Although mainstream quantum computing is still years off, agencies are starting to experiment with D-Wave machines for optimization problems.
While quantum computers are still a ways from becoming a standard component of government IT, that they will eventually get there seems no longer in doubt. D-Wave Systems, a Canadian company, is trying to get ahead of that trend with a U.S. subsidiary specifically targeting sales of its systems to the American government.
Historically, explained Bo Ewald, president of D-Wave Systems U.S., around half of the high-performance computing market has been in the United States, with the dominant customer being the federal government -- mainly the intelligence and classified agencies. He expects the model for D-Wave sales to follow that.
All of the company’s current customers are in the United States. Lockheed Martin bought the first-ever D-Wave System and has already upgraded it a few times. Another was bought by Google and a third by the Los Alamos National Laboratory, a long-time leader in the use of high-performance computers.
D-Wave offers both the computers themselves as well as a set of professional services that customers can use when developing approaches to quantum computing.
“Most of that work is classified today,” Ewald said, “but in the future we expect to work also on unclassified projects such as looking to adapt algorithms for D-Wave systems, and to develop software tools that will make it easier for people to use the machines.”
In addition to D-Wave, IBM and Microsoft also are developing quantum systems – though they are based on different methods.
IBM is trying to develop a so-called universal quantum computer that uses a gate logic approach similar to that of current non-quantum computers and that can be programmed to solve many different kinds of problems. A useful computer along those lines would have to be around 50 to 100 qubits in size, the company believes, and that system is probably still a decade or so away from production.
IBM recently made its five-qubit quantum processor available through the cloud for anyone to experiment with, in the hope of accelerating the understanding and development of applications for its system.
How qubits work
In classical computing a bit can take only two values, either 0 or 1. A quantum bit, or qubit, can be in both states at the same time, and it’s that property that allows quantum computers to operate much more quickly than classical computers.
D-Wave uses an approach that was first posited in the 1950s. A method called “simulated annealing” was proposed to iteratively refine complex calculations so that a probable solution could be reached rapidly. It was first suggested as a possible basis for quantum computing in the late 1990s, around the same time that D-Wave was being formed.
Consider D-Wave’s machine operating in a 3D landscape of mountains and valleys as an analogy, Ewald said. Problems are described in N dimensions, where N is the number of qubits the system contains. When looking for the lowest possible energy solution among all possible variations of mountains and valleys in that landscape, the D-Wave system can searching this landscape thousands of times a second, quickly finding the best range of low-energy solutions.
“It’s a probabilistic machine that can find the best possible answer among a range of solutions and answers, and all the machine’s capacity can be devoted to that,” Ewald said.
Given D-Wave’s approach, its systems are naturally geared to a class of calculations called optimization problems. Finding the best possible route for a traveling salesman to take to reach as many people in a particular period of time, for example, or being able to pick an all-star team of baseball players based on stats for individual strengths and skills and, crucially, how well they are likely to play with and against each other.
A D-Wave system gives a pretty good approximate answer within a fraction of a second to a problem a that would take a classic computer the lifetime of the universe to compute, according to Scott Pakin, a computer scientist at Los Alamos who has over two decades of high-performance computing experience and who is now evaluating D-Wave’s system. “Our current (classical) high-performance machines are subscribed 24x7x365 in helping with the kinds of problems we have, such as simulating nuclear explosions for our weapons research," he said. "Those need incredible computing power, and sometimes it’s better to get a good enough solution very fast than an excellent solution slowly.”
A lot of the current work is in trying to develop programs for problems that take advantage of the D-Wave’s optimization capabilities. Pakin also said he thinks working with a quantum computer will help LANL develop a better approach for its classical computers, such as what code will work best on one processor versus another, given different memory configurations. Having the computer could also boost LANL staff recruiting, he thinks, since many smart people will be attracted to working with the quantum machines.
Even though there’s a “rising tide” of interest in quantum computing generally, Pakin said he believes it will still take a long time for it to get to a useful level of maturity. It’s hard to forecast how large quantum systems need to be to solve the kinds of problems that classical computers now tackle, he said.
There are areas that Ewald thinks are ripe for D-Wave machines to tackle now. The company is working with both Google and NASA Ames to find a way to match this optimized computing capability to machine learning, for example, so that automated machines and robots can better recognize images. It’s also a good fit for attacking the Monte Carlo type of calculations that the financial community now uses to assess various levels of risk in investments.
The company recently introduced a 2,000 qubit machine that greatly increases the potential computing power of its systems. However, Ewald admitted that D-Wave’s quantum systems are still mainly in the research and experimental phase, though the intent now is to work on developing algorithms that can use the machines to look at real-world problems.
There’s definitely been a major increase in interest, Ewald said, and D-Wave is now getting invitations to conferences and events to explain its approach because people can see the challenges that traditional computing techniques will have as Moore’s Law seems to be reaching its end.
“Just a few years ago there was still a lot of skepticism because of the uniqueness of the quantum approach, but now many of the questions are how to craft an application to run on D-Wave machines,” he said. “Many more people now do accept that it’s real.”