Temperature-concentration-phase diagrams serve as an informative tool for materials design, however numerous synthesis and measurements are necessary to construct an experimental phase diagram. This work aims to develop a fully automated framework for calculating complete phase diagrams directly from ab-initio simulations, thereby enabling the discovery of new sustainable materials. Our approach combines Machine-Learned Interatomic Potentials (MLIPs), such as the Atomic Cluster Expansion (ACE)[1], with non-equilibrium thermodynamic integration implemented in Calphy[2] to efficiently compute absolute free energies with ab-initio accuracy. Furthermore, we systematically evaluate different approximations, including point defects and their entropic contribution to quantify their effect in reference to experimental phase diagrams. This data-driven approach is implemented in the pyiron[3] workflow manager to enable reproducible high-throughput calculations.

On the poster we demonstrate the construction of the Al-Mg phase diagram using different thermodynamic approximations. Interpolating free energies from molecular dynamics calculation with MLIPs across different concentrations yields smooth phase boundaries, though the γ-phase is suppressed—an effect reflected in formation energy versus concentration plots. By systematically extending the approximations in our automated free-energy workflow, our framework provides a pathway to reliable, ab-initio–based phase diagram prediction in complex alloys. Such capability holds promise for accelerating high-throughput, computational discovery of sustainable materials.

References:

[1] R. Drautz; Physical Review B, 2019, 100, 249901.
[2] S. Menon, et. al.; npj Computational Materials, 2024, 10, 261.
[3] J. Janssen, et. al.; Computational Materials Science, 2019, 163, 24-36.
[4] B. Hallstedt, et. al.; Calphad, 2023, 82, 102577.