Machine Learning to Predict Quasicrystals from Chemical Compositions

Quasicrystals have emerged as the third class of solid-state materials, distinguished from periodic crystals and amorphous solids, which long-range order without periodicity exhibiting rotational symmetries that are disallowed for in most cases. To date, more than one hundred stable quasicrystals been reported, leading to discovery many new exciting phenomena. However, pace has lowered recent years, largely owing lack clear guiding principles synthesis quasicrystals. Here, it is shown can be accelerated with a simple machine-learning workflow. With list chemical compositions known quasicrystals, approximant crystals, ordinary prediction model trained solve three-class classification task its predictability compared observed phase diagrams ternary aluminum systems evaluated. The validation experiments strongly support superior predictive power machine learning, overall accuracy reaching ≈0.728. Furthermore, analyzing input–output relationships black-boxed into model, nontrivial empirical equations interpretable by humans describe conditions necessary quasicrystal formation identified.