Modern computers needed just weeks to correctly solve models that took human theoretical physicists six years to figure out.
OKINAWA, Japan — Machine learning, or the ability of AI systems and computers to learn and improve from experiences, has made some incredible leaps and bounds in recent years and is already starting to make its way into various industries, often times completely reinventing what would have been considered impossible a few decades ago. One such example would be the growing popularity and prevalence of self-driving cars. AI systems have also recently made headlines for besting the top-ranked human chess players in the world, or solving a Rubik’s cube in an absurdly insignificant amount of time.
Now, an international study conducted at the Okinawa Institute of Science and Technology Graduate University finds that modern computers can also solve complex scientific problems just as accurately as human theoretical physicists — only much, much faster.
It took such physicists six years to identify unusual magnetic phases within what’s known as a pyrochlore model. But, with the help of a machine, scientists were able to accomplish the same feat in a matter of weeks!
“This feels like a really significant step,” says Professor Nic Shannon, leader of the Theory of Quantum Matter (TQM) Unit at OIST, in a release. “Computers are now able to carry out science in a very meaningful way and tackle problems that have long frustrated scientists.”
Every single atom within a magnet is associated with a tiny magnetic moment, usually called a “spin.” In typical magnets, such as the ones that are in all likelihood stuck on your fridge right now, these spins are ordered so that they all point in a singular direction. It’s this corresponding pattern that results in a strong magnetic field. This same phenomenon applies to solid materials as well, all the atoms in a particular object are ordered in one direction.
However, just like how matter can exist as a solid, a liquid, or a gas, so too can magnetic substances. Researchers in the quantum matter unit focus on especially unusual magnetic phases called “spin liquids.” Within these spin liquids, there are often competing, or “frustrated” interactions between individual spins, in which they constantly fluctuate in direction. This is similar to the disorder seen in liquid phases of matter.
In the future, these phases may prove incredibly useful in quantum computing.
The research team wanted to discover which of these spin liquids were capable of existing in frustrated pyrochlore magnets. To start, they built a diagram illustrating how different phases could occur as spins interacted in various ways when temperatures fluctuated. That diagram was completed in 2017, but actually reading and using the illustration to identify some semblance of rules governing these interactions between spins proved an incredibly difficult, and long, task.
“These magnets are quite literally frustrating,” quips Professor Shannon. “Even the simplest model on a pyrochlore lattice took our team years to solve.”
So, the research team decided to see if computers could help them out.
“To be honest, I was fairly sure that the machine would fail,” Professor Shannon says. “This is the first time I’ve been shocked by a result – I’ve been surprised, I’ve been happy, but never shocked.”
Researchers collaborated with machine learning experts from the University of Munich who had already developed a way to represent spin configurations in a computer. This innovation was then combined with a machine capable of categorizing complex data into different groups.
“This is the first time I’ve been shocked by a result – I’ve been surprised, I’ve been happy, but never shocked.” -Professor Nic Shannon
“The advantage of this type of machine is that unlike other support vector machines, it doesn’t require any prior training and it isn’t a black box – the results can be interpreted. The data are not only classified 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,” says Dr. Ludovic Jaubert, a CNRS researcher at the University of Bordeaux.
The machine was provided with 250,000 spin configuration variations. Remarkably, without being given any information on which phases were present, the machine successfully created an identical replication of the phase diagram.
Most importantly, when the research team looked into how the machine was able to classify all the different types of spin liquid, they discovered that it had calculated the exact mathematical equations representing each phase. A remarkable achievement that would have taken a team of humans years to accomplish, was completed by the machine within a matter of weeks.
“Most of this time was human time, so further speed ups are still possible,” said Prof. Pollet. “Based on what we now know, the machine could solve the problem in a day.”
“We are thrilled by the success of the machine, which could have huge implications for theoretical physics,” added Prof. Shannon. “The next step will be to give the machine an even more difficult problem, that humans haven’t managed to solve yet, and see whether the machine can do better.”
The study is published in Physical Review B.