The tug-of-war between quantum computers and classical computers is intensifying.
In just minutes, a special quantum processor, called a quantum annealing processor, solved a complex real-world problem that a classical supercomputer would take millions of years to complete, researchers claim March 12 in Science. And that supercomputer, the team reports, would consume more energy to run the whole computation than the entire globe uses in a year. However, another group of researchers claims to have already found a way for a classical supercomputer to solve a subset of the same problem in just over two hours.
Quantum computers leverage principles of quantum mechanics to potentially offer huge advantages in processing power and speed compared with the classical computers we’re familiar with in our daily lives. This capability theoretically allows quantum computers to tackle problems much faster than classical computers can.
The new, conflicting results follow similar claims made in recent years. The nascent field of quantum computing has been advancing in lockstep with techniques to make supercomputers more efficient, resulting in a closely matched rivalry. While quantum computers have demonstrated the ability to solve truly random problems faster than classical computers, they have yet to come out on top for physical problems relevant to real-world systems.
In the latest match-up, researchers at D-Wave Quantum Inc. in Burnaby, Canada, used a quantum computer equipped with a quantum annealing processor. Annealing processors differ from other, more typical quantum processors and have shown promise in conducting specific tasks. These processors are better equipped to tackle large problems because their quantum bits, or qubits, are coupled to many other qubits instead of just one, as in other types of quantum processors. But they are useful only for specific types of problems, such as optimization problems, and D-Wave’s computers have attracted scientific skepticism in the past.
For the new result, the D-Wave researchers used a quantum annealing processor to simulate quantum dynamics by using arrays of magnetized disordered pieces known as spin glasses. This setup is relevant to materials science, where understanding the evolution of such systems can help in designing new metals.
“This is a simulation of magnetic materials,” says Mohammad Amin, chief scientist at D-Wave. “Magnetic materials are very important in industry and daily life,” appearing in devices such as cell phones, hard drives and specialized medical sensors.
The researchers simulated the evolution of such systems in two, three and infinite dimensions. After trying to solve the problem with approximations on a supercomputer, they concluded that it couldn’t be done within a reasonable timeframe.
“It’s a milestone result in quantum computing,” says Andrew King, a quantum computer scientist at D-Wave. “We’ve demonstrated quantum supremacy for the first time on an actual problem of real interest.”
Physicist Daniel Lidar, director of the quantum computing center at the University of Southern California in Los Angeles, agrees that the D-Wave team hit a milestone. “It’s very impressive work,” says Lidar, who was not involved in either study but works with a D-Wave device. “They really managed to perform quantum simulations on their hardware that are beyond the reach of current classical methods.”
But the claim isn’t without controversy. King and his colleagues posted a preliminary draft of their paper about a year ago on arXiv.org, providing another group of researchers the opportunity to scrutinize the findings.
Quantum computer scientist Joseph Tindall of the Flatiron Institute in New York City and colleagues simulated part of the same problem using a classical computer. They developed a method that repurposed a 40-year-old algorithm called belief propagation, commonly used in artificial intelligence. Their results, submitted to arXiv.org on March 7 but not yet peer-reviewed, claim to be more accurate than the quantum computer’s for certain cases of the two- and three-dimensional systems.
“For the … spin glass problem at hand, our classical approach demonstrably outperforms other reported methods,” the group writes in a draft of their study. “In [two cases] we are also able to reach errors noticeably lower than the quantum annealing approach employed by the D-Wave Advantage2 system.”
The classical simulations focused on only a subset of the D-Wave results, and the two groups are at odds as to whether the classical simulations can reproduce all the abilities of the quantum computer simulations, particularly for the three-dimensional system.
However, the quantum computer indisputably excelled with the infinite-dimensional system. Although not strictly physical, this system is useful for improving artificial intelligence. Simulating it classically would require an entirely different approach compared with the methods used for the two- and three-dimensional systems, Lidar says. Whether that can be done remains an open question.