Talk:D-Wave Systems

Some earlier versions of this article cited D-Wave's articles to counter balance statements that none of D-wave's has been shared with the physics community. Thus maintaining a NPOV. However, these were citations to the a pre-print server, whichis not what most would consider peer reviewed -- but it does have a referal system to qualify contributors. So citations were removed. However, it is clear that D-Wave has many peer review articles -- see below. Perhaps, these can be worked in beside comments like "which has not been published or shared with the physics community"

Phys. Rev. A 79, 022107 (2009); Int. J. Quant. Inf. 7, 725 (2009); Phys. Rev. Lett. 101, 117003 (2008); Phys. Rev. Lett. 100, 197001 (2008); Phys. Rev. Lett. 100, 130503 (2008); Quantum Information Processing, 7, pp. 193-209 (2008); Phys. Rev. A 78, 012352 (2008); Phys. Rev. Lett. 100, 060503 (2008); Appl. Phys. Lett. 90, 022501 (2007); Physical Review Letters, Volume 98, 177001 (2007); Phys. Rev. Lett. 98, 057004 (2007); Quantum Information Processing 6, pp. 187-195 (2007); Phys. Rev. B 74, 104508 (2006); Phys. Rev. Lett. 96, 047006 (2006); Phys. Rev. A 74, 042318 (2006); Science 309, p. 1704 (2005); Europhys. Lett. 72(6), pp. 880-886 (2005); New J. Phys 7, p. 230 (2005); Phys. Rev. B 72, 020503(R) (2005); Phys. Rev. B 71, 140505(R) (2005); Phys. Rev. B 71, 144501 (2005); Low Temp. Phys. 30, 661 (2004); Phys. Rev. A 70, 032322 (2004); Fizika Nizkikh Temperatur 30, 823 (2004); Phys. Rev. Lett. 92, 017001 (2004); Phys. Rev. B 70, 212513 (2004); Phys. Rev. Lett. 93, 037003 (2004); Phys. Rev. B 71, 024504 (2005); Phys. Rev. B 71, 064503 (2005); Phys. Rev. B 71, 064516 (2005); Europhys. Lett. 65, 844 (2004); Turk J Phys 27, 491 (2003); Phys. Rev. B 68, 014510 (2003); Phys. Rev. B 68, 134514 (2003); JETP Letters 77, 587 (2003); Phys. Rev. B 69, 060501(R) (2004); Phys. Rev. Lett. 91, 097906 (2003); Phys. Rev. Lett. 91, 097904 (2003); Phys. Rev. Lett. 90, 117002 (2003); Phys. Rev. B 68, 144514 (2003); Phys. Rev. B 67, 100508(R) (2003); Phys.Rev. B 67, 155104 (2003); Phys. Rev. Lett. 90, 127901 (2003); Phys. Rev. Lett. 90, 037003 (2003); Phys. Rev. B 66, 214525 (2002); Phys. Rev. B 66, 174515 (2002); Quantum Information Processing 1, 155 (2002); Quantum Information Processing 1, 55 (2002); Physica C 368, 310 (2002); Physica C 372-376 P1, 178 (2002); IEEE Tran. Appl. Supercond. 12, 1877 (2002); Physica B 318, 162 (2002); Low Temp. Phys. 27, 616 (2001); Phys. Rev. Lett. 86, 5369 (2001); Phys. Rev. B 63, 212502 (2001)

Shadesofgrey (talk) 01:22, 10 August 2009 (UTC)

Ndickson did not remove any of D-Wave's criticism (though many of them don't apply), but simply added facts, supported by peer-reviewed and published papers. If D-wave critics have a reason to doubt the correctness of any of the papers please share them with the scientific community. Otherwise stop undoing modifications that introduce uptodate and factual information to the wiki page.

Verbal, do you realise that saying something like "The problem is the lack of independent reliable sources" and at the same time removing references to published papers is a contradiction? You want to avoid anything that disproves your pre-conceived ideas, based on some opinionated blog entries? You are the one who has to defend yourself. Either prove the published papers wrong, or stop keeping this page out-of-date. —Preceding unsigned comment added by 216.13.217.230 (talk) 19:20, 7 June 2010 (UTC)

See WP:PRIMARY. Please propose changes you would like in a neutral manner. Verbal chat 19:40, 7 June 2010 (UTC)

It seems that bashing D-Wave has become an ego thing for some people, but Wikipedia is not a place for that. Let's keep D-Wave's page as factual as possible. Take comments like "I think D-Wave is this" or "I haven't seen any evidence for that" to personal blogs, where such speculations belong.

