The new approach, currently at the simulation stage, suggests that a more practical route can be followed to build effective quantum computers. Current methods use bulky and expensive equipment such as nuclear magnetic-resonance imaging systems, and the quantum states used to encode quantum bits, or "qubits,"are maintained at temperatures close to absolute zero.

However, at room temperature, photons exhibit quantum behavior, and a lot of known technology can manipulate them. "The double-slit experiment, where a single photon goes through whichever parallel slit you put a photodetector behind, clearly demonstrates the quantum-mechanical aspects of photons," said Los Alamos National Laboratories researcher Emanuel Knill. "Others thought you needed a new kind of nonlinear optical component to make quantum computers with photons. We have shown that all you need is feedback."

Knill's work was done with another Los Alamos researcher, Raymond Laflamme, and with professor Gerard Milburn of the University of Queensland, St. Lucia, Australia.

Photons can act as the data in quantum computers by virtue of their dual wave/particle nature. The famous double-slit experiment sends a single photon toward two parallel slits and locates a single photodetector behind first one slit and then the other. No matter which slit the photodetector is put behind, it always detects the single photon.

How does the photon "know"which slit to go through? The answer is that it is acting as a wave instead of a particle, and thus goes through both until it is measured by the photodetector. The act of measurement instantaneously localizes the "particle" aspect of the photon — essentially causing it to "condense" behind whichever slit the measurement is made.

For the optical quantum computer blueprint provided by the labs, the phase state — as polarized either vertically or horizontally — works off the ability of photons to represent 1s and 0s. With all quantum bits, the phase of a photon's wave can simultaneously represent both 1 and 0, since its phase can differ depending on the exact moment it is measured. Afterward that is no longer possible; the phase has become fixed as one or the other by the very act of measurement.

"Until our work, it was thought that the only way to get photons to interact with each other was with nonlinear optics, which is very difficult to implement,"said Knill. "Nonlinear media work fine if you send laser beams through them, but if you only send single photons through, essentially nothing happens."

Nonlinear coupling

To provide the necessary nonlinear coupling among qubits, using photons, the team of Knill, Laflamme and Milburn fell back on one of the most useful engineering techniques ever invented — feedback.

By employing feedback from the outputs of the photodetectors, they were able to simulate the effect of nonlinear media without the disadvantages of actually using them. Essentially, the optical components capable of handling single photons were bent to the service of nonlinear couplings through feedback.

"People never thought to use feedback from the result of a photodetector, but that is where our nonlinearity comes from — it was there all along," Knill explained. This technique was not tried because researchers assumed they could not reuse measurements in quantum computations.

"We discovered that you can use feedback, and that you can replace a nonlinear component with it," said Laflamme.

As in all quantum-mechanical systems, the most important principle has been to preserve "coherence" — that is, to make sure that the qubits remain "unobserved" in their nebulous superposition of both 1 and 0 during a calculation. Once a measurement is made of a quantum-mechanical state, the system reverts to a normal digital system and the advantage of quantum computations is lost. That was why it was thought that feedback could not work — because it would destroy the quantum coherence that forms the basis for quantum algorithms.

However, Knill, Laflamme and Milburn have shown that systems that combine qubits with ordinary bits in the feedback loop can simulate nonlinear optical components. "What we do essentially is destroy coherence in one place and manage to indirectly reintroduce it elsewhere — so that only the coherence we don't care about gets lost in the measurement," said Knill.

The basic idea is that the original qubits to be used in a calculation can be prepared ahead of time by entangling them with what the researchers call "helper" qubits. Entangling ensures that the helper bits maintain the same state as the originals, even after they have gone through a quantum calculation. The helper qubits can then be independently processed with standard optical components, and after the calculation, they can be measured without destroying the coherence of the originals.

Additional errors

The results of measuring the helper qubits are introduced into the feedback loop, which then simulates a nonlinear optical component for a single photon. There is a price for the destroyed coherence of the helper bits, however. According to the researchers, the labs' quantum computer blueprint will make more errors than the already error-prone quantum computers designed elsewhere. To compensate, the team carefully architected their design to use built-in error correction in two subsequent stages.

"The most important discovery in quantum computing in the last five years has been quantum error correction," said Laflamme. "Using quantum error correction, we can mitigate the effect of the errors we introduce with our measurements."

The resulting architecture uses three distinct stages. In stage one, helper photons are generated by entanglement and teleported to a circuit running in parallel with the main calculation. Measurement of the helper bits, after the main calculation, is then introduced into the feedback loop to simulate the effect of a nonlinear coupling between two photons.

"We know when it succeeds by measuring the helper qubit. If the outcome is good, then we go on with whatever else we are going to do in the calculations, but if it fails then we forget about what we just did and start over," said Knill.

But calculations made in this way are successful only with a quantum probability of 1/4, which necessitates the second stage of the architecture.

In stage two, the success probability of stage one can be tuned arbitrarily close to 1. Unfortunately, however, the computing resources needed to achieve 100 percent accuracy can grow exponentially. To solve this problem, the researchers used a third error-correction stage drawing on the recent work of other scientists.

By freely providing the blueprint to the research community, they hope to interest engineers in setting up real-world experiments.