Thermodynamic constraints in models, reproducible data, inverse design and automated feedback: Tohoku University is designing an ecosystem where physics frames prediction.
Finding a material that stores enough hydrogen, releases it under exploitable conditions, reacts quickly and survives hundreds of cycles remains a problem with multiple stresses. An international team led by the Advanced Institute for Materials Research of the University of Tohoku argues that IA's acceleration of this research will not come from faster predictions, but from a framework that prevents predictions from being unverifiable.
The perspective, published in ACS Energy Letters August 14, 2026, released in press release on September 8, builds on discussions by 69 researchers from hydrogen storage, AI, scientific computing and autonomous laboratories. It articulates four elements: data management designed for reproducibility, models forced by thermodynamics and kinetics, reverse design driven by IA, and experimental validation integrated into the cycle rather than added to the end.
" A reliable discovery guided by AI in this area cannot come from faster predictions alone," sums up Seong-Hoon Jang, associate professor: it requires physical constraints, transparent data and a continuous return of real experiences, inscribed in the process from the start. Tohoku also pairs these tools with automated synthesis and measurement systems, as part of the JST GteX program.
Read the full article on the original source
Go to the source →