Advanced Computing in the Age of AI | Wednesday, July 24, 2024

Scientists Use Machine Learning To Prove The Existence of a Rare Phase Of Matter 

At first glance, glass and crystal may look similar, however, when viewed under a microscope, their structures are significantly different. While crystals have perfectly ordered and repeating patterns of atoms, glass has a fluid-like disordered structure. 

In the world of Physics, the glass phase is considered as a special form of matter. On short timescales, glass behaves much like a solid. However, over a longer period of time, glass behaves more like a liquid. It exists right at the intersection of solid and liquid, making their nature elusive to the traditional classification of matter.  

Quantum researchers and scientists have long been puzzled by the transition of glass from a liquid to a special solid form. Perhaps even more elusive has been the phase of matter called Bragg glass, which exhibits the structural characteristics of both ordered and disordered structures. 

The recent advances in the field of artificial intelligence and machine learning offer an opportunity for scientists to unravel long-standing scientific mysteries. AI and ML algorithms can sift through vast datasets to identify complex correlations and discover patterns that have not been possible with traditional analytics methods. 

Scientists from the U.S. The Department of Energy’s (DOE) Argonne National Laboratory along with collaborators from Stanford University and Cornell University have used ML to uncovered experimental evidence to prove the existence of Bragg glass in a material.   

Using large volumes of x-ray scattering data and a new ML data analytics tool developed at Cornell, the scientists were able to investigate the nature of glasses. While the theoretical prediction of Bragg glass phase has been there for over three decades, concrete experimental evidence has been missing, until now. 

“We can collect massive amounts of X-ray data in short periods of time, and analyzing the data manually can make it impossible to see the forest for the trees,” said Ray Osborn, a senior physicist in Argonne’s Materials Science division and an author on the study. ​“With the combination of cutting-edge X-ray and computational technology, we were able to uncover a signature that is unique to the Bragg glass phase.”

For this experiment, the scientists searched for the elusive Bragg glass state in a crystal base on ErTe3, which is known to have a particular long-range order to its structure that scientists refer to as a charge density wave (CDW).

About three decades ago, it was theorized that CDW materials could host Bragg glass state if some “chaos” could be introduced to the otherwise ordered state of the structure. For this experiment, the scientists randomly distributed palladium atoms to the structure to create some disorder.  

(NicoElNino/Shutterstock)

X-ray scattering was performed on the disordered samples and the 3D structural data for each crystal was recorded. Samples of data were collected at temperatures ranging from 30K to 300 K to analyze how the structures change. 

The hundreds of gigabytes of data were then analyzed using machine learning tools, which confirmed that at a certain transition temperature, the samples froze into a state with a significant amount of long-range order, while also displaying the local features. This confirmed the experimental detection of the Bragg glass phase. 

The insights derived from this experiment also highlight the power of AI and ML algorithms for scientific discoveries in the digital era. This discovery can contribute to a better fundamental understanding of phase transitions in matter. It can also help with advances in the field of superconductivity and magnetism. In addition, the results can lead to the development of new materials for various applications. 

Related Items 

Argonne’s EXSCLAIM! Software Can Retrieve Microscopy Images from Science Articles 

Oxford University Study Shows Large Language Models (LLMs) Pose Risk to Science with False Answers 

Cost-effective Fork of GPT-3 Released to Scientists 

 

EnterpriseAI