thank you. Please listen to this article using the above players. ✖
Want to listen to this article for free?
Complete the form below to unlock access to all audio articles.
At the forefront of discoveries that cutting-edge scientific questions tackle, there is often not much data. Conversely, successful machine learning (ML) tends to rely on large, high-quality datasets for training. So how can researchers effectively utilize AI to support research? Published in Physical Review Researchscientists work with ML to explain an approach to tackling complex questions in condensed material physics. These methods address difficult problems that could not be solved previously through physicist simulations and ML algorithms alone.
Researchers were interested in frustrating magnets, or magnetic materials whose competing interactions lead to exotic magnetic properties. Studying these materials has helped to promote understanding of quantum computing and shed light on quantum gravity. However, it is very difficult to simulate a frustrating magnet due to the constraints that arise from the way magnetic ions interact.
Here, the Japanese, French and German teams were interested in how the properties of certain types of magnetic materials change as they cool towards absolute zero. Their attention focused on a particular phase called “spin fluid.” This spin liquid freezes to different types of magnetic states just as liquid water freezes to ice. However, when it came to identifying the state, they were unable to understand the results of the simulation.
“Recently, physicists are excited by a kind of quantum spin fluid that helps them understand fault-resistant quantum computers,” explained Professor Nic Shannon, director of the theory of quantum matter units at the Okinawa Institute of Science and Technology (OIST) and co-author of the study. “In 2020, we realized that this spin fluid can occur naturally in a class of magnetic materials known as the “respiratory pillow croix.” However, at low temperatures, I couldn't understand what happened to the spin fluid. ”
OIST researchers collaborated with ML experts at LMU Munich, who developed an ML algorithm that could classify traditional magnetic orders.
“Our method is very interpretable. This means that humans are easy to decipher the decision-making process and do not rely on previous training of the model. This makes it more suitable for applications where data is limited compared to other forms of machine learning. “We never applied it to spin liquids before teaming up with OIST, so we were excited to see if it would help us gain insight into such challenging physics problems that have failed.”
To model the cooling of the spin liquid, the team used a computational technique called Monte Carlo simulation. By running the simulation data via the ML algorithm and processing the results, researchers were able to see patterns appearing from within the ML output. These results were used to reversely run the Monte Carlo simulation, seeding the simulation at low temperatures with patterns found in ML, and heating previously unknown phases to simulate the transition in the opposite direction. These new simulations confirmed the characteristics of this phase and provided new understandings in this field of quantum research.
“What's interesting is that neither humans nor machines alone were able to solve this problem. It's like our colleagues are working together, and the algorithms find what we don't have, and vice versa, and we build together towards this full understanding photo. “It's exciting because there are many complex problems to solve in condensed material physics that can be achieved by combining such human and AI approaches.”
reference: Sadoune N, Liu K, Yan H, Jaubert LDC, Shannon N, Pollet L. Human-Machine Collaboration: The ordering mechanism of RANK-2 spin fluids on the respiratory Pyrochlore lattice. Phys Rev Res. 2025; 7 (3): 033061. doi:10.1103/c6z1-wh6l
This article has been republished from: Note: Materials may have been edited for length and content. For more information, please contact the source quoted. You can access the press release publishing policy here.
