Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

MM: Fachverband Metall- und Materialphysik

MM 13: Data-driven Materials Science: Big Data and Workflows I

MM 13.4: Vortrag

Dienstag, 10. März 2026, 11:00–11:15, SCH/A251

Automated Prediction of Phase Stability with ab-initio Accuracy — •Prabhath Chilakalapudi, Marvin Poul, Jan Janssen, and Jörg Neugebauer — Computational Materials Design, Max-Planck-Institute for Sustainable Materials, Düsseldorf

Developing sustainable metallic alloys-free of toxic elements and compatible with circular synthesis-requires novel and efficient ways to explore large composition spaces. A key bottleneck is the automated, ab-initio-accurate prediction of temperature-composition phase diagrams, where experimental phase data is limited or difficult to obtain.

We present a reproducible, automated workflow that uses Machine-Learned Interatomic Potentials (MLIPs) such as Atomic Cluster Expansion (ACE)[1] and non-equilibrium thermodynamic integration (Calphy[2]) to compute free energies and phase stabilities. By analysing key approximations including point-defect models, different entropic contributions, and free-energy interpolation schemes, we quantify the reliability of the calculated phase boundaries and provide meaningful ``error bars'' on the diagram. The workflow is demonstrated for representative binary alloys and is structured for gradual scaling to multicomponent systems. We leverage the pyiron[3] workflow framework for reproducible and efficient automation, to accelerate the discovery of sustainable materials.

[1] R. Drautz, Phys. Rev. B 100, 249901 (2019).

[2] S. Menon et al., npj Comput. Mater. 10, 261 (2024).

[3] J. Janssen et al., Comput. Mater. Sci. 163, 24-36 (2019).

Keywords: Sustainability; Phase diagrams; Workflows; Machine learning