Daily AI Paper

Today’s standout paper is about making finite-temperature materials discovery much easier. In solid-state science, we usually want free energies, because they tell us which crystal phases are actually stable at a given temperature and pressure. The problem is that free energy is expensive to compute: it often requires ensemble simulations or specialized thermodynamic calculations, so large-scale screening has mostly relied on simpler ground-state energies instead. This paper introduces thermodynamic interatomic potentials, or TIPs, which extend a standard interatomic potential into a model that predicts Gibbs free energy directly and remains thermodynamically consistent. The key idea is to use automatic differentiation so temperature and pressure responses come out of the model naturally. With a single evaluation, the system can estimate an equation of state and identify phase transitions, including dynamically stabilized phases. That matters because it turns free-energy-based phase stability into something much more accessible, potentially accelerating high-throughput searches for new alloys and crystalline materials.