TY - JOUR
T1 - Structure and properties of alkali aluminosilicate glasses and melts
T2 - Insights from deep learning
AU - Le Losq, Charles
AU - Valentine, Andrew P.
AU - Mysen, Bjorn O.
AU - Neuville, Daniel R.
N1 - Publisher Copyright:
© 2021 The Authors
PY - 2021/12/1
Y1 - 2021/12/1
N2 - Aluminosilicate glasses and melts are of paramount importance for geo- and materials sciences. They include most magmas, and are used to produce a wide variety of everyday materials, from windows to smartphone displays. Despite this importance, no general model exists with which to predict the atomic structure, thermodynamic and viscous properties of aluminosilicate melts. To address this, we introduce a deep learning framework, ‘i-Melt’, which combines a deep artificial neural network with thermodynamic equations. It is trained to predict 18 different latent and observed properties of melts and glasses in the K2O-Na2O-Al2O3-SiO2 system, including configurational entropy, viscosity, optical refractive index, density, and Raman signals. Viscosity can be predicted in the 100–1015 log10 Pa·s range using five different theoretical frameworks (Adam-Gibbs, Free Volume, MYEGA, VFT, Avramov-Milchev), with a precision equal to, or better than, 0.4 log10 Pa·s on unseen data. Density and optical refractive index (through the Sellmeier equation) can be predicted with errors equal or lower than 0.02 and 0.006, respectively. Raman spectra for K2O-Na2O-Al2O3-SiO2 glasses are also predicted, with a relatively high mean error of ∼25% due to the limited data set available for training. Latent variables can also be predicted with good precisions. For example, the glass transition temperature, Tg, can be predicted to within 19 K, while the melt configurational entropy at the glass transition, Sconf(Tg), can be predicted to within 0.8 J mol−1 K−1. Applied to rhyolite compositions, i-Melt shows that the rheological threshold separating explosive and effusive eruptions correlates with an increase in the fraction of non-bridging oxygens in rhyolite melts as their alkali/Al ratio becomes larger than 1. Exploring further the effect of the K/(K + Na) ratio on the properties of alkali aluminosilicate melts with compositions varying along a simplified alkali magmatic series trend, we observe that K-rich melts have systematically different structures and higher viscosities compared to Na-rich melts. Combined with the effects of the K/(K + Na) ratio on other parameters, such as the solubility, solution mechanisms and speciation of volatile elements, this could ultimately influence the eruptive dynamics of volcanic systems emitting Na-rich or K-rich alkali magmas.
AB - Aluminosilicate glasses and melts are of paramount importance for geo- and materials sciences. They include most magmas, and are used to produce a wide variety of everyday materials, from windows to smartphone displays. Despite this importance, no general model exists with which to predict the atomic structure, thermodynamic and viscous properties of aluminosilicate melts. To address this, we introduce a deep learning framework, ‘i-Melt’, which combines a deep artificial neural network with thermodynamic equations. It is trained to predict 18 different latent and observed properties of melts and glasses in the K2O-Na2O-Al2O3-SiO2 system, including configurational entropy, viscosity, optical refractive index, density, and Raman signals. Viscosity can be predicted in the 100–1015 log10 Pa·s range using five different theoretical frameworks (Adam-Gibbs, Free Volume, MYEGA, VFT, Avramov-Milchev), with a precision equal to, or better than, 0.4 log10 Pa·s on unseen data. Density and optical refractive index (through the Sellmeier equation) can be predicted with errors equal or lower than 0.02 and 0.006, respectively. Raman spectra for K2O-Na2O-Al2O3-SiO2 glasses are also predicted, with a relatively high mean error of ∼25% due to the limited data set available for training. Latent variables can also be predicted with good precisions. For example, the glass transition temperature, Tg, can be predicted to within 19 K, while the melt configurational entropy at the glass transition, Sconf(Tg), can be predicted to within 0.8 J mol−1 K−1. Applied to rhyolite compositions, i-Melt shows that the rheological threshold separating explosive and effusive eruptions correlates with an increase in the fraction of non-bridging oxygens in rhyolite melts as their alkali/Al ratio becomes larger than 1. Exploring further the effect of the K/(K + Na) ratio on the properties of alkali aluminosilicate melts with compositions varying along a simplified alkali magmatic series trend, we observe that K-rich melts have systematically different structures and higher viscosities compared to Na-rich melts. Combined with the effects of the K/(K + Na) ratio on other parameters, such as the solubility, solution mechanisms and speciation of volatile elements, this could ultimately influence the eruptive dynamics of volcanic systems emitting Na-rich or K-rich alkali magmas.
KW - Aluminosilicate melt
KW - Glass
KW - Machine learning
KW - Magma
KW - Thermodynamic properties
KW - Viscosity
KW - Volcanology
KW - grey-box neural networks
UR - http://www.scopus.com/inward/record.url?scp=85115954687&partnerID=8YFLogxK
U2 - 10.1016/j.gca.2021.08.023
DO - 10.1016/j.gca.2021.08.023
M3 - Article
SN - 0016-7037
VL - 314
SP - 27
EP - 54
JO - Geochimica et Cosmochimica Acta
JF - Geochimica et Cosmochimica Acta
ER -