AI beats theoretical physicists at their own game
Over the past few decades, machine learning has revolutionized many sectors of society, with machines learning to drive cars, identify tumors and play chess, often surpassing their human counterparts.
Now, a team of scientists based at the Okinawa Institute of Science and Technology Graduate University (OIST), the University of Munich and the CNRS at the University of Bordeaux have shown that machines can also beat theoretical physicists at their own game, solving complex problems as precise as scientists, but considerably faster.
In the study, a machine learned to identify unusual magnetic phases in a pattern of pyrochlore – a naturally occurring mineral with a tetrahedral lattice structure. Remarkably, when using the machine, fixing the problem only took a few weeks, whereas before, OIST scientists needed six years.
“It looks like a really big step,” said Professor Nic Shannon, who leads the Theory of Quantum Matter (TQM) unit at OIST. “Computers are now able to carry out scientific activities in very meaningful ways and solve problems that have long frustrated scientists.”
The source of frustration
In all magnets, each atom is associated with a small magnetic moment – also called “spin”. In conventional magnets, like those that stick to refrigerators, all the spins are ordered to point in the same direction, creating a strong magnetic field. This order is like the order of atoms in a solid material.
But just as matter can exist in different phases – solid, liquid, and gas – so can magnetic substances. The TQM unit is interested in more unusual magnetic phases called “spin liquids”, which could have applications in quantum computing. In spin liquids, there are competing or “frustrated” interactions between spins, so that instead of ordering themselves, the spins continually fluctuate in direction – similar to the disorder seen in the liquid phases of matter.
Previously, the TQM unit set out to establish what different types of spin liquid might exist in frustrated pyrochlore magnets. They constructed a phase diagram, which showed how different phases could occur as spins interacted in different ways as temperature changed, with their findings published in Physical examination X in 2017.
But reconstructing the phase diagram and identifying the rules governing the interactions between spins in each phase was an arduous process.
“These magnets are literally frustrating,” joked Professor Shannon. “Even the simplest model on a pyrochlore array took our team years to solve.”
Enter the machines
With the increasing advances in machine learning, the TQM unit was curious if machines could solve such a complex problem.
“To be honest, I was pretty sure the machine would break down,” Prof Shannon said. “It’s the first time I’ve been shocked by a result. I was surprised, I was happy, but never shocked.”
The OIST scientists teamed up with machine learning experts from the University of Munich, led by Professor Lode Pollet, who had developed a ‘tensor kernel’ – a way of representing spin configurations in a computer. The scientists used the tensor core to equip a “support vector machine”, capable of categorizing complex data into different groups.
“The advantage of this type of machine is that unlike other support vector machines, it requires no prior training and is not a black box, the results are interpretable. Data is not only categorized into groups; you can also interrogate the machine to see how it made its final decision and learn about the distinct properties of each group,” said Dr. Ludovic Jaubert, CNRS researcher at the University of Bordeaux.
The Munich scientists fed the machine with a quarter of a million spin configurations generated by the OIST supercomputer’s simulations of the pyrochlore model. Without any information about the phases present, the machine succeeded in reproducing an identical version of the phase diagram.
Importantly, when the scientists deciphered the “decision function” the machine had built to classify different types of spin liquid, they discovered that the computer had also independently determined the exact mathematical equations that illustrated each phase – the whole process taking a matter of weeks.
“Most of this time was human time, so other accelerations are still possible,” Prof Pollet said. “Based on what we know now, the machine could fix the problem within a day.”
“We are delighted with the success of the machine, which could have huge implications for theoretical physics,” added Professor Shannon. “The next step will be to give the machine an even harder problem, which humans have yet to solve, and see if the machine can do better.”
Greitemann et al. (2019) Identifying Emergent Stress and Hidden Order in Frustrated Magnets Using Tensor Core Machine Learning Methods. Physical examination B. DOI: https://doi.org/10.1103/PhysRevB.100.174408
This article was republished from the following materials. Note: Material may have been edited for length and content. For more information, please contact the quoted source.