Machine learning modeling of the atomic structure and physical properties of alkali and alkaline-earth aluminosilicate glasses and melts

Abstract

The first version of the machine learning greybox model i-Melt was trained to predict latent and observed properties of K$_2$O-Na$_2$O-Al$_2$O$_3$-SiO$_2$ melts and glasses. Here, we extend the model compositional range, which now allows accurate predictions of properties for glass-forming melts in the CaO-MgO-K$_2$O-Na$_2$O-Al$_2$O$_3$-SiO$_2$ system, including melt viscosity (accuracy equal or better than 0.4 log$_{10}$ Pa$\cdot$s in the 10$^{-1}$-10$^{15}$ log$_{10}$ Pa$\cdot$s range), configurational entropy at glass transition ($\leq$ 1 J mol$^{-1}$ K$^{-1}$), liquidus ($\leq$ 60 K) and glass transition ($\leq$ 16 K) temperatures, heat capacity ($\leq$ 3 \%) as well as glass density ($\leq$ 0.02 g cm$^{-3}$), optical refractive index ($\leq$ 0.006), Abbe number ($\leq$ 4), elastic modulus ($\leq$ 6 GPa), coefficient of thermal expansion ($\leq$ 1.1 10$^{-6}$ K$^{-1}$) and Raman spectra ($\leq$ 25 \%). Uncertainties on predictions also are now provided. The model offers new possibilities to explore how melt/glass properties change with composition and atomic structure.