Benchmarking foundation models for simulating radiation damage in iron alloys

Recent advances in machine-learned interatomic potentials (MLIPs) allow them to be trained across many elements, giving foundation models that can be used for complex alloys and a wide range of properties. In this poster, we benchmark six pre-trained foundation models - two MLIP frameworks, MACE [2] and GRACE [3], trained on three databases each - for modelling typical high energy environments seen in iron collision cascades, and solute-defect interactions in the iron lattice for the main six alloying elements of Eurofer steel for fusion reactors.