Structure and properties of alkali aluminosilicate glasses and melts: Insights from deep learning

Abstract

The full-text may be used and/or reproduced, and given to third parties in any format or medium, without prior permission or charge, for personal research or study, educational, or not-for-prot purposes provided that:• a full bibliographic reference is made to the original source • a link is made to the metadata record in DRO • the full-text is not changed in any way The full-text must not be sold in any format or medium without the formal permission of the copyright holders.

Modeling Viscosity of Volcanic Melts With Artificial Neural Networks

“…For the 32 compositions in the KNAS system in our training/validation database (Table 1), our model performs better than the one by Le Losq et al. (2021), with values of RMSE = 0.29 and RMSE = 0.39, respectively (Figure S2 in Supporting Information ). The model of Le Losq et al.…”

“…For the 32 compositions in the KNAS system in our training/validation database (Table 1) in Supporting Information S1). The model of Le Losq et al (2021) can not be applied to natural volcanic melts in a meaningful way, as these typically contain divalent cations, most importantly MgO, CaO and FeO as well as water.…”

“…However, with a scarcity of η data, this approach cannot be used for volcanic melts. The problem of sparse data is even more pronounced when considering the effect of water on η (Duan, 2014; Giordano et al., 2008; Hui & Zhang, 2007), ignored by all previous ANN‐based models (Cassar, 2021; Le Losq et al., 2021; Tandia et al., 2019). Therefore, we do not follow the previous ANN studies with a gray box approach, but use a black box to utilize as many data as possible; this approach maps the input (composition x and T , where we treat H 2 O on equal footing with the oxide components) directly to viscosity.…”

“…Recently, Tandia et al (2019) have shown the capabilities of ANN to accurately fit and predict melt η for a database of technical glasses. Based on this success, Cassar (2021) and Le Losq et al (2021) produced ANN architectures using gray box approaches, trained on mostly technical glasses and the K 2 O-Na 2 O-Al 2 O 3 -SiO 2 (KNAS) system, respectively. In contrast to a black box approach, which directly maps the input parameters to the desired quantity (here η), their gray box approaches predict input parameters of certain viscosity equations (e.g., VFT or MYEGA) which are then used to compute η.…”

Structural changes in silicate melt: A record from high-field strength elements in the Himalayan Cenozoic leucogranites

“…Regardless of whether magma is primary or derived, it has a specific melt structure (Hou et al, 1992). The structure of magma melt has a significant impact on the thermodynamic properties and physical properties of the magmatic system (such as density and viscosity), thereby affecting related dynamic processes of magma (such as crystallization, rising speed, diffusion, and distribution of elements) (Mysen, 1983;Mo, 1985;Bottinga et al, 1996;Losq et al, 2021). Changes in magma composition, temperature, and pressure are accompanied by changes in the melt structure, which inevitably lead to changes in the physical and chemical properties of the melt (Zhu and Zhang, 1996).…”