A new method to characterize atomic-scale damage in electron microscopy images relies on so-called distortion states, mathematical representations that compactly capture how a crystal lattice deforms relative to its ideal configuration. The approach not only quantifies the degree of alteration already present in a sample but also predicts its evolution, something that has been difficult to extract from the data volumes produced by modern microscopes.

Electron microscopy now yields atomic-resolution images containing vast amounts of information on defects, dislocations and amorphized regions in materials. Yet translating those images into comparable metrics of structural degradation remains a bottleneck: conventional analyses rely on manual inspection or ad hoc descriptors that do not generalize well across materials or irradiation conditions. Having a robust, transferable representation of the distortion state would help assess materials destined for extreme environments, such as nuclear reactors and fusion components.

The work proposes using distortion states as embeddings that summarize the local deformation of the lattice in a low-dimensional space. From there, damage can be measured reproducibly and fed into predictive models that anticipate how the structure will evolve under continued irradiation. The method's key feature is its robustness to noise and to variations inherent in experimental acquisition, making it applicable to real microscopy data and not only to simulations.

The ability to forecast, rather than merely describe, atomic damage opens the door to rapid screening of radiation-resistant materials and to early-diagnosis strategies for structural ageing. The authors note that the framework is extendable to other characterization techniques and to different families of materials, although it will need validation on independent datasets and under varied experimental conditions to establish its true scope.