The Quantum Pontiff

Life-Size Quantum Computers

D-wave systems, whose paracomputer, err, I mean quantum-maybe computer, which sparked quite a bit of controversy earlier this year, is back in the news. This time D-wave is at the big superconducting conference (SC07) being held in Reno, Nevada and is demonstrating a 28-qubit quantum-maybe computer. Paint me an ivory tower skeptic, but I don’t think their system will work as they expect it to. Of course, this being, D-wave, the news article makes for some entertaining reading.

First up we have this beautiful quote from Geordie Rose, D-Wave founder and CTO,

“We have been collaborating with Hartmut Neven, founder of the image-recognition company, Neven Vision, just after Google acquired it last year,” said Rose. “Neven’s original algorithms had to make many compromises on how it did things–since ordinary computers can’t do things the way the brain does. But we believe that our quantum computer algorithms are not all that different from the way the brain solves image-matching problems, so we were able to simplify Neven’s algorithms and get superior results.”

Sweet, not only is D-wave’s quantum-maybe computer going to be a real quantum computer, but it’s going to be a quantum brain! I can see Roger Penrose jumping up and down from here.

Now if comparing a quantum computer to the brain wasn’t biological enough for you, then you only need to read a bit further to find Rose really bringing in the life sciences (the [sic]s are EE Times errors, not Geordy’s, unless he’s decided to become a Merminite):

“It will house enough qbits[sic] to begin solving mathematical problems that are intractable today. D-Wave’s current prototypes are not amenable to scaling up to hundred of qbits[sic], but with the knowledge we’ve gained over the last year, we feel that the last remaining technical obstacles to life-size quantum computers have been removed.”

“Life-size” quantum computers? Sweet next time I want to talk about building a scalable quantum computer, I’m going to use the word “life-sized.” Then maybe I’ll get some NIH funding (or is that the key to venture capital funding?) But seriously, what is the admission that the current prototypes are not amenable to scaling up to hundreds of qubit?

Comments

  1. #1 Domenic
    November 12, 2007

    Clearly, a life-sized quantum computer is just a quantum system.

    You know, like, a thing. In the universe.
    :-P

  2. #2 Tyler DiPietro
    November 12, 2007

    “But we believe that our quantum computer algorithms are not all that different from the way the brain solves image-matching problems, so we were able to simplify Neven’s algorithms and get superior results.”

    Which would be quite interesting to cognitive scientists, who are still trying to figure out the bolded bit. Not only does he have a bunch of new quantum imaging processing algorithms, he’s got some reason to believe that this is how the brain does it!

  3. #3 Chris Granade
    November 13, 2007

    What’s really sad is seeing Google get mixed up with D-Wave. At least some of the press is being a little more critical (the key word being little): http://www.zdnetasia.com/news/hardware/0,39042972,62034291,00.htm

    Then again, they quote Aaronson in that ZDNet Asia article, so that’ll help keep things grounded.

  4. #4 nextquant
    November 13, 2007

    The statement concerning the ‘working relationship’ between Google and D-Wave was apparently removed.

  5. #5 Michael J. Biercuk
    November 13, 2007

    I’ll start with this straightforward statement: I’m not a D-WAVE fan.

    Moving on, I wanted to reiterate a comment I made on the ZDNET story. Geordie Rose, CTO of D-Wave often falls back on the notion that outside validation of their claims is unimportant as only the performance of their system on a problem of interest ultimately matters.

    Unfortunately this retory .

    Analog computers outperform digital computers on a variety of problems. Are they therefore quantum computers? The Pentium 4 is faster than the Pentium III – should we call it a quantum Pentium? D-Wave seems able to snow most of the uninformed press because they’re using superconducting circuits – if only people knew more about RSFQ…

    I wish D-Wave success in building an analog machine which performs classical simulated annealing and hence outperforms digital processors for some set of problems. They can build it, sell it, become wildly successful, whatever. Claiming, however, in an unsubstantiated manner that any gains are due to the fact that they have built a “quantum computer” is not only misleading, but seriously detrimental to the field.

    I wonder – if someone eventually bought time on the system, and it was later shown to in fact only be performing a classical analog computation, would there be grounds for a false-advertising lawsuit?