Ndickson has laid out the general direction. In more concrete terms I suggest the following modifications for the near future:

1) Integrating NDickson's changes: (*) adding a picture of the 128-qubit chip to show this is no "vaporware" and mention that Orion has been outdated. (*) Adding the "New Technology" section to inform Wikipedia readers about D-Wave's latest technology, backed by peer-reviewed papers.

2) Adding a section about the 2009 Google demo. I propose the following text (quite factual):

== 2009 Google Demo ==

On Tuesday, December 08, 2009 at the Neural Information Processing Systems (NIPS) conference, a Google research team lead by Hartmut Neven used D-Wave's processor to train a binary image classifier. The theory behind the classifier is explained in http://arxiv.org/abs/0912.0779, while descriptions of the implementation details appear in http://static.googleusercontent.com/external_content/untrusted_dlcp/www.google.com/en//googleblogs/pdfs/nips_demoreport_120709_research.pdf

3) Adding a link to D-Wave's list of peer-reviewed, published papers, because honestly, removing published papers from this Wikipedia page and at the same time complaining that D-Wave people don't share information with the scientific community is funny and sad at the same time. —Preceding unsigned comment added by 216.13.217.230 (talk) 17:52, 8 June 2010 (UTC)

First problem: The google text you propose would require an independent WP:RS (not google or dwave) to support the text beyond the event happening. There should not be a link to all publications as we don't do that for any other company involved in research. Let's stick to WP:RS and not primary sources. Verbal chat 18:03, 8 June 2010 (UTC)

I'm neither for nor against the above text, but it might be worth re-reading WP:PRIMARY before you dismiss things. e.g.: "Primary sources that have been reliably published may be used in Wikipedia, but only with care, because it is easy to misuse them. Any interpretation of primary source material requires a reliable secondary source for that interpretation. A primary source can be used only to make descriptive statements that can be verified by any educated person without specialist knowledge. For example, an article about a novel may cite passages to describe the plot, but any interpretation needs a secondary source. Do not make analytic, synthetic, interpretive, explanatory, or evaluative claims about material found in a primary source." The above text does not interpret the sources it cites, it states explicitly material in those sources. Secondary sources are much more likely to contain editorialized opinion than fact, but there are many secondary sources that reported on the demonstration with Google, if you'd prefer. —Preceding unsigned comment added by Ndickson (talk • contribs) 18:14, 8 June 2010 (UTC) Ndickson (talk) 18:15, 8 June 2010 (UTC)

That doesn't conflict with what I have said. The second sentence should therefore be removed, leaving the whole thing unsourced. Verbal chat 18:56, 8 June 2010 (UTC)

True, it doesn't directly conflict with what you said here, and I fully agree that the sources the anonymous user above listed for the collaboration with Google are not so good (certainly not peer-reviewed). I also agree that for describing the Google collaboration, there are many secondary sources that could be preferable. However, I'm not aware of any secondary sources on the technology that could be argued to be more reliable than the peer-reviewed (i.e. "reliably published") primary sources. There are more blog entries, such as NextBigFuture, which summarize some of the content of papers, if you'd just like to have a third party reference. e.g. They'd be slightly easier to digest than the full papers, but only give partial information, which may be sufficient. I'll continue this in the above section, where tempers appear a bit less heated. Ndickson (talk) 20:56, 8 June 2010 (UTC)

What intrigues me about D-Wave's page at its current form is that most of the criticisms are sourced from blog entries and other non-peer-reviewed sources (basically personal opinions and speculations), while any attempt at adding scientifically verified information is met by resistance.

Removing references to Google demo papers is not a problem. The result would be:

-----

== 2009 Google Demo ==

On Tuesday, December 08, 2009 at the Neural Information Processing Systems (NIPS) conference, a Google research team lead by Hartmut Neven used D-Wave's processor to train a binary image classifier.

