Crystallisation Instability in Glassforming Mixtures

Crucial to gaining control over crystallisation in multicomponent materials or accurately modelling rheological behaviour of magma flows is to understand the mechanisms by which crystal nuclei form. The microscopic nature of such nuclei, however, makes this extremely hard in experiments, while computer simulations have hitherto been hampered by their short timescales and small system sizes due to limited computational power. Here we use highly-efficient GPU simulation techniques to access system sizes around 100 times larger than previous studies. This makes it possible to elucidate the nucleation mechanism in a well-studied binary glassformer. We discover that the supercooled liquid is inherently unstable for system sizes of 10,000 particles and larger. This effect is due to compositional fluctuations leading to regions comprised of large particles only which rapidly nucleate. We argue that this mechanism provides a minimum rate of crystallisation in mixtures in general, and use our results to stabilise a model binary mixture and predict glassforming ability for the CuZr metallic glassformer.