Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MM: Fachverband Metall- und Materialphysik
MM 22: Data-driven Materials Science: Big Data and Workflows III
MM 22.7: Vortrag
Mittwoch, 11. März 2026, 12:00–12:15, SCH/A216
MACE-based Machine Learning Interatomic Potentials for Iron-Nickel Alloys: Validation Across Composition and Pressure Ranges — •Kushal Ramakrishna1, Mani Lokamani1, and Attila Cangi1,2 — 1Helmholtz-Zentrum Dresden-Rossendorf (HZDR), D-01328 Dresden, Germany — 2Center for Advanced Systems Understanding (CASUS), D-02826 Görlitz, Germany
Machine-learned interatomic potentials have emerged as powerful tools bridging quantum-level accuracy with mesoscale simulations in computational materials science. We present a comprehensive evaluation of MACE models for iron-nickel alloys across a wide range of compositions and pressures, with direct relevance to Earth's core modeling and industrial applications. We construct special quasirandom structures (SQS) to simulate random iron-nickel alloy configurations and train MACE models on density functional theory datasets combined with experimental validation data. Extensive short-range order analysis confirms improved chemical randomness for larger supercells, critical for faithful property sampling. Multiple MACE flavors are systematically compared against experimental measurements for structural and elastic properties in both body-centered cubic and face-centered cubic phases. Our results demonstrate that fine-tuned MACE models achieve remarkable predictive accuracy for equation-of-state behavior and elastic properties across all compositions. This approach successfully bridges computational predictions with experimental observations, enabling accelerated materials discovery for technologically relevant transition metal alloys.
Keywords: Machine Learning Interatomic Potentials; Density Functional Theory; Computational Materials Science; Iron-Nickel Alloys