Analyzing quantum materials using machine learning

Electrons and their behavior raise attractive questions for quantum physicists. Recent innovations in sources, equipment, and equipment may give researchers access to more information encoded in quantum materials.

But these innovations are producing unprecedented and previously unreadable amounts of data.

“The information content of a single material can quickly exceed the total information content of the Library of Congress, which is about 20 terabytes,” said Kim Eun-a, a professor of physics at the Faculty of Arts and Sciences. At the forefront of quantum material research, we use the power of machine learning to analyze data from quantum material experiments.

An example of 3D X-ray diffraction data that causes a phase transition during cooling. The magenta plot shows the special points associated with the formation of charge density waves revealed by the machine learning algorithm X-TEC.

“The limited capacity of traditional analytical modes (mainly manual) is quickly becoming a major bottleneck,” Kim said.

The group, led by Kim, uses machine learning techniques developed by computer scientists at Cornell University to analyze large amounts of data from the quantum metal Cd2Re2O7, resolve discussions about this particular material, and future machine learning. Set the stage for your home school.

The paper “Interpretable and Unsupervised Machine Learning Utilization to Address Big Data from Modern X-ray Diffraction” was published in Nature on June 9th.

Physicists and computer scientists at Cornell University have collaborated to build an unsupervised machine learning algorithm, XRD Temperature Clustering (X-TEC). Next, researchers applied X-TEC to investigate the major elements of the pyrochlore oxide metal Cd2Re2O7.

X-TEC analyzed 8 terabytes of X-ray data across 15,000 Brillouin zones (uniquely defined cells) in minutes.

Kilian Weinberger, a professor of computer science at Cornell University, said: “We used an unsupervised machine learning algorithm, which is great for transforming high-dimensional data into clusters that make sense to humans. S Bowers College of Computing and Information Science.

Thanks to this analysis, researchers have discovered what is known as the pseudo-goldstone mode and have discovered important insights into the behavior of electrons in materials. They sought to understand how atoms and electrons are arranged in an orderly manner in order to optimize their interactions within the astronomically large “community” of electrons and atoms.

“In complex crystalline materials, unit cells, which are specific structures of multiple atoms, are repeated in a regular arrangement, like in a high-rise condominium,” Kim said. “The relocations we have discovered will take place on the scale of each apartment unit throughout the complex.”

She said it would be difficult to detect this relocation from the outside, as the placement of the units is the same. However, the relocation almost spontaneously breaks the continuous symmetry, resulting in a pseudo-goldstone mode.

“The existence of pseudo-goldstone mode can reveal the secret symmetry of the system, which is otherwise difficult to see,” Kim said. “Our discovery is made possible by X-TEC.”

According to Kim, this discovery is important for three reasons. First, we show that machine learning can be used to analyze large amounts of X-ray powder diffraction (XRD) data. It serves as a prototype for scaling up X-TEC applications. Available to researchers as a software package, X-TEC will be integrated into the Synchrotron as an analytical tool for Advanced Photon Source and Cornell HighEnergy Synchrotron Source.

Second, the discovery resolves the physics debate on Cd2Re2O7.

“As far as we know, this is the first example of goldstone mode detection using XRD,” Kim said. “This atomic-scale insight into complex quantum matter variability is only the first example of answering important scientific questions associated with the discovery of new phases of matter using large amounts of information-rich diffraction data. “

Third, this finding shows what collaboration between physicists and computer scientists can achieve.

“The mathematical internal behavior of machine learning algorithms is often not different from physics models, but it applies to high-dimensional data,” Weinberger said. “It’s a lot of fun working with physicists because physicists are so good at modeling the natural world. When it comes to data modeling, they really set foot on the ground.”

Co-authors include Geoff Pleiss, MS ’18, Ph.D. ’20; Jordan Benderley, MS ’17, Ph.D. ’19; Krishnanand Mallayya, a postdoctoral fellow in the Atomic and Solid-State Physics Laboratory. Michael Matty, a PhD candidate in physics. The study was conducted in collaboration with a colleague at Argonne National Laboratory.

This research was supported by a grant from the National Science Foundation and a grant from the Department of Energy.

Kate Blackwood is a writer in the Faculty of Arts and Sciences.

Leave a Comment