-----

As a source, one can use http://www.maximumpc.com/article/news/google_dwave_claim_have_created_first_quantum_computing_image_search or http://www.tech-talks.com/google-demos-quantum-computer-at-nips-2009/ or any of the many articles one can find over the Internet about this demo. —Preceding unsigned comment added by 216.13.217.230 (talk) 21:28, 8 June 2010 (UTC)

That would be a good addition I think. Verbal chat 07:46, 9 June 2010 (UTC)

I added the section. Thanks. —Preceding unsigned comment added by 216.13.217.230 (talk) 19:15, 9 June 2010 (UTC)

Here is a link to D-Wave's list of peer-reviewed, published papers: http://www.dwavesys.com/index.php?page=publication-list It does include papers published prior to 2005 (invalidating a claim on the Wikipedia page). Any suggestions as to the best way of making this list available on the Wikipedia page? —Preceding unsigned comment added by 216.13.217.230 (talk) 20:45, 10 June 2010 (UTC)

Thanks for the addition of the list Verbal. I edited the page to emphasize that they are peer-reviewed, and also removed the statement about pre-2005 publications not being available. —Preceding unsigned comment added by 216.13.217.230 (talk) 17:20, 11 June 2010 (UTC)

Here is an article in IEEE Spectrum about it: http://spectrum.ieee.org/computing/hardware/loser-dwave-does-not-quantum-compute/0

Key quotes:

"If this were the real thing, we would know about it. D-Wave hasn't demonstrated signatures believed to be essential to quantum computers, such as entanglement, a coupling between qubits." --Christopher Monroe, a quantum-computing researcher at the University of Maryland, in College Park.

"Even the best prototypes can't keep more than 10 qubits in entangled states for long. Because of this I am very skeptical of D-Wave's claims that it has produced a 128-qubit quantum computer, and talk of reaching 10 000 qubits at this point is advertising hype." -- Paul Benioff, physicist who pioneered quantum computing at Argonne National Laboratory, in Illinois,

"D-Wave has made claims that have not been generally regarded as substantiated in the community." -- Anthony Leggett, Nobel laureate in physics, University of Illinois at Urbana-Champaign. Guy Macon (talk) 07:16, 22 May 2011 (UTC)

More quotes, this time from Scott Aaronson, Associate Professor of Electrical Engineering and Computer Science at MIT, expert on the limitations of quantum computing.

"it's notable that, now that D-Wave has happily joined the ruling-out-the-null-hypothesis club, we're down from 128 qubits back to 8. This paper also makes no claims to demonstrate entanglement, which is almost certainly necessary for any interesting quantum speedup." -- http://www.scottaaronson.com/blog/?p=639

"Any claims by D-Wave that the practical value of quantum annealing has already been demonstrated need to be taken with a huge grain of salt." -- http://blogs.forbes.com/alexknapp/2011/05/24/q-and-a-with-prof-scott-aaronson-on-d-waves-quantum-computer/ Guy Macon (talk) 18:12, 27 May 2011 (UTC)

You all might be interested in an article in Nature (12 June 2011, p. 18) titled "First sale for quantum computing". It also discusses the "mistrust" some folks have about D-Wave. - J. Johnson (JJ) (talk) 23:33, 17 July 2011 (UTC)

This is very silly. Arguing about something that can be measured is a waste of time. It appears that some people are just not ready to accept a quantum break-though yet, no matter what. Skepticism is healthy, as long as it's not just an excuse to be lazy.--SkipMcCormick (talk) 01:42, 4 March 2012 (UTC)