  6. #6 Michael J. Biercuk
    November 13, 2007

    Sorry – reposting as a sentence got cut – probably selected the text accidentally…

    I’ll start with this straightforward statement: I’m not a D-WAVE fan.

    Moving on, I wanted to reiterate a comment I made on the ZDNET story. Geordie Rose, CTO of D-Wave often falls back on the notion that outside validation of their claims is unimportant as only the performance of their system on a problem of interest ultimately matters.

    Unfortunately this retort does not address the question at hand – is the system truly performing a quantum computation of any kind?

    Analog computers outperform digital computers on a variety of problems. Are they therefore quantum computers? The Pentium 4 is faster than the Pentium III – should we call it a quantum Pentium? D-Wave seems able to snow most of the uninformed press because they’re using superconducting circuits – if only people knew more about RSFQ…

    I wish D-Wave success in building an analog machine which performs classical simulated annealing and hence outperforms digital processors for some set of problems. They can build it, sell it, become wildly successful, whatever. Claiming, however, in an unsubstantiated manner that any gains are due to the fact that they have built a “quantum computer” is not only misleading, but seriously detrimental to the field.

    I wonder – if someone eventually bought time on the system, and it was later shown to in fact only be performing a classical analog computation, would there be grounds for a false-advertising lawsuit?

  7. #7 Ian
    November 13, 2007

    There’s already a life-size quantum computer – it’s called “The Universe”!

  8. #8 Dave Bacon
    November 13, 2007

    Hey Michael, I think my view is close to yours. However I would add that it wouldn’t be surprising that if for very small versions of their computer there aren’t some small quantum effects. Whether these persist and have any effect for a thousand qubit machine: well color me a skeptic.

  9. #9 Geordie
    November 13, 2007

    @Tyler: You mean cognitive scientists like Hartmut, who used to be neuroscientists and became world leaders in object recognition with biologically-inspired pattern matching approaches?

    @Michael & Dave: An argument for why the systems we’ve build to date are adiabatic quantum computers is going to be presented at MIT, NRC & IQC starting on the 20th of November. The basic argument is this: the integrated low frequency noise in our qubits, which we developed a technique for directly measuring, is less than the gap between the ground state and first excited state for all the problems we’ve run so far. In other words the gap persists in the presence of noise–broadening of energy levels is not sufficient to make the groundstate overlap with the first excited state anywhere during evolution. In addition to this (which when you see the presentation is pretty compelling) there are a series of arguments presented related to quantum annealing and what is expected to happen for large (500+) numbers of qubits. Dave: if you want a copy of the slides in advance contact Mohammad, I’m sure he’d appreciate the chance to go over them with you in advance.

  10. #10 Michael J. Biercuk
    November 13, 2007

    Geordie:
    I’d be very pleased to see serious technical results relating to your system. While you may be able to show that the relevant energy scale in your system does not allow excitation contributions from the integrated noise power spectral density of the circuit, can you prove (not in the mathematical sense of course) experimentally that this necessarily implies you are performing quantum simulated annealing rather than classical annealing or something else? What concerns me is that all of the papers your organization has published in, say, PRL, are essentially demonstrating completely classical (QCA-style) operations/interactions. While lovely from a technical perspective (and I do legitimately appreciate the results) they answer no questions about the quantum nature of your system.

    Dave & Geordie: (Indicating the limits of my knowledge) What is the fundamental experimental test which would demonstrate the system is not simply undergoing a classical, incoherent process? If Dave proposes the demonstration experiment would D-Wave perform it in the open and publish the results regardless of outcome?

  11. #11 Michael J. Biercuk
    November 13, 2007

    Geordie: I’ll also ask this –
    “the integrated low frequency noise in our qubits, which we developed a technique for directly measuring, is less than the gap between the ground state and first excited state for all the problems we’ve run so far…broadening of energy levels is not sufficient to make the groundstate overlap with the first excited state anywhere during evolution”

    The level broadening does not produce overlap with the first excited state to what level of suppression? One part in 10, 100, 1000, 10^9? Noise is a stochastic phenomenon – saying that on average the integrated noise is smaller than the gap is not as meaningful as demonstrating quantitatively that no excitation can occur from the ground to first excited state to, say, one part in 10^6 (or whatever value).

