This project aims to confirm the presence of multiple new strong magnetic compounds, or families thereof, characterized by their saturation magnetization, anisotropy and curie temperature, and…

The Department of Energy Advanced Research Projects Agency awarded the University of Oregon a $1.946 million cooperative agreement on July 30, 2026, under the Advanced Research Projects Agency - Energy program (CFDA 81.135) to discover and characterize new strong magnetic compounds for technological viability assessment.

The University of Oregon will combine machine learning with experimental synthesis and characterization to identify novel magnetic materials. The approach leverages existing data on binary and ternary compounds to train predictive models that forecast phase spaces containing strong-spin compounds, then uses solution growth techniques and single-crystal synthesis to validate predictions and map multinary phase diagrams. Electronic structure calculations will elucidate the chemical origins of magnetic behaviors in promising materials, creating a closed-loop feedback cycle between computation and experiment. The work will yield generalized design rules for phase diagram prediction that accelerate discovery and establish a reusable platform for developing high-performance magnets and related material systems.

Performance occurs in Eugene, Oregon through July 29, 2029.

This project aims to confirm the presence of multiple new strong magnetic compounds, or families thereof, characterized by their saturation magnetization, anisotropy and curie temperature, and assessed for their technological viability. This accomplishment follows from the other deliverables of maps, including generalizable design rules for phase diagram prediction that accelerates materials discovery of strong-spin phase spaces to create a closed-loop cycle for developing high-performance magnets, and creates a general platform to realize other non-magnetic materials. The overall technical approach combines theory and computation with experiment and active learning. Maps leverages the wealth of information known for binary and ternary compounds as the initial training data for machine learning models to predict phase spaces having strong magnetic compounds. Here, machine learning is used in conjunction with thermodynamic and chemical principles to enable predictions of high-dimensional systems not present in the training set. This predictive framework is coupled tightly to experiment, using solution growth techniques to robustly map multinary phase diagrams and identify new phases.6 a method of reusable chemical reactions makes phase space mapping efficient, and the results will form a closed-loop between computation and experiment as feedback to our models. When promising magnetic materials are identified, the team has expertise in solid-state synthesis to grow single crystals appropriate for crystallographic and magnetic characterization,6,810 while electronic structure calculations aid in determining the chemical origins of magnetic BEHAVIORS11 that inform phasespace predictions.