Slashdot Mirror
Quantum Computer Works Better Shut Off
waimate writes "A New Scientist article relates how its possible to get answers from a quantum computer even when your program isn't running." From the article: "With the right set-up, the theory suggested, the computer would sometimes get an answer out of the computer even though the program did not run. And now researchers from the University of Illinois at Urbana-Champaign have improved on the original design and built a non-running quantum computer that really works."
Maybe we should look at what the definition of running is for a Quantum computer. Once it is assembled is it at that point running? Are applications an add on for functions pertaining to specified data?
I read slashdot and the linked article, I think ridiculous things and see voodoo. I read the Nature (very honorable science journal) article, there's no voodoo and everything is straightforward. BTW, I'm a physics graduate student. I wish slashdot and other sites didn't present serious science with such careless descriptions. I give slashdot a F- on this.
I'll admidt I really don't understand what the article is talking about,
Neither did whoever wrote the article.
It works like this:
1) Define some classical terms, like "running" and "actually"
2) Apply them incorrectly to quantum situations
3) ob. ????
4) Profit!
The components of the photon wavefunction that are "not actually running the program" become entangled with the components of the photon wavefunction that "are actully running the program", and therefore they carry information regarding the state of those components.
If we think about this in classical terms, where we incorrectly and falsely imagine that each component of the wavefunction represents a classical trajectory through the apparatus, we could incorrectly and falsely say that photons that have not followed classical trajectories through the part of the apparatus that does the quantum computation have not run the program.
But the clear contradiction of that statement makes the slippery bullshit marketing-speak of the article clear: of course a photon that has followed any classical trajectory whatsoever has not run the quantum program. And to claim that "a photon whose wavefunction is entangled with the program has not run the program" too obviously has the same epistemological and moral status as giving away "free" products that only require a "small" processing fee to claim.
One is motivated to ask, "Why doesn't entanglement with the program state count as 'really' running the program? What is this 'real' thing you keep talking about?" Admittedly, entangling things in this way is a different way of running the program, and is really rather clever, but to promote the results in this way is just attention-grabbing marketing, unworthy of the name of science.
This kind of abuse of language is similar to that of the "quantum teleportation" folks, whose deliberately misleading claims often make it sound like something other than the ontologically-problematic quantum state is being "teleported."
Blasphemy is a human right. Blasphemophobia kills.
No problem. The concept of "counterfactual computation" the article refers to is based around the ideas of the Elitzur-Vaidman bomb testing problem. Imagine you have a computer with an on/off switch. If it's a quantum computer, then it's possible to put the switch in a superposition of both "off" and "on". If you try to read the switch, you'll always see "off" or "on"; however, it's possible to run an algorithm on the quantum computer that preserves this combination of "off" and "on". Using clever quantum interference effects, it's possible to end up with a situation where the algorithm gives you the right answer -- and yet, when you measure it, the switch is "off" and so the computer wasn't switched on.
One way of thinking about it (if you believe in the many-worlds interpretation of quantum mechanics) is that you can use the results of the algorithm in the "other universes" in which the computer was switched on, even though in our universe, the computer was always switched off.
The contribution in this paper is that the authors propose a new method for taking better advantage of this effect, and have implemented it in the toy problem of searching an unsorted database of 4 elements -- which can be done with only one database query on a quantum computer. Amazingly, it seems that this approach of "non-running" a computer can help protect against quantum decoherence, which is the big enemy of quantum computation.
Second: you clearly don't understand the experiment, so why accuse the authors of 'bullshit marketing-speak'? 'On and Off' are not necesarily classical notions; the method to implement on and off is quite simple -- you just use a beamsplitter.
You also seem to have the impression that just because two photons are entangled, that they somehow talk to one another nonlocally. Well, this does not happen in this experiment because there is only ever a single photon, which has an amplitude to go towards the computer ('on') or not towards it ('off'). There is no way for the amplitude for 'on' to communicate with the amplitude for 'off'.
The point of the experiment is that by having the computer not run you can still gain information about which element of the database is the marked element. This effect can be enhanced by utilizing the quantum Zeno effect, which they describe in the Nature article and the supplementary material -- though they don't seem to perform the experiment with the Zeno enhancement, which is a shame.
As for teleportation: you need words to describe the situation you produce. I think the word teleportation is a perfectly good word to describe the effect of transferring the quantum state of a photon instantaneously. I think it is wrong to assume the Star Trek definition of transportation is the correct one. If you are confused by what experiments are actually showing: read the article or ask a physicist.
Disclaimer: IAAQIS (I Am A Quantum Information Scientist).
The actual journal article that New Scientist is referring to was just published: Nature 439 949. I'm not sure if that link will work if you're not at an institution that has a subscription, but you'll probably at least get to see an abstract.
A few bits of background: New Scientist's coverage of quantum information is sometimes horrible. Therefore, it's not surprising that the New Scientist article makes no sense but contains lots of exciting fluff. That said, these guys do have something interesting.
I skimmed through the Nature article, and it looks interesting. It's especially nice that they have an experimental implementation. Nonetheless, the bit about the quantum computer being "off" is just silly.
Here's a summary of how it works, stripped of some hyperbole and converted into something more like plain english (note: qubit means quantum bit).
(1) Create a "control qubit" and some output qubits, with the control qubit initially set to 0, which we will take to mean off.
(2) "Rotate" the control qubit into a superposition of 1 (on) and 0 (off), with most of the "amplitude" being for the 0 state (the qubit is mostly off)
(3) Apply whatever algorithm to the data and the output qubits, conditional on the control bit being on. (Note: we don't actually measure anything here--this is entirely a unitary operation).
(4) Perform a weak measurement on the output qubits, which has the effect of reducing the amplitude of the output qubits being in something other than their initial state (which can only happen if the control qubit was on and the algorithm was applied), since the amplitude for that was small to begin with.
(5) Repeat (2) - (4) N times, such that, if the output bits are unmodified after each algorithm application, you end up with the control qubit in the 1 (on) state. Otherwise, you get the 0 (off) state.
(6) Profit!
This is the simple version, in which you only get to learn whether the application of the algorithm to the data gives you the default output or not. There's a more sophisticated version in which you learn more about the data.
There are a few catches here. One is that N has to be reasonably large, or the probability of an "error" in step 4 becomes an issue (by error, I mean that the weak measurement gives us the wrong outcome). Specifically, the probability of an error is 1 - cos^2N (pi / 2N), which scales as O(N^(1-4N)). Fortunately, that is exponential suppression of error, which is pretty good scaling. Another catch is that their particular experimental implementation used a non-scalable encoding. This isn't a major issue, but it means we should wait for an experiment using a scalable encoding before we really break out the champagne.