OK, the problem is that "Quantum Computing" in the well-established field that treats of the theory of quantum algorithms refers to a theoretical device that is assumed to be capable of certain things. The pragmatic question is how close you can get to that ideal and even if you don't get particularly close, whether you can still be useful, i.e. still extract some of the speed-up you'd get from the idealized device. This is pretty finicky for all sorts of reasons I won't go into. So there is a gray area of what constitutes "pretty good: especially if we are still in the stage where proof of principle can already get you into Nature, and what is "pretty good" in those papers is not usually portable to worthwhile practical purposes. Anyway, a fair consensus is that there are various devices with a handful of qbits that are pretty good in that generous sense. After all, everybody is pioneering and the Wright brothers did not exactly build a Concorde etc... If it gets into dozens or scores of qbits (or Qbits, people get worked up over the difference but it does not really matter here), there's usually some restrictions on what algorithms the device can do, so we are getting further and further away from the idealized universal device. But if you cry foul at this point you get the reasonable reproach that you should have cried foul from the get go, since even the best 2-qbit devices are really nowhere near the idealised device. So thus far we have the idealized quantum computer, for which, in anticipation of a really good approximation being one day engineered, we have already developed all the algorithms. Then we have the various handful-of-qbits (sometimes more) experimental devices which are really not very good but nonetheless cutting-edge science. And then we have the DWave machine with 100s, 1000s, 10,000s or whatever of qbits. What gives? Well, simply that it is not even remotely an approximation of the idealized device that is the theoretical construct of the QC field. So not a quantum computer in that sense. Is it a quantum computer in any other sense? Well, yes, if only because every physical device that is used as computer is a quantum computer in the some sense! The question is, does its operation "explicitly" exploit quantum phenomena, and the answer is a cautious yes. It is certainly the tool of choice for some types of problems. But remember that there are many examples of problems are quickly "solved" approximately by suitable mechanical contraptions. So the real question is how much you can do with DWave machines, and whether it's cheaper/faster than classical alternatives. Figuring out ways to massage as many different kinds of problems into a form that the machine can work with is one sensible thing, and that task is made lighter by scaling up how many of these flux qbits they fit into the machines. So that's what they've been doing.137.205.101.77 (talk) 16:51, 8 March 2017 (UTC)

Following the addition of the recent QC vs PC comparison source, I added the following text to the article (This claim is attributed to Catherine McGeoch in the article, although not quoted):

However, she admitted that the comparison is "not quite fair, because generic computers will always perform less well than a device dedicated to solving a specific problem"

Based on a paragraph from the article:

The speed tests are also not quite fair, because generic computers will always perform less well than a device dedicated to solving a specific problem, says McGeoch. "A next step would be to build a conventional processor optimised for this task, for a fairer comparison," says O'Brien.

An anonymous editor changed it to:

However, it is not unusual for general-purpose algorithms to perform less well than a device dedicated to solving a specific problem

There are multiple problems with this formulation, all of which seem to erode its value:

• The attribution of McGeoch was removed, who is the authority in this case. Positive statements are attributed to McGeoch, and so should the negative, if she said them.
• The admission that the comparison is unfair. As far as I can tell this was a comparison between regular PC CPUs against QC -- which I find unfair, because it's not the best that classical computers have to offer. Comparing QC to GPUs (vector processors) and ASICs would be fairer.
• "is not unusual" is very awkward formulation. Consider "It is not unusual for CPUs to perform less well than a GPU at rendering 3D graphics". While technically true, it suggests that sometimes CPUs may be faster -- it's misleading. The source says "will *always* perform less well"

Note that McGeoch was hired as a consultant by D-Wave, with the goal of coming up with a favorable comparison between quantum and classical computers. It's not a surprise to be that the comparison is unfair, but it's better to state explicitly in the article for balance. -- intgr [talk] 17:34, 17 May 2013 (UTC)

I think the reason why the "algorithm" expression was used instead of the "conventional processor" expression is that it has been shown that you can get as good or better performance than D-Wave using a conventional processor with custom software packages designed to solve the problems D-Wave is good at solving. See my edit in the controversy section for the reference. 130.216.63.250 (talk) 00:26, 2 July 2013 (UTC)

The criticism section is pretty flawed. It implies that after the google/NASA deal, all academic critics have stopped criticizing the company. This is not true. In general, the criticism section should be about the specific complaints of critics, and the response by the company - and a new impressive deal with google+NASA doesn't count as a response.

