Calculation of thermodynamic properties and phase diagrams through atomistic simulations can provide valuable insights for the design and assessment of new materials and their properties. Accurate computation of phase diagrams involves determining the Helmholtz and Gibbs free energies for various phases, along with their dependencies on thermodynamic state variables. However, conventional computational approaches to free energy estimation are often technically complex, computationally demanding, and consist of a number of interdependent steps that require human expertise.

In this work, we present algorithms aimed at facilitating the computation of multicomponent phase diagrams. First, relevant phases are identified through an integrated use of materials databases and universal machine learning interatomic potentials. The free energies of these phases are estimated using atomic cluster expansion potentials via a two-step methodology: temperature-dependent free energies are obtained through non-equilibrium thermodynamic integration, while composition-dependent variations are evaluated by alchemical sampling. Both vibrational and configurational entropy contributions are systematically taken into account. Additionally, our method seeks to detect phase transformations during simulations using local atomic environment descriptors. Finally, stability and coexistence regions of the various phases are derived from the computed free energies, allowing for the construction of a composition-temperature phase diagram.

We illustrate the methodology by computing several binary temperature-composition phase diagrams and provide corresponding computational tools. Our workflows are designed to be independent of the interatomic potential and material system, supporting broader use and offering a step toward making the computation of thermodynamic phase diagrams more accessible in the context of atomistic simulations.