Deciphering diffuse scattering with machine learning and the equivariant foundation model: the case of molten FeO

Area of Science:

• Condensed matter physics
• Materials science
• Computational physics

Background:

• Bridging the gap between scattering data and predicted structures from atom-atom pair potentials in disordered materials is a persistent challenge.
• Traditional methods use approximate interatomic potentials, relating 3D structural models to measured structure factors and pair distribution functions.
• Machine learning potentials show promise, especially for ionic and oxide systems, with recent advances integrating scattering data for improved model accuracy.

Purpose of the Study:

• To evaluate the efficacy of a novel equivariant foundation model, MACE-MP-0, in modeling disordered materials.
• To validate the model's performance against experimental scattering data for molten iron(II) oxide (FeO).
• To assess the potential of foundation models to overcome limitations of classical interatomic potentials.

Main Methods:

• Overview of traditional approaches using approximate interatomic pair potentials.
• Application of a newly introduced equivariant foundation model, MACE-MP-0.
• Validation of model predictions against high-quality experimental diffuse x-ray or neutron scattering data for molten FeO.

Main Results:

• MACE-MP-0 demonstrates potential to surpass traditional limitations of classical interatomic potentials.
• Preliminary results show improved agreement between model predictions and experimental scattering data for molten FeO.
• The model shows promise for accurately capturing local polyhedral bonding and connectivity in disordered systems.

Conclusions:

• Equivariant foundation models like MACE-MP-0 represent a significant advancement in modeling disordered materials.
• Direct integration of scattering data into model development is crucial for accurate predictions.
• This approach holds promise for improving our understanding of complex materials at the atomic level.