more than finding materials that can hold large amounts of hydrogen. We need to understand why their properties determine how hydrogen is retained and released. By combining reliable experimental data with interpretable AI, we can begin to uncover these underlying physical relationships.
Hao Li
Exposing the Capacity-Pressure Trade-Off in Metal Hydrides
The resulting physical map exposed an interconnected hierarchy of material characteristics that dictate both storage capacity and equilibrium pressure. Lighter host atoms generally favor higher storage capacity by weight.
When hydrogen enters a metal lattice, the physical structure must deform and expand. Stiffer materials resist this deformation, raising the elastic energy needed to house the hydrogen.
That resistance can destabilize the hydride and push its equilibrium pressure higher. By connecting electronegativity, thermal conductivity, and resistance to compression, the team clarified why boosting storage capacity often alters the conditions required for hydrogen release in practical applications.
Material Designers Must Evaluate Storage Capacity with Equilibrium Pressure
The findings show that material designers must evaluate hydrogen storage capacity and equilibrium pressure together rather than treating individual properties in isolation when engineering new solutions.
The important next step is to use these insights to guide the development of new materials and test whether their predicted properties hold up in experiments. By connecting data-driven analysis with the synthesis and evaluation of candidate materials, we hope to accelerate the search for practical hydrogen storage solutions.
Shin-ichi Orimo
The study was published online in Applied Physics A on September 4, 2026. The research group intends to expand the database, pinpoint promising material compositions, and synthesize physical candidates for laboratory testing to refine their predictive models.
