Atomistic Insights into Alloy Solidification using Machine-Learning Potentials

Abstract

Alloy solidification is a critical process in materials design and manufacturing, as it governs the formation of microstructures that determines the mechanical, thermal, and chemical properties of materials. However, direct in situ observation remains extremely challenging due the need for high spatial and temporal resolution under elevated temperatures. On the theory side, solidification is a complex phenomenon often studied using phase-field simulations, which rely on empirically fitted parameters and simplified assumptions about interfacial kinetics, limiting their predictive capability. Capturing this process at the atomistic level can yield more fundamental insights, but is hindered by the need for interatomic models that are both accurate and computationally efficient across relevant timescales and length scales. To overcome these challenges, this thesis develops and applies machine-learning interatomic potentials (MLPs) that capture the chemical complexity of metallic alloys, providing a physically accurate and computationally efficient backbone for large-scale atomistic simulations of complex alloy solidification. We first address a foundational challenge in deploying MLPs: the systematic construction of robust and transferable training datasets. Using CrCoNi as a model system, we evaluate various strategies for training MLPs to capture chemical short-range order (SRO), a critical feature in high-entropy alloys, and its effects on materials quantities of relevance for mechanical properties, such as stacking-fault energy and phase stability. It is demonstrated that energy accuracy on test sets often does not correlate with accuracy in capturing material properties, which is fundamental in enabling large-scale atomistic simulations of metallic alloys with high physical fidelity. Based on this analysis we systematically derive design principles for the rational construction of MLPs that capture SRO in the crystal and liquid phases of alloys. The resulting MLPs are validated against experimental measurements on key thermophysical properties, including melting points, heat capacities, thermal expansion coefficients, and enthalpy of SRO formation, confirming their suitability for predictive simulations. With these validated potentials, we then investigate the evolution of SRO during rapid solidification processes. Our simulations reveal that alloy processing can lead to nonequilibrium steady states of SRO that differ qualitatively from any equilibrium configuration. We attribute this behavior to an inherent ordering bias introduced by nonequilibrium dynamics during solidification. These findings suggest that conventional manufacturing processes offer new opportunities to tailor alloy properties by accessing a broader spectrum of nonequilibrium SRO states, expanding the alloy design space beyond the equilibrium spectrum. Finally, we conduct predictive solidification simulations of chemically complex alloys across experimentally relevant growth rates (0.15–2 m/s) , alloy compositions, interface orientations, and undercooling levels. These simulations capture the dynamic build up of solute partitioning at the solid-liquid interface and reveal kinetics-dependent segregation patterns that deviate markedly from equilibrium predictions. The developed framework enables direct evaluation of key kinetic properties under realistic growth conditions, including interface mobility, liquid diffusivity, and solute trapping. Altogether, this thesis develops machine-learning potentials capable of capturing the chemical complexity of metallic alloys with near DFT-level accuracy, and establishes a framework for extracting key kinetic properties through predictive simulations of alloy solidification. When combined with emerging advances in continuum-scale modeling, these results lay the groundwork for truly multiscale investigations of alloy solidification, enabling DFT-level predictive capabilities at scales directly comparable to experimental alloy design and additive manufacturing processes.