Another point which is often raised in discussions is that "pro D-Wave" views are based on published peer-reviewed articles, and the criticism is from blog posts and private discussions. While this is true, it risks leading to a dangerously distorted article. The peer review of D-Wave's articles doesn't mean that they are accepted as true, or as telling the whole story, but mostly that they are interesting and likely to be true, and to contain new science. On the other hand, criticizing D-Wave will rarely be publishable because it is hard to say "this company is full of BS" in a way that has new scientific content that a journal will want to publish. So we should be careful not to confuse publication with credibility here. Nevertheless, it is interesting that nevertheless people have found ways to make their D-Wave criticism into scientifically interesting articles; see e.g. arXiv:0705.1115. Aram.harrow (talk) 01:14, 22 May 2013 (UTC)

At least some of the "peer-reviewed articles" are, in fact press releases, or have no content which is falsifiable. Still, we do have a problem with blogs and unpublished papers (arXiv are not published); namely what is the field we are looking for an "expert" in, per WP:SPS. "Quantum computing" (we don't know who the experts are), "experimental physics" (probably, IMO, not the right field, and we don't know who the expert are), "computational complexity" (the right field for some claims, but not others), or ???. If D-Wave employees are the only (otherwise) credible "experts", we could not use their statements, either. — Arthur Rubin (talk) 01:11, 27 June 2014 (UTC)

Aram, the article you mention is from 2007 which is ages and ages before any D-Wave machines were released to the public, it has only 22 citations on Google Scholar making it one of Bravyi's least cited articles of all time, and the title is "Classical approximation schemes for the ground state

energy of quantum and classical Ising spin Hamiltonians on planar graphs" which means we do not need to read the rest of the paper to know that this "criticism of D-Wave" is not strong at all. It was known since 1965 that Jack Edmonds' "blossom algorithm" can solve planar graph Ising models in polynomial time... but D-Wave's connectivity graph is non-planar. If the problem is planar: use a classical computer to optimize it. If the problem is non-planar, annealing on D-Wave might be faster (and probably WILL be more energy efficient) than solving the problem on a classical computer. Finally, I think any "criticism" of D-Wave right now is a bad idea because humans as a species still don't have enough knowledge to determine whether or not the claims of those critics will turn out to be relevant. You raise an interesting point about blog posts vs peer-reviewed articles, but I do think peer reviewed articles are in general much more valuable than blog posts and criticisms should be publishable in the journals just like you say that pro-DWave papers can be (and still be false or irrelevant, etc.). The criticisms of Umesh Vazirani, Scott Aaronson and Matthias Troyer in the article are all terribly wrong in my opinion. Umesh: "That speed up unfortunately does not hold in the setting at hand" ... why not? I see no reason why we can't get square root speed-up for a Grover-like search. Umesh: "and therefore D-Wave's "quantum computer" even if it turns out to be a true quantum computer, and even if it can be scaled to thousands of qubits, would likely not be more powerful than a cell phone" well now it has been scaled to thousands of qubits and it beat calculations on Google supercomputers that Hartmut Neven said in his talk at the APS conference in 2016 "spent millions of dollars of computer time to do" so Umesh's criticism about the "cell phone" has already been proven to be misleading. Scott: "blowing results out of proportion on press releases that claim speedups of three orders of magnitude, in light of a paper[40] by scientists from ETH Zurich reporting a 128-qubit D-Wave computer being outperformed by a factor of 15 using regular digital computers and applying classical metaheuristics (particularly simulated annealing) to the problem that D-Wave's computer was specifically designed to solve" show me one classical computation that can find the ground state of the Ising models in Katzgraber's paper faster than D-Wave's machine did. Show me just one. Matthias: "Matthias Troyer at the Swiss Federal Institute of Technology in Zurich, said that they found "no quantum speedup" across the entire range of their tests" ... funny, his 2014 Science paper shows quite clearly a figure with the label "DW2 Faster" and another label "SA faster" meaning simulated annealing is faster.... and there were points labeled in the "DW2 Faster" region that were SEVERAL ORDERS OF MAGNITUDE away from the cross-over line... furthermore Katzgraber's and Denchev's papers BOTH confirm that Troyer's simulated annealing is not only orders of magnitude slower than D-Wave, but also scales with a worse slope, on a log graph. Dr. Universe (talk) 00:02, 6 December 2018 (UTC)

