From electrons to phase diagrams with machine learning potentials using pyiron based automated workflows

Automated pyiron workflow linking interatomic potentials with calphy free-energy calculations, enabling ab initio phase diagram construction for Al–Li and other alloy systems in materials science. Image by Menon et al. on npj Computational Materials, Licensed under CC-BY-4.0 (Link).

Link to DemonstratorLink to Publication

This demonstrator illustrates how interatomic potentials can be used to calculate thermodynamic properties and construct phase diagrams for materials science applications. The workflow highlights the integration of machine learning potentials (MLPs) and empirical models into automated simulation pipelines.

Using calphy, a tool for the automated calculation of Helmholtz and Gibbs free energies, the demonstrator shows how free energies derived from atomistic simulations can be directly linked to phase stability and phase diagram construction. All steps are managed in pyiron, ensuring reproducibility and seamless integration of database generation, potential fitting, validation, and thermodynamic assessment.

As an example, the demonstrator focuses on the Al–Li alloy system, an important lightweight material for aerospace applications. By combining DFT-based datasets, MLP training (HDNNP, ACE, EAM), and free-energy methods, the workflow bridges electronic-structure accuracy with mesoscale thermodynamic predictions.

This demonstrator highlights how pyiron workflows can connect simulation data, machine learning, and automated thermodynamics tools into a coherent environment, accelerating the adoption of data-driven and AI-assisted approaches in materials science.