Systematic and accurate anharmonic formation free energies of metastable defects via a constrained Bayesian Adaptive Biasing Force framework

In the invisible world of atoms, materials are never perfectly still. Even in a solid block of metal, atoms vibrate, jostle, and occasionally swap places. When a material is damaged by radiation or extreme heat, tiny imperfections called defects appear. These are not just holes or missing pieces; they are complex rearrangements where atoms crowd together or leave gaps behind. Scientists have long known that to predict how a material will behave under stress, they must understand the energy of these defects. However, calculating this energy is notoriously difficult because the atoms do not vibrate in simple, predictable patterns. Instead, their movements are chaotic and interconnected, a phenomenon known as anharmonicity. Furthermore, many of these defects are metastable, meaning they are stuck in a temporary, precarious state that can easily collapse into something else if the atoms move just a little too far. For decades, this complexity has prevented researchers from building a complete library of how these defects behave at different temperatures, leaving a gap in our ability to design materials for nuclear reactors or space exploration.

A team of researchers at the Université Paris-Saclay has now developed a practical way to solve this problem. They created a computational framework that can accurately measure the formation free energy of these unstable defects without losing track of them. In their study, they focused on iron, a material critical for nuclear energy, and examined thousands of different ways that four extra iron atoms could cluster together, as well as how four missing atoms (vacancies) could arrange themselves. The challenge was that standard computer simulations often fail here: as the temperature rises, the atoms vibrate so vigorously that they escape the specific defect shape the scientists are trying to study, causing the calculation to collapse into a different, unrelated state. The researchers overcame this by using a method called constrained Bayesian Adaptive Biasing Force. This technique acts like a gentle, invisible guide that keeps the atoms within the boundaries of the specific defect shape they are investigating, while mathematically correcting for the fact that the atoms were being held back. This allows the computer to sample the chaotic, high-temperature vibrations of the atoms without the defect falling apart.

The results of this work are a massive, systematic database of energy values for hundreds of distinct defect configurations. The researchers tested their method using two different types of mathematical models for how atoms interact: a traditional model used for many years and a newer, data-driven model based on machine learning. They found that the traditional model produced energy landscapes that were rough and irregular, making it difficult to get consistent results, especially at higher temperatures. In contrast, the machine learning model provided a much smoother landscape, allowing the method to work with exceptional stability and precision. The team performed thousands of independent calculations, covering a wide range of temperatures, and found that their method failed in fewer than one percent of cases. This level of reliability is unprecedented for systems of this size, which contain roughly one thousand atoms. They achieved a level of accuracy where the uncertainty in their energy measurements is less than one-thousandth of an electron volt per atom, a precision required to make real-world predictions about material stability.

Perhaps most importantly, the study revealed a deep connection between the simplicity of a defect's shape and the complexity of its energy. The researchers discovered that the chaotic, high-temperature behavior of the atoms could be predicted with surprising accuracy just by looking at the simpler, low-temperature vibrations of the defect. This suggests that the complex, messy reality of atomic motion at high heat is not entirely random but is instead a scaled-up version of the simpler patterns seen at lower temperatures. This finding opens the door to using simpler, faster calculations to estimate the behavior of defects in extreme conditions. By proving that this constrained method works reliably across thousands of different defect shapes and two different types of atomic models, the researchers have provided a general tool that can be applied to almost any material. This work removes a major bottleneck in materials science, offering a clear path to understanding how the smallest imperfections in a material determine its strength and durability in the harshest environments.