The Chinese Academy of Sciences team says it has developed an algorithm to accomplish a task that “was considered impossible for classical computing.” The researchers claim that the 1 million uncorrelated samples obtained using their method are more accurate than Google’s quantum computer.
Scientists in China are once again challenging Google’s claims to “quantum supremacy” after the research team said it had developed an algorithm to perform a task that “was considered impossible for classical computing.”
The Chinese team said their non-quantum classical computer completed the sampling task “in about 15 hours” with higher accuracy than Google’s model of Sycamore’s real-life quantum computer, which took 200 seconds to complete the same task.
According to the team, the accuracy of 1 million uncorrelated samples created with their method was 0.0037, while the accuracy of Google’s quantum computer model was 0.002.
In a paper that will be submitted to a scientific journal for peer review, scientists at the Institute for Theoretical Physics of the Chinese Academy of Sciences said their algorithm on classical computers completed simulations for Sycamore quantum circuits “in about 15 hours using 512 graphics processing units (GPUs).”
“We propose a new method for the classical solution of this problem by compressing the corresponding tensor network once, [it is] much more efficient than existing methods at obtaining a large number of uncorrelated samples with a given accuracy,” they said.
In October 2019, Google announced that its Sycamore processor was the first to achieve quantum supremacy, completing in three minutes and 20 seconds a task that would have taken the best classic IBM supercomputer Summit 10,000 years. This claim – especially how Google scientists came to the conclusion of “10,000 years” – has been challenged by some researchers.
In Beijing, a team from the Chinese Academy of Sciences argued that “the computation time estimated by Google is based on a specific classical algorithm … and not on a theoretical boundary that applies to all possible algorithms.”
“So, in principle, there may be algorithms that perform much better than the algorithm used by Google, which rejects the claim of quantum superiority,” the team said. “We propose such an algorithm based on the tensor network method.”
Team Leader Zhang Pan, a professor at the institute, told the South China Morning Post that it is currently important in quantum computing to combine classical and quantum computing, which suffer from “noise,” or noise that degrades their accuracy, for real-world applications.
Unlike classical computing, quantum computing is error prone because environmental factors can affect subatomic behavior.
“Our new algorithm and the use of advanced classical computing resources, including over 500 GPUs, is why our device is comparable to Google’s quantum computer in this task of randomly sampling quantum circuits,” he said. “To further speed up the task and improve accuracy, an even better algorithm can be developed and applied to larger and higher-performance equipment such as a classic supercomputer.” In March, the team published an article about their test using 60 GPUs that completed Google’s experiment in “about five days.”
In their latest article published this month, the team stated:
“If our simulation of quantum superiority circuits can be implemented on a modern conventional computer with high efficiency, in principle, the total simulation time can be reduced to a few tens of seconds, which is faster than Google’s hardware experiments.”