“The ball is in D-Wave’s court”

Google and American defence company Lockheed Martin paid more than USD 10 million for a quantum computer, although its exact capabilities are unknown. A team headed by ETH professor Matthias Troyer examined the question of how to properly test such devices, creating quite a stir in the process.

troyer  
Professor Matthias Troyer set off an echo heard around the world with his tests of the D-Wave quantum computer. (Photo: ETH Zurich)

Experts debated the pre-publication of Matthias Troyer and his co-authors’ work on the arXiv document server. This made press headlines and was met with fierce criticism from the Canadian company D-Wave months before the scientific journal Science published the paper, which is now online. The international team of researchers tested the quantum computer purchased from D-Wave by Google and found it to be no faster than a conventional computer (cf. ETH News)

ETH News: Professor Troyer, D-Wave accuses you of having run the wrong tests. Is this a legitimate criticism?
Matthias Troyer: We performed the same tests that the company itself performed in the past to show that its machine is much better than a conventional computer. We selected test problems that fit perfectly with the hardware of the machine. We didn’t see any speedup; however, this may have been because a quantum device might not be better for these tests.

D-Wave claims its machine is built for more complex applications. Were the tasks you performed on the quantum device too simple?
We selected complex but random test problems not derived directly from applications because the machine cannot currently solve any complex application problems. This has to do with its size and the coupling of the quantum bits used to build the machine. The 512 quantum bits in the machine we tested are coupled only with quantum bits in their vicinity. Hard application problems require a lot more quantum bits and, ideally, arbitrary couplings between any of the quantum bits. The ball is now in D-Wave’s court. If the company claims that we performed the wrong tests, it should demonstrate how the machine is able to solve complex problems better than a conventional computer. Our work shows how these kinds of tests need to be conducted in order to make reliable assertions.

Why have companies such as Lockheed Martin and Google paid more than USD 10 million on a device that may not offer any advantages?
This is not really surprising. Some companies spend millions of dollars every day to solve optimisation problems. We encounter these questions, for example, when planning a flight path or improving a portfolio. This is exactly why D-Wave was built. The machine has the potential to be able to solve these types of problems more quickly than a conventional computer. It is much like a mountain landscape in which you are trying to find the deepest valley. If you do this using a traditional machine, you have to climb over the mountain to find a new valley. A quantum device lets you tunnel under the mountain. The question is whether tunnelling is faster than climbing. If the mountains are high and narrow, it might be helpful. However, no one knows what this landscape actually looks like in the case of real-world application problems.

Would you invest in D-Wave?
If I represented a big company with numerous optimisation problems that needed to be solved, I would say that this sort of high-risk investment is reasonable. But no one knows yet whether it will pay off since there are still a number of hurdles.

The manufacturer pitches its machine as the world’s first commercially available quantum computer. Critics initially considered this a nonsensical claim. Has the mood now turned?
The experts were sceptical at first, but the question now is no longer whether it’s bogus. The tests at Lockheed Martin and Google have shown that the machine works and uses quantum mechanics in the process. This is an accomplishment. But can quantum mechanics help solve optimisation problems? This is now the exciting, unanswered question.

So can D-Wave really be described as a quantum computer?
It’s more of a physics experiment, a prototype that solves specific problems using some quantum mechanics. The device is not a universal quantum computer that can do anything. But as a specialised device, it can be referred to as a quantum computer. It’s like the controls in a car, toaster or refrigerator that solve only one specific problem but can also be called computers.

Why have you concerned yourself with D-Wave?
This machine exists, but no one knows what it does and what the technology can do. Instead of speculating about whether it works, it should be tested. This is not easy, however. Our paper shows how to perform proper, sound testing to determine whether a quantum device performs better. We describe what needs to be taken into consideration and where there may be hidden traps. For example, D-Wave is an analogue machine. It requires more hardware and more quantum bits to solve more complex problems. In order to compare two different devices, their hardware must be comparable in size. So, you have to allow a conventional computer more hardware to solve a bigger problem.

The D-Wave computer is the first quantum computer to be marketed by a private company. Why haven’t scientists working in this field in state-funded research labs succeeded in doing this yet?
The company needed more than USD 100 million to build this machine. An academic research group can’t muster this kind of money. When we write research proposals, our projects must be easier to predict in terms of potential success and not exposed to such a high level of risk. D-Wave built this machine although the outcome of the project was uncertain. As a researcher, you wouldn’t take that risk. But if it works, there are tremendous ramifications. It would be a huge breakthrough.

About

Matthias Troyer  has been professor of computational physics at the Institute for Theoretical Physics at ETH Zurich since 2005. His research interests focus on the development of new simulation algorithms for quantum systems and the numerical simulation of quantum phase transitions, strongly correlated fermionic systems, ultracold atomic gases and quantum computers.

Literature reference

Rønnow TF, Troyer M, Wang Z, Job J, Boixo S, Isakov SJ, Wecker D, Martinis JM: Defining and detecting quantum speedup, Science, 19 June 2014, doi: 10.1126/science.1252319

Leave a comment

We are interested in your comments. Please use our form to send your comment to us:

2 Comments

  • Sol Warda 28 June 2014 17:04

    Dr. Troyer seems to be of two minds. On the one hand, he takes part, with eight other scientists, and finds that D-Wave 2 isn't any faster than classical PCs and have the results published in Science magazine. On the other hand, he takes part in a new study led by Dr. Itay Hen of the USC's ISI. Dr. Hen's findings show that D-Wave consistently outperforms the most optimised software available (namely, Alex Selby's software), sometimes by a significant margin! How does Dr. Troyer explain these contradictions? Did he and his colleagues use wrong tests and/or methodology in the paper just published? And what does he think of the results of Dr. Hen? Dr. Troyer and Dr. Ronnow are both listed there as collaborators of Dr. Hen. I would like to know his thoughts on this matter. Here is the latest study of Dr. Hen:
    http://www.isi.edu/sites/default/files/top_level/events/aqc2014/day4-talk4-Hen.pdf

     
    • Matthias Troyer 7 July 2014 09:10

      There is no contradiction at all but the results are fully consistent. The slides by Itay Hen show a 4-10 times advantage of D-Wave 2 compared to a single core of a CPU. If one uses the full CPU, employing all eight cores, this advantage vanishes and the D-Wave machine performs similar to the classical codes (always ignoring the overhead to program the D-Wave device). This is the similar to what we find in our Science paper, where we also see comparable performance. The more important issue reported in the Science paper is that of speedup, referring to how the time to solution increases with problem size. There we did not see any scaling advantage of D-Wave over classical codes and neither did we find that so far Itay Hen's preliminary unpublished results. We need to keep looking!

       
       
     
 
 
Page URL: https://www.ethz.ch/en/news-and-events/eth-news/news/2014/06/d-wave.html
05.07.2015
© 2015 Eidgenössische Technische Hochschule Zürich