Machine learning is used to analyze quantum materials

Quantum physicists are fascinated by the behavior of electrons.

Machine learning is used to analyze quantum materials

Quantum physicists are fascinated by the behavior of electrons. Recent innovations in instruments, sources, and facilities have allowed researchers to access more information in quantum materials.

However, these research innovations are producing unprecedented--and until now, indecipherable--volumes of data.

Eun-Ah Kim is a professor of Physics in the College of Arts and Sciences. He is at the forefront of quantum materials research and machine learning to analyze quantum material experiment data.

Kim stated that the "limited capacity of the traditional method of analysis--largely manually--is rapidly becoming the critical bottleneck."

Kim's group used machine learning techniques developed with Cornell computer scientists to analyze large amounts of quantum metal Cd2Re2O7 data. This helped settle a dispute about the material and set the stage for future machine-learning aided insights into new phases.

The paper "Harnessing Interpretable, Unsupervised Machine Learning to Address Big Data From Modern Xray Diffraction" was published in Proceedings of National Academy of Sciences on June 9.

Computer scientists and Cornell physicists collaborated to create XRD Temperature Clustering (XTEC), an unsupervised, interpretable machine-learning algorithm. The researchers used X-TEC to study key elements of Cd2Re2O7, a pyrochlore-oxide metal.

X-TEC analyzed 8 terabytes worth of X-ray data spanning 15,000 Brillouin areas (uniquely identified cells) in just minutes.

Professor of computer science at Cornell Ann, Kilian Weinberger said, "We used unsupervised Machine Learning algorithms, which were a perfect fit for translating high-dimensional data into clusters, that make sense to people." S Bowers College of Computing and Information Science.

The researchers gained important insight into the electron behavior of the material through this analysis. They discovered what is now known as the pseudo Goldstone mode. The researchers were interested in understanding how electrons and atoms behave to maximize interaction within an astronomically large "community".

Kim stated that "In complex, crystalline materials, there is a specific structure made up of multiple atoms. The unit cell repeats itself in a regular arrangement, like in high-rise apartment buildings." The repositioning that we found occurs at the scale of each apartment unit and across the entire complex.

She said that the order of the units remains the same so it is hard to see the repositioning from the outside. The repositioning can occur almost automatically, and results in pseudo-Goldstone modes.

Kim stated that pseudo-Goldstone mode could reveal secrets in the system that are difficult to see. "X-TEC enabled our discovery."

Kim stated that this discovery is important for three reasons. It shows machine learning can be used for analyzing voluminous X-ray Powder Diffraction (XRD), data. This prototype will also serve as a model for X-TEC's future applications. X-TEC will be made available to researchers as a program package. It will also be integrated into the synchrotron's analysis tools at the Advanced Photon Source as well as the Cornell High Energy Synchrotron Source.

The discovery also settles the debate about the physics behind Cd2Re2O7.

Kim stated that this was the first time that XRD has been used to detect a Goldstone mode. Kim stated that this atomic-scale insight into fluctuations in complex quantum materials will only be the first instance of answering key scientific questions associated with any discovery of new phases or matter... using information rich voluminous data.

The third is that the discovery demonstrates what collaboration between computer scientists and physicists can achieve.

Weinberger stated that the mathematical inner workings behind machine-learning algorithms often resemble models in physics, but are applied to high-dimensional data. It's a lot of fun to work with physicists, as they are so adept at modeling the natural environment. They are very efficient when it comes to data modeling.