Crystal Nucleation in Eutectic Al-Si Alloys by Machine-Learned Molecular Dynamics

Solidification control is crucial in manufacturing technologies, as it determines the microstructure and, consequently, the performance of the final product. Investigating the mechanisms occurring during the early stages of nucleation remains experimentally challenging as it initiates on nanometer length and sub-picoseconds time scales. Large scale molecular dynamics simulations using machine learning interatomic potential with quantum accuracy appears the dedicated approach to complex, atomic level, multidimensional mechanisms with local symmetry breaking. A potential trained on a high-dimensional neural network on density functional theory-based ab initio molecular dynamics (AIMD) trajectories for liquid and undercooled states for Al-Si binary alloys enables us to study the nucleation mechanisms occurring at the early stages from the liquid phase near the eutectic composition. Our results indicate that nucleation starts with Al in hypoeutectic conditions and with Si in hypereutectic conditions. Whereas Al nuclei grow in a globular shape, Si ones grow with polygonal faceting, whose underlying mechanisms are further discussed.

💡 Research Summary

This paper presents a comprehensive study of early‑stage crystal nucleation in eutectic Al‑Si alloys using a high‑dimensional neural network potential (HDNNP) trained on density‑functional‑theory (DFT) based ab‑initio molecular dynamics (AIMD) data. The authors first generated a diverse AIMD dataset covering the full composition range (xSi = 0.10, 0.25, 0.50, 0.75) and temperatures spanning the liquid, supercooled, and solid states. For each composition, 750 configurations were extracted, yielding a total of several thousand atomic environments. These were described with Behler‑Parrinello atom‑centered symmetry functions (radial G2 and angular G5) and split into 90 % training and 10 % test sets. Training of the HDNNP achieved an energy root‑mean‑square error below 2 meV/atom, and validation against the test set showed excellent agreement in radial distribution functions (RDF) and mean‑square displacements (MSD) when compared to the original AIMD trajectories.

With the validated potential, large‑scale molecular dynamics simulations were performed using LAMMPS equipped with the ml‑hdnnp package. System sizes reached up to two million atoms, allowing the authors to explore nucleation under realistic thermodynamic conditions. Two nucleation scenarios were investigated: (i) homogeneous nucleation in a hypoeutectic alloy (5 at % Si) subjected to rapid quenching, and (ii) heterogeneous nucleation in a hyper‑eutectic alloy (>12 at % Si) seeded with a pre‑formed diamond‑Si particle. The simulations employed an initial NVT equilibration followed by an NPT production run to let the volume adjust during the liquid‑solid transition.

The results reveal a composition‑dependent nucleation pathway. In the hypoeutectic regime, Al atoms aggregate in Si‑depleted regions, forming compact clusters that quickly adopt a globular, near‑spherical morphology. Their growth follows a diffusion‑controlled kinetic law (R ∝ t¹/²), and polyhedral template matching (PTM) analysis shows rapid propagation of the fcc‑Al lattice into the surrounding melt. Conversely, in the hyper‑eutectic regime, Si atoms nucleate as faceted polyhedra, predominantly exposing {111} and {110} facets characteristic of the diamond cubic structure. Growth in this case is interface‑controlled, with the Si nucleus maintaining sharp edges and corners, and the surrounding liquid rapidly reorganizes into the diamond‑Si lattice.

The authors further validated the potential by comparing simulated total structure factors S(q) with experimental X‑ray and neutron diffraction data for liquid Al‑Si alloys; the agreement confirms that the HDNNP captures both static and dynamic liquid properties. The observed morphological dichotomy—globular Al nuclei versus faceted Si nuclei—matches long‑standing experimental observations of microstructures in cast Al‑Si components, where hypoeutectic alloys display Al‑rich globular particles and hyper‑eutectic alloys exhibit Si‑rich polygonal particles.

In conclusion, the study demonstrates that a machine‑learned interatomic potential trained on a modest AIMD dataset can achieve quantum‑level fidelity while enabling simulations at scales necessary to observe nucleation events. The work provides atomistic insight into why Al nuclei tend to be spherical and Si nuclei tend to be faceted, linking these tendencies to differences in surface energy anisotropy and diffusion behavior. These findings have direct implications for alloy design and solidification processing, offering a predictive tool for tailoring microstructures in additive manufacturing, casting, and other industrial processes involving Al‑Si alloys.