  12. #12 lolDwave
    November 13, 2007

    Dont believe the D-wave hype. When they solve a Qubo machine learning problem with billions of 1000 variable training sets, then you have something truly special, until then its, what was the term; hucksterism to a exponential degree?

    also if you patent everything you make, which im sure D-wave does, and the barriers to entering the market are ridiculously high, which i assume they are, what harm is there in giving the acedemics a taste of the goods? why BS your crowd with stupid press releases and then not follow through with UNQUESTIONABLE evidence? cmon D-wave and G.Rose you have no competitors and no-one can take your AQC tech and use them for 100 years. just get on with it and give yourself some credibility for once.

  13. #13 Tyler DiPietro
    November 13, 2007

    “You mean cognitive scientists like Hartmut, who used to be neuroscientists and became world leaders in object recognition with biologically-inspired pattern matching approaches?”

    So Hartmut has some reason to believe that an algorithm operating on quantum principles is “not all that different from how the brain does” pattern matching? I’ll have to look that up. Although I can think of a few cognitive scientists who’d be far more interested in any evidence he has backing that up.

  14. #14 Michael J. Biercuk
    November 13, 2007

    Wow, I stayed away from these posts for a while because they made me get a bit too worked up. Happening again…

    As an experimentalist I have the following question for Geordie, which I post here because I have been told that he removes unflattering/challenging posts from his blog:

    The original AQC paper by Farhi requires that the multiqubit system be placed in a superposition of all possible “clauses” to the satisfiability problem. For example, in 3-SAT the system is initialized as |000> in the X basis, i.e. a superposition of |1> and |0> in Z for each qubit. Is there any evidence, provided by D-Wave, shown by others, etc. that D-Wave has the ability to make a superposition state with a single qubit?

    As far as I understand (Dave, others please correct me) AQC requires quantum coherence in the initial state and throughout the adiabatic evolution. Instead of jumping right to the many-body physics problem of multiple “qubits” with a variety of couplings, what about showing that you can make a superposition state in a single bit and probe its coherence via a Rabi oscillation. Or Ramsey Fringes. In fact, if D-Wave claims that they don’t have such capabilities because it’s outside of their mission statement, are there any labs which would volunteer to measure a D-Wave “qubit,” with D-Wave personnel? My guess, is that like most Nb SQUID-based qubits, without materials and process optimization (e.g. tuning junction critical current densities) any measurable coherence time will be on the order of ns. Better still, my guess is that T1 for these qubits would be similarly short.

  15. #15 Geordie
    November 14, 2007

    Michael: If you think I remove unflattering comments from my blog you obviously haven’t been reading it!

    Alright back to your question, which I’m going to answer with a question: In what basis does decoherence happen in?

  16. #16 nextquant
    November 14, 2007

    Hi Geordie!

    How was your presentation?! Any chance to see the slides and the video soon?

  17. #17 Michael J. Biercuk
    November 14, 2007

    Geordie: I have read it, and I took note of a thread a number of months ago in which two people were debating technical points with you. One had the name (which stood out for humor’s sake) Donald Duck. You removed some of those comments after a while.

    Pick a basis. Decoherence is either longitudinal or transverse. Make a superposition, show it persists in time. If your point is about phase being unimportant or something as such, at least prove that T1 is long.

  18. #18 Robin Blume-Kohout
    November 20, 2007

    Michael, Geordie:

    Maybe it will help the audience if I clarify one thing about decoherence. It doesn’t have to happen in a specific basis. A lot of work assumes — explicitly or implicitly — the existence of a pointer basis, defined so that pointer basis states are unaffected, but superpositions of them don’t. In practice it’s not usually so pretty. Even the T1/T2 picture is a little simplified.

    However, since the challenge here is “Show that you’re doing something quantum”, there’s a pretty easy way to cut the Gordian Knot (I’m so tempted to say “Geordian”… must… resist). If you pick any two noncommuting bases, and show that the states in both of them persist with negligible decoherence, then you’ve got coherence. If you need to replace “negligible” with “fairly small”, then there’s a bunch of caveats… but the point remains that if you can show stability of any basis + its conjugate basis, then you’re home free — without ever needing to ask “What’s the pointer basis?”

    Too bad I’m chiming in 6 days too late…

  19. #19 Dr. Frankenstein
    November 28, 2007

    What if it turns out to be an evil brain!

    Mwaaaaa….ha ha ha!!

    ….