Dresden 2026 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

MM: Fachverband Metall- und Materialphysik

MM 5: Topical Session: Physics-driven Artificial Intelligence for Materials I

MM 5.7: Talk

Monday, March 9, 2026, 12:30–12:45, SCH/A251

Towards automated calculation of phase diagrams with machine learning interatomic potentials — •Sarath Menon and Ralf Drautz — ICAMS, Ruhr University Bochum, Germany

Calculation of thermodynamic properties and phase diagrams through atomistic simulations provides valuable insights for the design and assessment of new materials. Accurate phase diagram prediction requires determining the Helmholtz and Gibbs free energies for relevant phases and understanding their dependence on thermodynamic state variables, yet conventional approaches remain technically complex and computationally demanding.

In this work, we introduce algorithms that streamline the computation of multicomponent phase diagrams. Relevant phases are identified using a combination of materials databases and machine learning interatomic potentials, and their free energies are computed with atomic cluster expansion potentials. Temperature and composition effects are assessed through non-equilibrium thermodynamic integration and alchemical sampling, including both vibrational and configurational entropy contributions.

We demonstrate the methodology by computing unary pressure-temperature and several binary temperature-composition phase diagrams as well as phase equilibria in ternary materials. We provide all corresponding computational tools. The workflows are designed to be independent of the interatomic potential and material system, supporting broader use and advancing the accessibility of thermodynamic phase diagram computation in atomistic simulations.