Deep learning helps understanding and predicting the properties of glasses and volcanic lavas

What is it about?

How lavas and magmas flow influences how volcanoes behave. In addition, lavas and magmas have chemical composition close to that of the silicate glasses, obtained by rapid quench of melts, used in everyday objects (windows, lenses, cellphone screen, etc.). Therefore, silicate melts and glasses are of prime importance for materials and Earth sciences. This is why there is an active field of research focusing on understanding how physical and chemical properties of silicate glasses and melts change with their chemical composition and atomic structure. A model that can predict the relationship between chemical composition, atomic structure and physical properties of such material would find great use in many areas, including glass making industry, volcanology, and even for the understanding of primordial magma oceans. In this publication, we combined deep neural networks with physical equations to predict the viscosity (resistance to movement) of silicate melts, as well as many other properties of the melt (configurational entropy, glass transition temperature...) and of the glasses (Raman spectra, density, optical refractive index) interesting for geo- and materials sciences.

Featured Image

Why is it important?

Before this work, there was no easy-to-use model linking multiple properties of melts/glasses with their composition and atomic structure. Here, we present a model using Python and Pytorch, easy-to-use, which combines machine learning with theoretical knowledge to predict multiple properties at once. As such, it represents a step forward, toward better models that can be used in routine.

Perspectives