The article says nothing at all about quantum algorithms. Yet a quantum computer can only run asymptotically faster than a classical computer by using a quantum algorithm such as Shor's factoring method, Grover's search method, etc. D-Wave claims that their quantum computer efficiently solves an NP-hard problem, namely quadratic unconstrained binary optimization or QUBO. However there does not currently exist any known quantum algorithm for solving NP-hard problems usefully faster than existing classical algorithms. Furthermore if any efficient quantum algorithm existed for NP-hard problems its implications for quantum computers would be vastly more profound than Shor's algorithm. Yet the article does not say anything about the quantum algorithm D-Wave is using. Something is very wrong here. Vaughan Pratt (talk) 19:02, 18 October 2013 (UTC)

The lede does not reflect the content of the article, since it summarizes the press-releases of the company and presents them as facts. The article itself is more balanced.--Ymblanter (talk) 21:59, 27 November 2013 (UTC)

Six years later, it's presumably worse. 73.222.1.26 (talk) 22:14, 19 March 2019 (UTC)

I propose that D-Wave Two be merged into this article. What isn't a press release or a quote from a press release is not much longer than what is already here. (I haven't researched whether "Two" was split from this article. It doesn't really matter, as they should be together.) — Arthur Rubin (talk) 00:52, 27 June 2014 (UTC)

Agree. The significance or notability of the company and the systems it builds does not seem to extend to the individual models of that system. ~ J. Johnson (JJ) (talk) 20:58, 21 October 2014 (UTC)

• Merge notability for company entangled with products. For now, per WP:SPINOFF. Widefox; talk 18:46, 1 January 2015 (UTC)

I agree with the merge. I have just gone through and made some clean-up edits to the D-Wave Two article. How do we go about actually merging it in? It has been flagged for a long time --Strongguy1 (talk) 21:14, 20 March 2015 (UTC)

If there is no dissent soon, I will be moving to perform this merge Strongguy1 (talk) 15:26, 31 March 2015 (UTC)

I'm removing this as a stale merge. Widefox; talk 21:27, 5 May 2016 (UTC)

The criticism section is quite flawed and the quote by Umesh Vazirani is very out of date and flawed. It’s about time it gets removed. The Talk page has a criticism section, but it’s all discussion and no talk about change. VPL Strathcona (talk) 20:49, 31 March 2019 (UTC)

Has the question of speedup over classical computation been resolved? There seems to be a common thread about the noise levels that I would be hesitant to remove, so I'm interested in what you think are the best post-2015 sources on the topic. EllenCT (talk) 21:16, 31 March 2019 (UTC) @VPL Strathcona: ? EllenCT (talk) 21:12, 2 April 2019 (UTC)

@EllenCT:, everything in the section is pre-2015. This 2016 paper by Google is crystal clear: See Fig. 4 https://arxiv.org/pdf/1512.02206.pdf. Clearly the time it takes D-Wave to solve the problem is 10^7 times faster than the classical algorithms ("simulated" annealing, and QMC). VPL Strathcona (talk) 01:16, 10 April 2019 (UTC)

@VPL Strathcona: Great. Can you help me find literature reviews other than and ? I'd also like your comments on , in particular, "To date [July 2018] classical heuristics have consistently outperformed quantum annealing based approaches." And which states, "A huge obstacle we are facing is that operating a single fault-tolerant qubit can require many thousands of physical qubits. In a recent study, it was estimated that quantum computing could achieve a significant speedup (in absolute time), but this advantage vanished when the classical processing required to implement error-correction schemes was taken into account [37]." EllenCT (talk) 21:38, 18 April 2019 (UTC)

@EllenCT: Unfortunately I don't know what reviews to give you. The two reviews you just gave me are quite different from each other. What type of review are you looking for? Those two are cut from a completely different cloth! The paper takes forever to load and then it says "service unavailable", can you give me an arXiv link or just the title and authors of the paper? As for the last paper, they are talking about circuit based quantum computation, which has nothing to do with D-Wave. It is true that the number of physical quits is often something like 15,000 times the number of logical qubits. This is far from the case for D-Wave annealers. The closest thing to "quantum error correction" for D-Wave annealers is "quantum annealing correction" where there's only 4 physical qubits per logical qubit, however that is a topic is not really explored in much detail yet.

