Curated hydrogen database links storage capacity to release pressure

Hydrogen is often described as the fuel of a cleaner future, but the promise of a hydrogen economy rests on an unglamorous problem: finding solid materials that can hold large quantities of hydrogen and then give it back under practical conditions. A research team led by Tohoku University, working in collaboration with the University of Tokyo, has now taken a substantial step toward solving that problem with data rather than intuition. The group has built the Digital Hydrogen Platform, or DigHyd, a curated database containing more than 30,000 data entries drawn from over 4,000 experimental publications.

Using this resource, the researchers constructed a physical map that links measurable material properties to two quantities that matter enormously in hydrogen engineering: how much hydrogen a material can store, and the equilibrium pressure at which it absorbs and releases the gas. The study, published online in Applied Physics A on September 4, 2026, was led by Hao Li, Distinguished Professor at the Advanced Institute for Materials Research (WPI-AIMR) at Tohoku University, together with Shin-ichi Orimo of the same institute. According to the team, the central insight is that storage capacity and operating pressure should not be treated as separate design targets.

A material that packs in a lot of hydrogen by weight is not automatically useful if the thermodynamics demand extreme temperatures or pressures to release it. The new database and the interpretable machine learning analysis built on top of it are designed to make that trade-off visible, quantifiable and, ultimately, predictable. "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," says Li. That statement captures the methodological heart of the work.

Rather than training an opaque model to rank candidate materials, the team used a technique called symbolic regression, which searches for mathematical equations that describe relationships in data. The result is a set of formulas that scientists can read, inspect and connect back to known physics, rather than a black-box prediction they must simply trust. Building a database trustworthy enough to support that kind of analysis was itself a major undertaking.

The team did not simply scrape numbers from the literature with automated tools. They combined AI-assisted literature mining with careful human validation, checking extracted information against the original publications, standardizing units and analyzing experimental curves. For metal hydrides, the workhorse class of solid-state hydrogen storage materials that hold hydrogen within their crystal structures, the researchers manually curated 4,643 storage-capacity measurements and 652 pairs of enthalpy and entropy measurements.

Those thermodynamic quantities are what allow scientists to predict how hydrogen storage behavior shifts with temperature, and every curated entry retains a link to its original publication so that the data can be traced and verified. The raw experimental quantities behind those numbers are pressure-composition-temperature measurements, often abbreviated as PCT curves, which trace how much hydrogen a material takes up as the pressure is varied at a fixed temperature. Van't Hoff analyses of such curves yield the enthalpy and entropy of hydrogenation, the thermodynamic bookkeeping that determines whether a hydride forms readily or reluctantly at a given temperature.

By converting thousands of published curves and analyses into standardized, machine-readable entries, DigHyd turns decades of scattered laboratory work into a coherent dataset. That kind of curation is rarely glamorous, but it is the foundation on which any credible data-driven discovery effort in materials science must rest. With the curated dataset in hand, the researchers applied an AI method called GoodRegressor, a form of symbolic regression that identifies relationships in data and expresses them as explicit mathematical equations.

In five-fold cross-validation tests, the method achieved predictive performance comparable to leading black-box models while producing equations that remained interpretable. Matching the accuracy of black-box approaches while remaining transparent allowed the team to move beyond predicting material behavior and begin investigating the physical mechanisms behind it. The resulting physical map revealed an interconnected hierarchy of material properties that influence hydrogen storage capacity and equilibrium pressure.

Lighter host atoms generally favor greater hydrogen storage capacity by weight, a straightforward consequence of the fact that capacity per unit mass improves when the atoms doing the hosting are not heavy. In interstitial hydrides, where hydrogen occupies the spaces between metal atoms rather than bonding into distinct chemical units, the size of the atoms and the arrangement of atoms in the crystal lattice become critically important, because those geometric factors determine what kind of interstitial sites are available for hydrogen to occupy. Properties such as thermal expansion and thermal conductivity also provide clues to how a material accommodates hydrogen, adding further structure to the map.

One of the most physically revealing findings concerns mechanics. When hydrogen enters a metal lattice, the structure must deform and expand to make room for it. Stiffer materials resist that deformation, which increases the elastic energy required to accommodate the hydrogen.

That extra energy cost makes the resulting hydride less stable and raises its equilibrium pressure, meaning the material holds onto its hydrogen less willingly and requires different conditions to release it. The team also identified connections between electronegativity, thermal conductivity and resistance to compression among the metals examined, helping explain why pushing a material to improve its storage capacity can simultaneously change the conditions needed to get that hydrogen back out. Capacity and release pressure, in other words, are coupled through the same underlying physics.

That coupling is precisely why the researchers argue that hydrogen storage capacity and equilibrium pressure should be considered together when designing new materials. Rather than optimizing each material property in isolation, the physical map provides a framework for understanding how several properties interact to shape overall storage performance. It also demonstrates, more broadly, how symbolic regression can uncover previously unrecognized physical relationships hiding in experimental data.

"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," says Orimo. The team plans to expand and refine the database, identify promising material compositions and synthesize candidates for experimental testing.

Comparing measured hydrogen storage behavior with model predictions will help refine the approach and guide the discovery of more effective storage materials. The DigHyd platform is publicly available at https://www.dighyd.org, allowing researchers anywhere to interrogate the same curated data that underpinned the analysis. The study was published in Applied Physics A under the title "Digital hydrogen platform (DigHyd): a rigorously curated database for hydrogen storage materials empowered by AI-assisted literature mining," with the DOI 10.1007/s00339-026-09903-6.

The work was conducted within Japan's World Premier International Research Center Initiative, the MEXT program launched in 2007 to foster globally visible research centers, administered by the Japan Society for the Promotion of Science.

Subject of Research: A curated hydrogen storage materials database and interpretable AI analysis linking capacity and equilibrium pressure

Article Title: Curated hydrogen database reveals links between storage capacity and release pressure

Article References: Curated hydrogen database reveals links between storage capacity and release pressure. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: hydrogen storage, metal hydrides, curated database, symbolic regression, machine learning, thermodynamics, enthalpy, equilibrium pressure, materials science, Tohoku University, DigHyd, data curation