Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Quantum Supremacy Using a Programmable Superconducting Processor

drive.google.com

Post reply on HN
Print view View original

11–20 of 71 posts

Re: Quantum Supremacy Using a Programmable Superconducting Processor

#11

Post by reikonomusha »

The_rationalist wrote: post #9
acqq wrote: post #7

Reading as somebody who is not in the field, and writing this down to read the responses of those who are more qualified: How I understand it: the "computation" is actually "sampling" the cubits? And then sampling results in the sets of random numbers, which don't have uniform but some specific distribution (specific for quantum effects). Then they claim that such a distribution could not be achieved using classical…

I guess you just clarified why this is bullshit and why quantum supremacy still doesn't exist.

“Quantum supremacy” should be taken as a mathematical term with a mathematical definition in the same way we have definitions for what it means for a problem to be “hard” in the complexity theoretic sense.

Re: Quantum Supremacy Using a Programmable Superconducting Processor

#12

Post by reikonomusha »

The_rationalist wrote: post #5

How does performance compare with an ASIC specialised at generating random numbers? Also what is the error rate?

If you “produce random numbers”, against what distribution should they be generated? Usually RNGs produce uniformly distributed random numbers. The challenge of quantum supremacy is to produce them against another kind of distribution, called the Porter -Thomas distribution. If you run your quantum computer and you get uniformly random output (akin to what toy might do if you build an ASIC), you have a broken quantum computer.

Re: Quantum Supremacy Using a Programmable Superconducting Processor

#13

Post by someguy12342 »

acqq wrote: post #7

Reading as somebody who is not in the field, and writing this down to read the responses of those who are more qualified: How I understand it: the "computation" is actually "sampling" the cubits? And then sampling results in the sets of random numbers, which don't have uniform but some specific distribution (specific for quantum effects). Then they claim that such a distribution could not be achieved using classical…

No, the distribution is thought to be hard for any classical algorithm (this is still being worked on - there are still ongoing theory development for showing it is harder and harder), and by extension, any classical process.

Yes, the problem is very contrived - the distribution being sampled is given by the distribution you get from sampling from a randomly sampled quantum circuit. Of course this is easy on a quantum computer because you can just throw on the randomly chosen gates and measure it. I don't think it will be very useful in a application sense. This problem was designed for the very purpose of showing a "quantum supremacy" as soon as possible.

Also, from what I know, there is still a problem in the whole verification aspect for this problem. That is, even if it can sample from the distribution, it is difficult to verify that in fact the correct distribution was sampled. This is in contrast to the problem of factoring, where to verify the computation we just need to multiply the numbers together to check these are indeed the factors. A remedy I have heard of this (I think from Scott Aarsonson) is that we can modify the parameters so that the output is just within the limits of verifiability with a big supercomputer, and that should be good enough to show supremacy was achieved (EDIT: this is precisely what OP's article says it will do). I think this might be bending what is meant by the term "quantum supremacy" too much, but it's a pretty arbitrary point in computational power anyways , so I'm not too concerned.

If you're interested in more about quantum supremacy, these two articles have a lot of information:

1. Quantum Computational Supremacy Aram W. Harror, Ashley Montanaro (2018) https://arxiv.org/pdf/1809.07442.pdf

2. Complexity-Theoretic Foundations of Quantum Supremacy Experiments Scott Aaronson, Lijie Chen (2016) https://arxiv.org/pdf/1612.05903.pdf

Re: Quantum Supremacy Using a Programmable Superconducting Processor

#15

Post by hhs »

lachlan-sneff wrote: post #14

What use will quantum computers have? Can they do things other than factoring large numbers quickly?

There was an article discussing how finance groups like DE Shaw and Two Sigma are investing in quantum computing to get an edge on spotting trends [0]. I wonder how finance groups will learn from this study?

[0]: https://www.ft.com/content/60fded2a-be5a-11e7-b8a3-38a6e068f...

Re: Quantum Supremacy Using a Programmable Superconducting Processor

#17

Post by the8472 »

someguy12342 wrote: post #13
acqq wrote: post #7

Reading as somebody who is not in the field, and writing this down to read the responses of those who are more qualified: How I understand it: the "computation" is actually "sampling" the cubits? And then sampling results in the sets of random numbers, which don't have uniform but some specific distribution (specific for quantum effects). Then they claim that such a distribution could not be achieved using classical…

No, the distribution is thought to be hard for any classical algorithm (this is still being worked on - there are still ongoing theory development for showing it is harder and harder), and by extension, any classical process. Yes, the problem is very contrived - the distribution being sampled is given by the distribution you get from sampling from a randomly sampled quantum circuit. Of course this is easy on a quantu…

> A remedy I have heard of this (I think from Scott Aarsonson) is that we can modify the parameters so that the output is just within the limits of verifiability with a big supercomputer, and that should be good enough to show supremacy was achieved.

That is what TFA does.

Re: Quantum Supremacy Using a Programmable Superconducting Processor

#18

Post by comnetxr »

acqq wrote: post #7

Reading as somebody who is not in the field, and writing this down to read the responses of those who are more qualified: How I understand it: the "computation" is actually "sampling" the cubits? And then sampling results in the sets of random numbers, which don't have uniform but some specific distribution (specific for quantum effects). Then they claim that such a distribution could not be achieved using classical…

There's additional piece to this that you are missing: while the classical simulation is _hard_ (i.e. increasingly hard for larger and larger number of qubits or circuit depths), it _can_ be computed with enough computational power for small number of qubits <=50 and depths<=20. (To get the classical simulation to 50 qubits was an achievement in itself and required a huge amount of compute power provided by google, without which they'd probably be only able to compute it for 30-40 qubits.) And where the classical computer can compute the result, the quantum computer does get the same answer with some probability (not a particularly high probability in absolute sense, but much higher than you would expect than if the quantum computer didn't work.)

Re: Quantum Supremacy Using a Programmable Superconducting Processor

#19

Post by PeCaN »

lachlan-sneff wrote: post #14

What use will quantum computers have? Can they do things other than factoring large numbers quickly?

They're pretty good at simulating molecules and molecular interactions. I think chemical engineering is probably the killer app for quantum computers.

Re: Quantum Supremacy Using a Programmable Superconducting Processor

#20

Post by someguy12342 »

the8472 wrote: post #17
someguy12342 wrote: post #13

Earlier quoted context omitted.

No, the distribution is thought to be hard for any classical algorithm (this is still being worked on - there are still ongoing theory development for showing it is harder and harder), and by extension, any classical process. Yes, the problem is very contrived - the distribution being sampled is given by the distribution you get from sampling from a randomly sampled quantum circuit. Of course this is easy on a quantu…

> A remedy I have heard of this (I think from Scott Aarsonson) is that we can modify the parameters so that the output is just within the limits of verifiability with a big supercomputer, and that should be good enough to show supremacy was achieved. That is what TFA does.

Sorry, somehow I missed that in the abstract.
Post reply on HN