Machine Learning-Developed Maps for Discovering New Phases Released

Inorganic materials are synthesized by reacting multiple elements. If a new material is successfully synthesized through an unprecedented combination of elements, and that phase exhibits special physical properties or useful functions, it has the potential to become a "treasure" that could be put to practical use as a novel material. However, many of the combinations absent from crystal structure databases are combinations that were previously attempted but simply failed to react, making the ability to predict synthesizability in advance a key factor for efficient discovery of new phases.

The research team developed 80 "elemental reactivity maps" in a 80 × 80 grid format that indicate the likelihood of phase formation from combinations of up to three types of elements, along with the presence or absence of known materials. These maps were created through machine learning using crystal structure data from more than 30,000 inorganic compounds and are published as an interactive web system accessible to anyone.
When the prediction results were validated using experimental crystal structure databases that include data on complex crystals and solid solutions, known compounds were 17 times more likely among combinations with high reactivity scores (≥0.95) compared to combinations with low reactivity scores (<0.05), demonstrating the validity of reactivity scores. Since more than 3,000 combinations of elements that exhibit high reactivity scores but are not present in the experimental databases were identified, the maps are expected to serve as a "treasure trove" containing hidden new phases. The research team also successfully discovered several dozen new phases, including the B20-structure alloy Co(Al,Ge), which attracts attention as a potential magnetic skyrmion or thermoelectric material, by actually utilizing these maps.

By utilizing these elemental reactivity maps, various new phases are expected to be discovered, with the potential for finding novel materials among them. Additionally, since element combinations that are unlikely to react can also be identified from these elemental reactivity maps, they can prove useful in identifying candidates for containers or electrodes that need to remain chemically inert.

• This research was conducted by a research team consisting of Yukari Katsura (Senior Researcher, Materials Modeling Group, Data-driven Materials Research Field, Center for Basic Research on Materials, NIMS; also Associate Professor, University of Tsukuba, and Visiting Scientist, RIKEN), Yuki Inada (Graduate School of Frontier Sciences, UTokyo; completed doctoral program in March 2024), Masaya Fujioka (Senior Researcher, AIST), Haruhiko Morito (Associate Professor, Institute for Materials Research, Tohoku University), and Tohru Sugahara (Professor, Faculty of Materials Science and Engineering, Kyoto Institute of Technology), as part of the JST-CREST Project "Development of Innovative Functional Materials Based on Large-Scale Search for New Crystals" (JPMJCR19J1).
• This research result was published in Chemistry of Materials (online version) on February 21, 2025, and in Vol. 37, Issue 6 of the journal on pages 2097–2105 on March 25, 2025 as the cover article of that issue.