Machine-learning improves understanding of glass formation in metallic systems

Abstract

Glass-forming ability (GFA) in metallic systems remains a little-understood property. Experimental work on bulk metallic glasses (BMGs) is guided by many empirical criteria, which are often of limited predictive value. This work uses machine-learning both to produce predictive models for the GFA of alloy compositions, and to reveal insights useful for furthering theoretical understanding of GFA. Our machine-learning models apply a novel neural-network architecture to predict simultaneously the liquidus temperature, glass-transition temperature, crystallization-onset temperature, maximum glassy casting diameter, and probability of glass formation, for any given alloy. Feature permutation is used to identify the features of importance in the black-box neural network, recovering Inoue's empirical rules, and highlighting the effect of discontinuous Wigner–Seitz boundary electron-densities on atomic radii. With certain combinations of elements, atomic radii of different species contract and expand to balance electron-density discontinuities such that the overall difference in atomic radii increases, improving GFA. We calculate adjusted radii via the Thomas–Fermi model and use this insight to propose promising novel glass-forming alloy systems.