Automated Atomistic Construction and Interface Stabilization of Fe3O4@Au Core–Shell Nanostructures via Machine Learning

Abstract The structural stability of metal oxide-metal core-shell nanoparticles at the nanoscale is governed by atomistic interface effects that are not captured by conventional continuum approximations. In this work, we introduce an automated “virtual synthesis” framework for the deterministic construction, relaxation, and validation of Fe3O4@Au core–shell nanoparticles using a machine-learning interatomic potential (CHGNet). The pipeline integrates geometric assembly based on crystallographic symmetry and Wulff construction, CHGNet-assisted energetic relaxation, and postrelaxation quality control to ensure chemically and structurally consistent interfaces. Systematic parametric analysis reveals strict structural viability thresholds inherent to nanoscale heterostructures. Specifically, core diameters below D < 7.0 Å fail to preserve spinel coordination, while Au shell thicknesses below T < 0.8 Å do not yield a percolated metallic network for a 20 Å particle, to generate a continuous shell. Radial distribution analysis confirms the formation of polycrystalline FCC Au shells anchored by strong Au–O interactions (≈2.20 Å), alongside a measurable lattice compression of ∼4%, indicative of curvature-induced surface stress. Together, these results establish quantitative lower bounds for structural feasibility in magneto-plasmonic core–shell systems and demonstrate the capability of graph neural network-based interatomic potentials to bridge the gap between continuum design and atomistic realism.