Institute of Science Tokyo Uses AI to Identify 22 Promising Neodymium Magnetic Crystal Candidates

Highlights

• Researchers used AI system GMMag trained on 3.54 million crystal records to generate and screen new magnetic materials.
• From 10 million computer-generated attempts, 22 neodymium-containing candidates passed both formation energy and magnetization tests.
• One neodymium-iron-cobalt candidate reached an estimated 1.99 tesla, but critical magnetic anisotropy properties were not established.
• The study is a preprint and has not yet undergone peer review; no physical magnets have been manufactured or tested.
• AI-driven discovery could accelerate rare-earth magnet research but does not replace laboratory validation or secure supply chains.