In 2019, Google announced that their quantum chip "Sycamore" had solved a task faster than a classical computer for the first time. Chinese researchers have now cracked the problem without quanta on a normal computer.
Promises of technology are unbroken
The promise of technology, at any rate, continues unabated. Research ecosystems on quantum computing are springing up all over the world. Start-ups are springing up like mushrooms, and companies such as Google and IBM are outdoing each other in connecting more qubits on a chip. Only the solution to a practical "problem" has yet to be found.
Because qubits, unlike the bits of an ordinary computer, can not only assume the states 0 and 1, but can also remain in a superposition of these states, a parallel representation of 253 states is possible with the 53 qubits of the Sycamore chip.
«Das Google-Experiment hat getan, was es tun sollte, nämlich dieses Rennen zu starten»
Who will ultimately win and develop a true universal quantum computer, on the other hand, remains to be seen.
Spectrum of Science
We are partners of Spektrum der Wissenschaft and want to make sound information more accessible to you. Follow Spektrum der Wissenschaft if you like the articles
.
In 2019, Google opened the race for the quantum computer. For the first time, the Google Quantum AI Lab team announced in Nature, a quantum chip had solved a specific computational task in 200 seconds that would take the world's best supercomputer 10 000 years. Newspapers around the world ran headlines at the time "Google hangs supercomputer" and "Proof of quantum superiority". "Spektrum" also asked, "The Sputnik moment of quantum physics?".
Now Chinese scientists have done the same calculation in a few hours with normal processors. A real supercomputer, they write in an article already available on the preprint server ArXiv and now appearing in Physical Review Letters, could even solve the task in a few seconds, outperforming Google's quantum chip "Sycamore" outright. Quantum superiority, goodbye?
This new algorithm takes at least some of the shine off Google's claim, Greg Kuperberg, a mathematician at the University of California told "Science": "It's less exciting to be 300 steps away from the summit after all than it is to actually reach the summit." Researchers at IBM had then challenged the claim that a supercomputer takes 10 000 years to calculate. The colleagues had not properly utilised the capacities of the "Summit" supercomputer at Oak Ridge National Laboratories when checking their calculation, they wrote. However, the authors of the paper are also direct competitors of Google employees.
Because the task Sycamore solved in 2019 was also designed to be extremely difficult for a conventional computer, but as easy as possible for a quantum computer. Put simply the test consisted of a completely useless calculation for complex random numbers. The Google researchers had a circuit of coupled qubits, the quantum mechanical equivalent of classical bits, perform many randomly selected arithmetic operations, repeated the sequence millions of times and recorded the results. For comparison, the whole thing was simulated on a conventional supercomputer.
Assuming that all qubits were initially set to 0, the Google researchers made the quantum chip spit out a random sequence several million times, each time making it perform 20 random computing operations. Afterwards, they read out the state of the qubits. Roughly speaking, quantum waves, which initially represent all possible outcomes, sloshed back and forth between the qubits. The interactions between the qubits created interference that amplified some outcomes and cancelled out others. Some outcomes are more likely to occur than others for this reason. How likely a single sequence of numbers is can only be said after countless runs. Eventually, a characteristic probability distribution emerged.
Classical computers that simulate the circuit must painstakingly test every conceivable sequence of computing steps. As the number of qubits increases, the effort required to do this becomes immeasurable. The theoretically predicted limit is about 48 qubits. For a quantum computer, on the other hand, the computing time remains manageable because - once it has undergone the computing operations - it can output a single random result practically at the touch of a button.
The Chinese researchers led by theoretical physicist Pan Zhang have now represented the 2019 task as a large three-dimensional network of so-called tensors. This network consists of 20 layers - one for each computational operation Sycamore went through at the time. And each layer consists of 53 points - one for each of the 53 qubits. Running the simulation was then essentially limited to multiplying all the tensors. The calculation took 15 hours on 512 GPUs and actually delivered the expected probability distribution. On a supercomputer, Zhang and his colleagues write, the calculation would even take only a few dozen seconds - ten billion times faster than the Google team had estimated in 2019.
.
It was to be expected that research on classical computers and the search for better algorithms would not stop either. The important thing now would be to finally find practical applications to demonstrate the quantum advantage. Or as Dominik Hangleiter, a quantum physicist at the University of Maryland, told Science: "The Google experiment did what it was supposed to do, which was to start this race."
