Noise filtering of cluster expansions in multicomponent systems

Multiscale modeling of materials based on cluster expansions has been performed hitherto assuming that the input data, either from first-principles or experimental observations, were noise-free. Our systematic analyses on archetypical systems subjected to Gaussian numerical noise revealed that cluster expansions act as noise filters. This unexpected and remarkable feature has a deep impact on the way we approach the modeling of material properties and how we construct future databases.