Paper Overview
Field: Machine Learning / Materials Science Authors: Juno Nam, Bowen Deng, Xiaochen Du arXiv: 2508.08536
Abstract
Free energies govern solid-state phase stability, yet computational materials discovery still relies largely on ground-state energies because free energy calculations require ensemble averages. We introduce the thermodynamic interatomic potential (TIP), which extends an interatomic potential from its static energy to a thermodynamically consistent Gibbs free energy model, with thermodynamic responses following from temperature and pressure by automatic differentiation. We implement TIP[UMA] using the universal potential UMA, train it on free energies from quasi-harmonic to molecular dynamics fidelity, and calibrate it to higher-resolution calculations or experiment. From a single evaluation, it returns the equation of state of a crystal and locates phase transitions among competing branches, including dynamically stabilized phases. Fine-tuning extends the model to alloy solubility limits and miscibility gaps. TIP makes free energies as accessible as potential energies, opening the door to high-throughput discovery of finite-temperature phase stability.
Chinese Summary (translated)
Free energy governs solid-state phase stability, but computational materials discovery still relies mainly on ground-state energies, since free-energy calculations require ensemble averages. TIP extends an interatomic potential beyond its static energy to a thermodynamically consistent Gibbs free-energy model, with thermodynamic responses derived via automatic differentiation in temperature and pressure. Implemented with the universal potential UMA, trained on free energies from quasi-harmonic to molecular-dynamics accuracy, and calibrated to higher-resolution computation or experiment, TIP[UMA] returns a crystal's equation of state from a single evaluation and locates phase transitions between competing branches, including dynamically stabilized phases. Fine-tuning extends it to alloy solubility limits and miscibility gaps.
Links
- arXiv: https://arxiv.org/abs/2508.08536