Sampling bosons

• Post by: admin
• January 9, 2013
• No Comment

If ever there was a paper the linear optics community got as excited about as the now famous KLM paper, it was Aaronson and Arkhipov’s “The computational complexity of linear optics“. Fast forward two years and we have just published a first experimental implementation of the BosonSampling task introduced by the two ‘As’ in Science.

This work, and a similar one by our friends and competitors in Oxford, has attracted quite a lot of attention in the science media. Here’s a (probably incomplete) list of articles about it:

Science, “New form of quantum computation promises showdown with ordinary computers”

Scientific American, “New machine bridges classical and quantum computing”

New Scientist, “Victorian counting device gets speedy quantum makeover”

arstechnica, “Can quantum measurements beat classical computers?”

Physicsworld, “‘Boson sampling’ offers shortcut to quantum computing”

photonics.com, “Rise of the BosonSampling computer”

IEEE spectrum, “New machine puts quantum computers’ utility to the test”

Physorg, “At the solstice: Shining light on quantum computers”

ABC Science, “Proving the need for quantum computers”

But there’s more. Since Andrew presented our preliminary results at last year’s QCMC conference, two other groups in Vienna and Rome also raced to get their results out and all four manuscripts appeared within a day on the arXiv.

Since the titles of our papers don’t offer much of an explanation of the differences between the results, an explanation might be in order. Let’s talk about the similarities first. All of our experiments utilized down-conversion photons sent through some sort of linear optical network. We all observed three-photon interference which is the minimum test of the BosonSampling idea. The team in Oxford also measured four-photon interference patterns, albeit in a limited sense, where instead of four photons being sent into four distinct optical modes, they simply used the sporadic double-pair emissions from a single downconverter.

One difference is that the groups in Oxford, Italy and Vienna realized their optical circuits via integrated waveguides, while we did it in a three-port fiber beamsplitter (with the polarization degree of freedom giving us a six-by-six network). The waveguides provide a stable network, they are however quite lossy which is why we probably have the best quality three-photon data. Another difference is that while these circuits are in principle tunable via thermal heaters, the circuits were actually fixed in the respective experiments. Our circuit can easily be tuned over a large range of interesting unitaries.

An aspect which sets our work apart, and which in my opinion important for testing the validity of BosonSampling, is that we used an different method of characterizing our photonic network. Instead of using two-photon interference for this characterization, which rests on the same assumption as BosonSampling itself and thus does not allow independent verification of predicted three-photon amplitudes, we used a simple classical method for characterizing unitary circuits we recently developed.