Rather than letting a model explore blindly, the American laboratory combines high-throughput simulations and reasoning AI to predict magnetic properties before any synthesis.
In an article published in Advanced Functional Materials and relayed on June 3, 2026 by the Ames National Laboratory, researcher Prashant Singh outlines a method to accelerate the discovery of permanent magnets devoid of rare earths. The approach combines three bricks: fundamental physical modelling, high-throughput simulations, and AI tools capable of reasoning about these results to guide research before any time in the laboratory.
The issue is what determines the quality of a magnet: atomic structure and electronic behaviour determine magnetization, stored energy, resistance to demagnetization, and temperature resistance. By integrating these physical relationships into the learning framework, rather than allowing a statistical model to extrapolate freely, predictions remain rooted in real-world behaviors. Singh also points to Ames' strength: a long-standing history of data on magnetic materials, which no other institution has at this level, and which serves as a training base.
The motivation is as much industrial as scientific. Rare earths give high-performance magnets their strength and stability, but they are expensive and come from supply chains that the United States considers fragile. These magnets are used in both energy production and defense applications. The work is part of the Department of Energy's Genesis Mission, which brings together national laboratories, industry and universities to discuss the use of AI for energy security and the supply of critical minerals.
Read the full article on the original source
Go to the source →