DeepMind Models Transition of Glass from Liquid to Solid with AI

In short, glass is made by melting a mix of sand and other minerals that then cools into the solid state that we know as glass. When we look at a solid glass pane under a microscope, we don’t see an orderly collection of molecules like we would see if we were examining a crystal. Instead, we see a jumble of molecules with no discernible structure in a state that looks very similar to the molecules of the original molten liquid mix. Following the glass transition from liquid to solid, it’s almost as if the molecules of the liquid state become flash-frozen in place in the solid state.

Figure 1: A liquid, when cooled too quickly past its crystallization point, turns into a supercooled liquid which, upon further cooling, turns into a disordered, amorphous glass. If cooled slowly enough, it may instead transform into an ordered crystal.

Whether this transition corresponds to a structural phase process (like in freezing water) has remained an unanswered question. Put well by Nobel laureate Philip W. Anderson:

The deepest and most interesting unsolved problem in solid state theory is probably the theory of the nature of glass and the glass transition.

Read More: The Future of Glass

Researchers at DeepMind have recently made a breakthrough in understanding the glass transition with a newly developed AI system that can predict the movement of molecules through the glass transition. Motivated by the fact that glass is a great environment for applying machine learning methods to physical problems, DeepMind set out to apply Graph Neural Networks to predict physical aspects of glass.

“MODEL ARCHITECTURE: A) FROM THE 3-D INPUTS, NODES AT DISTANCE LESS THAN 2 ARE CONNECTED TO FORM A GRAPH. AFTER PROCESSING, THE NETWORK PREDICTS MOBILITIES (REPRESENTED BY DIFFERENT COLORS) FOR EACH PARTICLE. B) THE GRAPH NETWORK’S CORE FIRST UPDATES EDGES BASED ON THEIR PREVIOUS EMBEDDING AND THOSE OF THEIR ADJACENT NODES, AND THEN NODES BASED ON THEIR PREVIOUS EMBEDDINGS AND THOSE OF INCOMING EDGES. C) THE GRAPH NETWORK CONSISTS OF AN ENCODER, SEVERAL APPLICATIONS OF THE CORE, FOLLOWED BY A DECODER. EACH APPLICATION OF THE CORE INCREASES THE SHELL OF PARTICLES CONTRIBUTING TO A GIVEN PARTICLE’S PREDICTION, HERE SHOWN IN COLOR FOR THE CENTRAL PARTICLE (DARK BLUE).” – DeepMind

After ample testing using multiple datasets, DeepMind found that their graph networks strongly outperform existing models when applied to simulated three dimensional glasses. We encourage you to learn more about DeepMind’s well-documented process here.

Read More: Smartglass and Artificial Intelligence

Physical Applications

DeepMind’s new research, published in Nature Physics, could help deepen our insights about the nature of glass and our understanding of the structural changes that occur near the glass transition. Speaking practically, these findings may help us answer questions about the mechanical constraints of glass (e.g. where a glass will break).

Further, this research could inform new insights in a diverse range of fields – from medicine to computer science. Graph networks are being applied to a variety of physical systems in contexts including traffic, crowd simulations, and cosmology. Going beyond quantitative predictions, DeepMind’s findings illustrate that machine learning can also be used to gain qualitative understanding of physical systems. The best part? DeepMind has made their models and trained networks available in open source in hopes that machine learning systems may eventually assist researchers in deriving fundamental physical theories, augmenting (instead of replacing) human understanding.

Read More: 10 Interesting Glass Facts

All graphics property of DeepMind.