Hacker News Reader

Comment by mellosouls | Hacker News Reader

Repeating my question from the other thread:

If IBM can cast doubt by describing a 2.5 day classical run, why not just do that instead, and blow the claim out of the water?

There were a couple of responses (thanks) but I still don't understand - it looks like it's not as easy as they imply.

View on HN

Such a demonstration would prove nothing of importance, partly because the issue is not how fast a particular run takes, but how running time grows with the size of the problem. More significantly, the issue is whether the paper's argument for its original claim is sound, and that would be best addressed by showing where it goes wrong. For a brief and clear explanation, read this excellent post:

https://news.ycombinator.com/item?id=21334025

If you don't follow what is being said there, reposting your question will not help.

I'm sure you didn't intend snark, but no need for the tone - that's a new post and not a response to my original question.

Thank you for the link to a potential explanation.

My apologies for the tone. I should have said that the best route to understanding the issue is probably via the linked post - it worked for me.

Thanks no worries

I stand corrected by Scott Aaronson! [1] One of the things he says is that it would be useful to run IBM's algorithm, not to show that it can be done, but to verify Google's 53-qubit and depth-20 result, as, so far, Google has only been able to present indirect evidence that they are getting the correct result at this size of problem.

[1] https://www.scottaaronson.com/blog/?p=4372

Yes, thanks, I've since read that - excellent and thorough article.

They didn't say 2.5 days on a laptop. They probably mean on a bunch of datacenters together. That is expensive.

Google implied their supremacy over all of the world computers combined together.

They are talking about 2.5 days on the world's most powerful super computer, which Google had previously claimed would take 10,000 years to complete the task.

These calculations of 2.5 day and 1000 years, they're calculations based on specific parameters, right? Could they just prove it on a smaller version of the problem instead? Can they do a version of the problem which takes maybe 1m on with IBMs algorithm and 1 year with Google's?

Aaronson's blog goes into this. The problem is exponentially difficult to simulate in the number of qibits. Each quibdit doubles the storage required. 25 quibits can be done on a laptop. 50 requires a supercomputer. 53 is just small enough to fit on the biggest supercomputer. 55 doesn't fit.

There's also an alternative algorithm that requires less space, but it takes longer.

The 2.5 day run is for Summit, the world's #1 supercomputer, owned by Oak Ridge National Laboratory. Getting use of the entire machine for 2.5 days may take some doing.

Reply on news.ycombinator.com