2020-07-16

The most common and familiar materials hide the deepest unsolved mysteries in solid state physics

By yqqlm yqqlm
getInterUrl?uicrIvZQ=f634f325892ceffc3da9cdcb5ce1a788 - The most common and familiar materials hide the deepest unsolved mysteries in solid state physics

For most materials The macroscopic characteristics are derived from the microstructure of the material. Diamonds are hard because the carbon atoms in them are arranged in a repetitive crystal pattern. This arrangement does not change with time. The flowing liquid water can be packed into containers of different shapes, because the liquid has no fixed structure, and their molecules can move randomly.

This difference between solid and liquid seems to be obvious. By cooling, we can turn the liquid into a solid, such as putting a glass of water in the freezer, it will become a solid ice cube. This is an intuitive physical process:the molecules in the fluid are mixed together and gradually become orderly in the process of becoming hard.

However, there is such a strange material, Under the microscope, their molecules are disordered like a liquid; while at the same time, they It is as hard as a solid, and the molecules in it remain almost immobile. This substance is what we are familiar with, glass, which is a mixture of sand and minerals melted at high temperature. When the glass cools from a liquid to a solid, the molecules in the liquid will slow down sharply, but their arrangement will not change significantly as other solids, but still in a disordered state.

getInterUrl?uicrIvZQ=6cf3ca060824583f87bd36fa64f519c0 - The most common and familiar materials hide the deepest unsolved mysteries in solid state physics

glass and Crystal liquid, supercooled liquid and solid. | Image source:DeepMind

Why does this process called glass transition exist and does it correspond to a structure Phase change? For more than 50 years, there has been no answer. Philip Warren Anderson, a pioneer in the field of solid-state physics and a Nobel Prize winner in physics, once said:”The most profound and interesting unsolved mystery in solid-state theory may be glass Properties and glass transition.”

Recently, this problem seems to have made new progress. DeepMind used artificial intelligence to study how the molecules of glass change during solidification. The results show that the artificial neural network used in the new research can use the local structure hidden in the molecule to predict the long-term dynamics of the molecule. Researchers say that although the microstructure of glass looks irregular, they may have more predictable dynamic changes than previously thought.

getInterUrl?uicrIvZQ=ce54946b78e929001c65bc87788b4eea - The most common and familiar materials hide the deepest unsolved mysteries in solid state physics

In this study, scientists considered two key issues, one is how the molecules in the glass are arranged in space, and the other is how slowly they follow Time moves. The concept involved in these two problems is called Dynamic Tendency, which refers to the possibility that they will move in a certain period of time based on the current location of a group of molecules How big is it. This is an ever-changing quantity. It can be obtained by calculating many molecular trajectories with random initial velocities and then averaging the results.

By modeling the molecular dynamics, a computer can and can only generate thousands of glass molecules on a time scale of one trillionth of a second.”Tendency graph”. It is known that molecules in glass move very slowly, so it is impossible for ordinary computers to calculate the propensity of these molecules on a time scale of seconds.

DeepMind researchers believe that it may be very appropriate to use the node graph structure network to simulate the interaction between glass molecules. By training artificial intelligence systems, they predict the trends in the glass without actually running the simulation and try to understand where these trends come from. Graph neural network takes graph as input, each node in the graph represents the position of the molecule in the glass, and the connection between the nodes represents the distance between the molecules.

The researchers first trained the artificial intelligence system:they first created a virtual glass cube composed of more than four thousand molecules, simulating different temperatures Next, the molecules evolved from 400 different positions dynamically, and the propensity in each case was calculated. After the neural network was trained to accurately predict these tendencies, they entered 400 new molecular configurations (glass molecular snapshots) into the trained system. With these configuration snapshots, the neural network predicted the molecular tendency at different temperatures, and its accuracy reached an unprecedented level, 463 times longer than the future predicted by the most advanced machine learning prediction methods.

getInterUrl?uicrIvZQ=f3162ad8872f17085b012393d9e64302 - The most common and familiar materials hide the deepest unsolved mysteries in solid state physics


That is to say, the graphical neural network in the new research can predict the long-term dynamic movement of glass molecules by simply describing the current structure of the molecule, which is the motivation for exploring glass and other materials. Learning provides a powerful new tool.

Using this method, the researchers found some new clues, which is an important concept in the study of phase change in physics-“ Relevant length” is relevant, and the relevant length can be understood as a measure of the distance at which a particle can affect other particles. They found that the graph neural network learned to encode the relevant length:when predicting the tendency at higher temperatures (molecular motion is more like a liquid), the prediction of each node in the network depends on 2-3 adjacent to the node Node information; but at lower temperatures, this number will increase to 5.

That is, as the temperature drops, the network will extract information from larger and larger particle swarms. In other words, when the glass transition occurs, the increase in the relevant length is a sign of the glass transition. It means that although the glass at different temperatures looks the same from a macro point of view, it actually appears differently at the molecular level.

Although at present, the information obtained from the neural network is difficult to be used for quantitative prediction, but it can help us to qualitatively understand these physical systems . The acquisition of the concept of related length by this graphical neural network shows that during the phase transition of glass, there must be some hidden order in the glass structure, and they are not really as disordered as liquid.

Reference source:

https://www.quantamagazine.org/why-is-glass-rigid-signs-of-its -secret-structure-emerge-20200707/

https://deepmind.com/blog/article/Towards-understanding-glasses-with-graph-neural-networks

https://www.quantamagazine.org/ideal-glass-would-explain-why-glass-exists-at-all-20200311/