Today’s quantum computers have a small computational range, with chips inside smartphones containing billions of transistors, while the most powerful quantum computers contain hundreds of quantum transistors equivalent. increase. They are also unreliable. If you do the same calculation over and over again, it’s likely that you’ll get a different answer each time.
Quantum computers, however, have an inherent ability to consider many possibilities at once, so they don’t need to be so massive to tackle certain thorny problems of computation. IBM researchers announced Wednesday that they have devised a way to manage unreliability in the following ways: You’ll get reliable and useful answers.
“What IBM has shown here is a truly amazing and important step in the direction of moving toward full-fledged quantum algorithm design,” said a computer science professor at the Hebrew University of Jerusalem, who was not involved in the research. Professor Dorit Aharonov said.
In 2019, Google researchers claimed to have achieved “quantum supremacy” (a task that runs much faster on a quantum computer than on a conventional computer), but IBM researchers are more modest. He said he had achieved something new and more useful, albeit with a fancy name.
“We are entering this phase of quantum computing, which I call utility,” said Jay Gambetta, vice president of IBM Quantum. “Age of Practicality”.
A team of IBM scientists working under Dr. Gambetta He described the results in a paper published Wednesday in Nature..
Modern computers are called digital or classical computers because they process bit information that is 1 or 0, on or off. Quantum computers perform computations on quantum bits (qubits) that capture more complex states of information. Just as a thought experiment by physicist Erwin Schrödinger postulated that a cat could be in both dead and alive quantum states, a qubit allows him to be both 1 and 0 at the same time. There is a nature.
This allows quantum computers to perform many computations in a single pass, whereas digital computers must perform each computation individually. By speeding up computation, quantum computers have the potential to solve large and complex problems in fields such as chemistry and materials science that are currently out of reach. Quantum computers may also have a darker side, threatening privacy through algorithms that break the protections used for passwords and encrypted communications.
When Google researchers claimed their supremacy in 2019, the company’s quantum computer said it performed in 3 minutes and 20 seconds a calculation that would take a conventional, state-of-the-art supercomputer about 10,000 years.
However, some other researchers, including IBM, dismissed the claim, saying the problem was man-made. Dr. Aharonov, who is also chief strategy officer at quantum computing firm Qedma, said, “Google’s experiment, it’s impressive, really impressive, but it’s doing things that aren’t interesting for any application.” rice field.
Google’s calculations also turned out to be less impressive than they first appeared. A team of Chinese researchers was able to conduct the experiment. The same computation in just over 5 minutes on a non-quantum supercomputerThis was much earlier than the 10,000 years the Google team estimated.
In the new study, IBM researchers performed another task of interest to physicists. They used a quantum processor with 127 qubits to simulate the behavior of 127 atomic-scale bar magnets (small enough to be governed by the eerie rules of quantum mechanics) in a magnetic field. bottom. It is a simple system known as the Ising model and is often used to study magnetism.
The problem is too complex for even the largest and fastest supercomputers to compute an exact answer.
Quantum computers took less than a thousandth of a second to complete the calculation. Each quantum computation is unreliable, and fluctuations in quantum noise will inevitably intrude and induce errors. But each calculation was so fast that it could be run repeatedly.
In fact, many of the calculations intentionally added noise, making the answers even less reliable. However, by varying the amount of noise, researchers were able to reveal specific characteristics of the noise and its impact at each step of the computation.
“We can amplify the noise very precisely and rerun the same circuit,” says Abhinav Kandala, manager of quantum capabilities and demonstrations at IBM Quantum and author of the Nature paper. “Once we have these results for different noise levels, we can estimate what the results would be without the noise.”
In essence, researchers can subtract the effects of noise from unreliable quantum computations, which they call error mitigation.
“We have to come up with very clever ways to reduce the noise and get around it,” says Dr. Aharonov. “And this is what they do.”
The computer performed a total of 600,000 calculations and converged on an answer for the global magnetization produced by 127 bar magnets.
But how good was the answer?
The IBM team turned to physicists at the University of California, Berkeley for help. The Ising model with 127 bar magnets is too large and has too many possible configurations to fit on a conventional computer, but classical algorithms can produce approximate answers. This is the same technique used in compressing JPEG images to reduce the amount of data by discarding less important data. Reduces file size while preserving most of the image detail.
Michael Zaretel, a physics professor at the University of Berkeley and author of the Nature paper, said when he started working at IBM, he thought his classical algorithms worked better than quantum ones.
“It turned out a little different than we expected,” said Dr. Zaletel.
Certain configurations of the Ising model can be solved exactly, and both classical and quantum algorithms agree on simpler examples. For more complex but solvable instances, quantum and classical algorithms produced different answers, and it was the quantum algorithm that was correct.
So for other cases where quantum and classical calculations diverge and the exact solution is unknown, “there is reason to believe that the quantum result is more accurate,” says a Berkeley graduate student, who is also involved in the study. Sajant Anand, who has done a lot, said. Classic approximation.
It is not clear whether quantum computing is the undisputed winner over the classical methods of the Ising model.
Anand is now introducing error-mitigated versions of classical algorithms that could match or exceed the performance of quantum computing.
“It’s not clear if they achieved quantum supremacy here,” says Dr. Zaletel.
In the longer term, quantum scientists hope that another approach, error correction, will allow them to detect and correct computational mistakes, thereby opening the door to accelerating quantum computers for a variety of applications. ing.
Error correction is already used in conventional computers and data transmission to fix garbled characters. But for quantum computers, error correction is probably years away, requiring better processors capable of handling more qubits.
Scientists at IBM believe that error mitigation is an interim solution that can be used today for increasingly complex problems beyond the Ising model.
“This is one of the simplest natural science problems that exists,” said Dr. Gambetta. “So it’s a good place to start. But the question now is how can we generalize it to tackle more interesting natural science problems?”
These may include characterizing rare materials, accelerating drug discovery, and modeling fusion reactions.