Next week, a commercial company has promised to bring the quantum computer out of laboratories into the server room. But why do we want them, and should we believe the claim?
Quantum computers - if they are ever built - will be fantastically powerful. They will be able to solve complex problems including - worryingly - some like factoring prime numbers, whose very difficulty is the basis of our current cryptographic security systems.
Multiple computers in parallel universes
A quantum computer acts as if it is many systems, all operating at the same time. In quantum mechanics, every state of an isolated system exists simultaneously. "It's like a musical chord," says Professor Seth Lloyd of MIT. "You can hear many notes at the same time." In a quantum computer every single possible input is there at the same time - effectively giving you billions of computers all working in parallel (in parallel universes according to some interpretations of quantum mechanics).
That was the fundamental idea of quantum computing, set out by David Deutsch in 1985. But quantum computers have turned out to be difficult to build because they have to be isolated from the outside world to do their work. Put simply, a quantum computer stops working if you look at it - any noise or interaction with the outside world, and the quantum states "collapse" to one.
This gives quantum computer scientists two parallel problems. Keeping the quantum computer isolated long enough to do isolated work - a problem that gets harder the more qubits you attempt to build into the system. And reading the quantum states at the end without destroying the result.
Quantum computing groups have taken a variety of different approaches to these problems, including quantum entanglement, trapped ions and tunnelling. Progress has been slow, and in a recent survey of IEEE members most believed quantum computing was 50 years away.
But a new approach emerged recently that some believe is able to sidestep the problems: the "adiabatic" quantum computer.
Doing it differently
Adiabatic quantum computers (AQCs) are ones which embrace change. An AQC is designed to solve one particular problem: given a set of inputs, it settles to a ground state that solves the problem for those inputs. Far from causing trouble, thermal noise helps, says D-Wave's chief technical officer, Geordie Rose.
The drawback is that AQCs only solve the one problem they are designed for - they are specialised hardware accelerators, not general purpose computers. D-Wave has chosen a particular hard problem for its system, and promises a conventional computer that will act as a front end, translating other "real" problems to match the one the quantum computer can solve.
D-Wave's system solves the 2D Ising model, derived from the theory of magnetism. This is an NP-complete problem, which means it can't be solved rapidly by conventional means - and it also means that any other NP-complete problem can be mapped to it - a class which includes well-known problems such as the "travelling salesman" and lucrative ones such as protein folding and financial optimisation.
Will it work?
D-Wave's computer only has 16 qubits, so it won't go beyond what a conventional system can do, but according to Geordie Rose's blog, it is made with straightforward lithography. It is cooled to within five-thousandths of a degree of absolute zero, but this again is "dependable" technology, he says.
Expanding the number of qubits should be feasible, and one can even imagine a quantum computer available online or as a server in a very well-funded computer room. But the question remains - will it work?
"Adiabatic quantum computing, properly understood, is real quantum computing," says Professor Andrew Steane of Oxford University's Centre for Quantum Computing. "It is fully computationally equivalent to other forms of quantum computing, as long as it can be implemented in a general-purpose way with the right degree of control."
But that is the rub. The qubits are allowed to interact with the outside world, but their interactions have to be stable, says Steane. Going adiabatic may help in some ways, but he believes it doesn't amount to a free lunch: "I doubt that this computing method is substantially easier to achieve than any other."
He sounds a warning note about "NP complete" problems. "The current status of quantum computer science is that we have no very efficient quantum algorithm for any NP complete problem." Quantum computers don't bring in any better algorithms - they work by brute force. "The quantum speed-up only becomes useful once the problem size is quite large," says Steane. "Say, one million quantum operations (logic gates)."
Can we use it?
In the end, this is a computer. Like any other invention in IT, there has to be a hand-off between science and technology, between "experiment" and "computation". Scientific experiments ask repeated questions for which we believe we know the answer. We want to understand the process. Computations do the reverse - they ask questions whose answer we don't know, using a process we understand.
"The distinction between computation and experiment is ultimately a smooth spectrum, in which greater and greater detail of control and knowledge tends to make the machine more like a computer," says Steane. "This is why one would have to look carefully at a given candidate before ruling it one way or the other."
But others are hopeful. Others are more enthusiastic: "The only way we know that could prove if it works, is to build a large scale AQC," says Professor Seth Lloyd of MIT, who co-wrote a key paper describing an AQC architecture.
It might not make money for D-Wave's investors, and it might not find its way into a server room any time soon. But it should give some interesting results for the scientists.