D-Wave’s Advantage2 quantum computer chip
Kent Kallberg/D-Wave
Quantum computing firm D-Wave says its devices can solve problems that would be virtually impossible for classical computers – but two separate research groups have now cast doubt on the claim.
As reported by New Scientist last year, D-Wave published a pre-print claiming that its Advantage quantum computers could calculate transverse field Ising model problems – a quantum version of a mathematical approximation of how matter behaves when changing states, such as from a liquid to a gas – that would be impractically difficult to solve on a traditional computer.
That article has now passed peer review and been published in the journal Science on 12 March, but at the same time, researchers working on classical algorithms have shown that these problems are readily accessible to ordinary machines.
Dries Sels at New York University and his colleagues say that they have performed similar calculations on a normal laptop in just two hours, using a field of mathematics called tensor networks. These networks essentially reduce the amount of data a simulation requires, drastically cutting the computational power required to run it.
Andrew King at D-Wave says that this does nothing to change the company’s original claim. “They didn’t do all the problems that we did, they didn’t do all the sizes we did, they didn’t do all the observables we did, and they didn’t do all the simulation tests we did,” says King. “So it’s a huge advance, these are great researchers… but it’s not something that refutes our supremacy claim.”
King says that after hearing about the Sels paper, he decided to run larger calculations involving up to 3200 qubits – quantum bits, the building blocks of quantum computers – well beyond the 54 simulated by Sels. He says this further demonstrates quantum supremacy, although the results are not yet published.
Sels calls this response “a bit petty”, saying his tensor approach could easily scale further. The time to run the algorithm scales linearly in proportion to the size of the problem, he says, so there is no need to test larger problems. “If that would really make them [D-Wave] super-happy, and then they would say ‘OK, you guys did it’, we could do it,” says Sels. “I don’t plan to. I don’t see the point.”
Separately, Linda Mauron and Giuseppe Carleo at EPFL in Lausanne, Switzerland, say the transverse field Ising model problems can be solved either without the need for quantum entanglement – a key source of a quantum computer’s supposed benefits – or by simulating a minimal amount of entanglement with an ordinary computer.
Carleo says the pair rushed to publish their paper to coincide with D-Wave’s Science publication, and he admits that it only focuses on one type of problem tackled by the company and does not reach the same scale. D-Wave’s paper suggests that such a computation would take as long as 200 years on a powerful supercomputer, says Carleo, but he and Mauron did it in three days using just four graphics processing units (GPUs) – a fairly modest amount of computation. That said, he says that within a week, it should be possible to exceed the size of problems solved by D-Wave.
“The lesson to be learned is that if you say, ‘this is beyond classical simulation’, then there will be a classical simulation that will do it,” says Carleo. “My suggestion, when they write these papers, is to avoid these claims, because they don’t need them.”
In response, a D-Wave spokesperson dismissed these calculations. “In our paper, we found simulations of this type to be too easy to make any strong claims about,” says the spokesperson. “While this paper does appear to be an advance, it does not challenge our claims whatsoever of beyond-classical quantum simulation.”
If the results are confirmed to have been overturned, it would not be the first time that quantum computers have been touted as unbeatable, only to be proved otherwise. In 2019 Google claimed that its Sycamore quantum computer could perform calculations that would take even the world’s most powerful classical supercomputer 10,000 years to complete. But in 2022 researchers used 512 GPUs to complete the task in around 15 hours, and in 2024 another team completed the same task in 14.22 seconds. Those classical speed-ups also relied on tensor networks.
Aleks Kissinger at the University of Oxford says that D-Wave was one of the first start-ups working on quantum computing, offering what it called the first commercially available quantum computer as early as 2011. But the company had been plagued in its early days by experts’ doubts about whether its computers were truly quantum, or just unusual classical machines that excelled at certain optimisation problems.
Questions about D-Wave’s quantumness have been more or less put to bed at this point, but it’s yet to be seen whether its devices can really solve problems that are impossible for regular machines. “I think, broadly, these days, they seem to be more respected than maybe in the old days when they were making these big claims and keeping all the details a bit under the hood,” says Kissinger. “These days you can see quite a lot of detail about what their devices actually do.”
Topics:
