24
$\begingroup$

Very recently, China claims achieving 'Quantum Supremacy' with the world's fastest Quantum Computer. Their computer was designed as a circuit of lasers, beam splitters and mirrors (see figure bellow) and to solve a very specific problem called Gaussian boson sampling.

The quantum computer from Google, for example, consists of microscopic circuits of superconducting metal that entangle 53 qubits in a complex superposition state. In the case of IBM, their quantum chips are formed by microwave resonators in combination with Josephson superconducting junctions. In case of the IBM, there is already, an open source SDK called Qiskit that expresses quantum computing concepts intuitively and concisely in Python. In both cases, it is possible to program the computers to solve different tasks.

My questions is: how honest is the Chinese claim of quantum supremacy?

$\endgroup$
29
$\begingroup$

"how honest is the Chinese claim of quantum supremacy?"

It's equally (or at least as) honest in comparison to Google's claim.

In a comment to this answer at Quantum Computing Stack Exchange, Craig Gidney (who works at Google and was a co-author on Google's Quantum Supremacy paper), confirmed that the classical computer would have been 2^(20*7/4) = 34359738368 times faster if Google's hardware used CZ gates instead of their obscure gates that no one knows how to program anything with. This would mean that the 10,000 years that they said it would take classical computers to do the calculation that Google did, would become 9 seconds.

Furthermore IBM pointed out that the estimations Google made about the time it would take the classical computer, assumed that the classical computer could only use RAM and no disk at all, and if you allow the classical computer to use disk (which in this case, since they used Oak Ridge National Lab's Summit supercomputer, has petabytes available rather than just terabytes) then even though I/O is slower on disk than RAM you will save so much time by not re-calculating things, that you will get several orders of magnitude further speed-up for the classical computer.

I'm not aware of anything like this for the Chinese experiment, but to be clear: boson sampling is not enough to do "universal" computation, however universal computation is not a requirement for quantum supremacy. Boson sampling allows us to solve #P problems for which no polynomially scaling classical algorithms are yet known.

In any case, "quantum supremacy" is a buzzword and until people are actually using quantum devices to solve useful problems, classical computers are superior because people actually find them valuable enough to purchase.

$\endgroup$
6
  • $\begingroup$ Re: "buzzword", I think that's why it's being dropped in favor of quantum advantage. $\endgroup$ – stafusa Dec 14 '20 at 19:10
  • 2
    $\begingroup$ @stafusa The reason the phrase is being discouraged is much more because of the historical context of the word "supremacy" being associated with racism. My paragraph about it being buzzword would remain the same if it was called "quantum advantage" because truly there is zero computational advantage of using a quantum computer right now, for any real-world computation. I don't deny though, the benefits of pursuing quantum technologies, since if we can improve them, they may truly have advantages over classical technology. $\endgroup$ – Nike Dattani Dec 14 '20 at 19:23
  • $\begingroup$ In reference to the claim that "no one knows how to program anything with [the gate that was used]": Section VII-F of the supplement to the paper explains how to decompose arbitrary two qubit interactions into the gates that were used and this conversion is available in Cirq via the cirq.google.ConvertToSycamoreGates() optimizer. I certainly won't say it's a convenient gate set, but people have used it. $\endgroup$ – Craig Gidney Dec 16 '20 at 18:41
  • $\begingroup$ @CraigGidney That comment is very much appreciated, and I will incorporate it into the next edit of my answer, thanks so much! The decomposition of CZ into fSim gates seems even more complicated than I ever imagined. It seems if I really wanted to solve a small version of a real-world problem on Sycamore with gates I know (like CZ), there would be an overhead due to these extra gates that are needed, which will slow down Sycamore's speed, not by 2^35 but the classical computer also won't be simulating CZ either (it would be solving the problem directly). Real supremacy needs lots more qubits! $\endgroup$ – Nike Dattani Dec 16 '20 at 19:01
  • $\begingroup$ @NikeDattani There is actually a more direct decomposition of arbitrary 2 qubit interactions that doesn't go through the CZ gate, which uses at most 4 Syc gates (instead of the 6 you'd need using the CZ decomposition as an intermediary). I think this is what is done in cirq, though it is not explained in the paper. For comparison, decomposing into CZs you'd need at most 3 operations to do arbitrary interactions. $\endgroup$ – Craig Gidney Dec 16 '20 at 19:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.