Google Claims quantum supremacy

[BWV]

TL;DR Summary

Google claims quantum supremacy

https://www.cnet.com/news/google-reportedly-attains-quantum-supremacy/
Google has reportedly built a quantum computer more powerful than the world's top supercomputers. A Google research paper was temporarily posted online this week, Financial Times reported Friday, and said the quantum computer's processor allowed a calculation to be performed in just over three minutes. That calculation would take 10,000 years on IBM's Summit, the world's most powerful commercial computer, Google reportedly said.

Google researchers are throwing around the term "quantum supremacy" as a result, Financial Times said, because their computer can solve tasks that cannot otherwise be solved. "To our knowledge, this experiment marks the first computation that can only be performed on a quantum processor," the research paper reportedly said.

 

[jedishrfu]

BATTLE of the TITANS for the MINDS of HUMANITY

IBM has a new article in a similar vein:

https://techcrunch.com/2019/09/18/ibm-will-soon-launch-a-53-qubit-quantum-computer/

https://crunchbase.com/organization/ibm continues to push its quantum computing efforts forward and today announced that it will soon make a 53-qubit quantum computer available to clients of its https://www.ibm.com/quantum-computing/network/members/. The new system, which is scheduled to go online in the middle of next month, will be the largest universal quantum computer available for external use yet.

The new machine will be part of IBM’s new Quantum Computation Center in New York State, which the company also announced today. The new center, which is essentially a data center for IBM’s quantum machines, will also feature five 20-qubit machines, but that number will grow to 14 within the next month. IBM promises a 95% service availability for its quantum machines.

IBM notes that the new 53-qubit system introduces a number of new techniques that enable the company to launch larger, more reliable systems for cloud deployments. It features more compact custom electronics for improving scaling and lower error rates, as well as a new processor design.

 

[f95toli]

I don't think you can directly compare the two machines. IBM has had a fairly "generic" 20 qubit processor open to the public for quite a while, but no one has been able to use all 20 qubits at once in an algorithm(in fact, afaik 6 is the practical limit). We don't know how well their new process will work

Google' s machine is quite possibly more specialised but they have succeded in using all 53 qubits at once, albeit with very low fidelity.

 

[jedishrfu]

Right now it’s pretty much the advertising of titans vying for supremacy and IBM has a way of always staying in the game and making loads of money with deep pockets and deep research though I think Google has the same deep financial resources to succeed here.

 

[f95toli]

Sure, IBM and Google are definitely in the lead
However, their approach is very different. IBM has a big team working on this and have actively been trying to get as many users of their machines as possible (starting from next month they will apparently have 14 machines in the cloud). A lot of effort is also going into their software (qiskit).

Google has a much smaller core team and have been focused on a single experiment for the past 2-3 years. Very few outsiders have access to their machine.

 

[nsaspook]

A simple engineer question.

How do you know that the quantum computer got the answer right in the first place? DO we just trust the answer with empirical proofs if classical computing the problem is impossible?

 

[BWV]

If its some kind of massive factoring problem, like 2048 bit RSA Encryption, it would be easy to check

 

[nsaspook]

If its some kind of massive factoring problem, like 2048 bit RSA Encryption, it would be easy to check

True but how many other problems that quantum computers are effective on will have this 'trap-door' factoring property?

 

[anorlunda]

True but how many other problems that quantum computers are effective on

I would love to see a list of such problems. I don't think weather fits. Perhaps some kinds of optimizations, but which kinds? DNA folding?

 

[jedishrfu]

Like any tech we test it with what we know and then test again on what we don’t but check if it’s valid. If it seems reasonable then we let it fly and fix as needed when bug reports come in.

 

[.Scott]

True but how many other problems that quantum computers are effective on will have this 'trap-door' factoring property?

In general, if the problem has a practical application, then there will be a practical way to see if it works.
If it is a situation where an optimal solution is being sot, then there would be two tests: 1) at least as good as any solution yet found; 2) a determination of whether the quantum algorithm being used would be like to fail by providing a close-to-optimum solution.

 

[.Scott]

I would love to see a list of such problems. I don't think weather fits. Perhaps some kinds of optimizations, but which kinds? DNA folding?

In the long run, there should be chemical and materials design problems. But with only 53 or 72 qubits, there may not be many of those to work on yet.

 

[.Scott]

According to Fortune magazine, the scientific paper describing what Google has done has not been (intentionally) released yet. It is still in review and is likely to remain the for some weeks. It was accidently posted by NASA.

 

[f95toli]

A simple engineer question.

How do you know that the quantum computer got the answer right in the first place? DO we just trust the answer with empirical proofs if classical computing the problem is impossible?

Eventually you don't but that is true for classical computers as well. Benchmarking problems for QC are generally made in such a way that you can gradually increase the size of the problem; for small sizes you can check the solution on a classical computer. You might also know that the solution should have some specific statistical properties.
In the (hopefully near) future we might also get to a point where we can solve quantum chemistry problems where the experimental values are known.

 

[f95toli]

If its some kind of massive factoring problem, like 2048 bit RSA Encryption, it would be easy to check

It is not. It is not a secret that the problem they've been working on for the past couple of years is the "Speckle" problem which is a type of random benchmarking, it is math problem with no real applications.

 

[f95toli]

In the long run, there should be chemical and materials design problems. But with only 53 or 72 qubits, there may not be many of those to work on yet.

Indeed, you probably need ~150-200 qubits to do something useful. You also need decent circuit depth which is the number of operations that processor can do within its coherence time (roughly). The reason for why you can only effectively use about 6 qubits on IBM's current 20 qubit processor is due to the limited effective circuit depth (which in turn depends on many factors, e.g. how the qubits are connected).

 

[kith]

Scott Aaronson has written a blog post on this.

He also comments on the lack of a direct statement from Google: Nature and Science demand that authors don't discuss accepted work until one week before it gets officially published. It probably will be at most a few more weeks until this happens.