@VPL Strathcona: That paper is on arxiv as . A more recent demonstration of limited quantum speedup is here and was published in Physical Review X. There were also D-Wave papers in Science and Nature last year, using D-Wave processors as programmable simulators of quantum systems and . Although these papers are not directly concerned with the speed of simulation, they constitute very clear evidence of quantum effects. Another note: The discussion of Pegasus not being open to the public is inaccurate; an open-source generator was already online as of April 2018 . I work for D-Wave so I'm very reluctant to edit the main page myself. Adking80 (talk) 12:23, 13 May 2019 (UTC)

@Adking80: Thanks for sharing the arXiv link. Now I know which paper EllenCT was talking about. Isn't that Katzgraber paper extremely deceptive? They are pretending that QUBO can be solved in polynomial time on a classical computer, but to do that they needed to use the blossom algorithm, which requires the graph to be planar. D-Wave graphs are not planar, so Katzgraber represented entire groups of 8 qubits as a single qubit. Isn't that cheating? So the 2048 non-planar qubits of D-Wave becomes 256 planar qubits in Katzgraber's simulation, isn't that extremely unfair and dishonest? It seems that to properly criticize D-Wave, you need to solve a problem on a non-planar graph, faster than the 2000Q, and as far as I know, this has never been done in any of the papers ..... they always say in their abstract that their classical algorithms are so great and everything, but then the plots always show D-Wave finding the solution faster. As for the "pegasus" I'm not sure ... I don't recall writing about Pegasus at all, but then looking at my edit history it looks like I did edit it (or someone else from the Vancouver Public Library, Strathcona branch, edited it if the account was signed in). You have provided a link to a discussion on GitHub from April 2018. Do you have a link to the code? VPL Strathcona (talk) 20:04, 13 May 2019 (UTC)

@VPL Strathcona: The paper is not deceptive because it is in response to a particular D-Wave paper on a set of instances that can be reduced to planar problems. The particular Pegasus generator code is found here . Adking80 (talk) 21:40, 14 May 2019 (UTC)

@Adking80: Thank you so much. I now concede that the pegasus generator was on GitHub since 2018. I'm not sure when it was made open-source (since it seems many D-Wave repositories are private, and one of the commits in that repository midway through the commit history, says "added explanation so that we can pre-release it to customers"). I'm sure there's a lot of people that are not D-Wave customers, but even assuming this repository was made public right at the start, how many people knew about it? It looks like Mark Fingerhuth did know about it, based on his comment here: https://quantumcomputing.stackexchange.com/questions/2536/what-is-d-waves-pegasus-architecture/5134#5134. But even a few months ago Neil de Beaudrap was constantly asking for a mathematical description of the Pegasus, in his comments there. I have looked deeper into the git repo you posted, and while I do see some formulas for the Pegasus edges added in a commit by user "Boothby", it's very hard for me to understand what is going on, with so little description of what the nodes are and what the indices mean. The phrases "orthogonal major offset" and "orthogonal minor offset" mean nothing to me. In any case, the paper by Dattani, Chancellor and Szalay uses a completely different 6-index scheme, compared to the 4-index scheme explained in Boothby's commit. The formulas are completely different and I see no correspondence between Boothby's formulas and theirs (probably mostly because the node definitions and meanings of the indices are completely different). In any case, the paper on arXiv explains each node and each edge in a way that I can understand, whereas the generator you sent to me can probably make Pegasus graphs, but the comments do not give me enough information that I would be confident that I could reproduce Pegasus on my own (indeed it looks like the authors did not use D-Wave's scheme for describing the edges). I actually searched online earlier (like Neil de Beaudrap I guess) to see if there was some explanation of Pegasus the way I understand Chimera so easily, but I couldn't find anything at the time. To conclude this paragraph, what do you recommend we add/change to the Wiki article? I believe that the GitHub repository is not able to replace the value of the arXiv paper, which seems to be a better "source" for describing Pegasus. About the Katzgraber paper, thank you for pointing out the context that this was in response to a D-Wave paper about planar graphs. I guess the point is that for planar graphs you should use the polynomially scaling blossom algorithm, and for non-planar graphs you should use D-Wave (if it embeds well enough). So I thank everyone here for all the references, but I see no reason why there should be so much outdated "controversy" in the controversy section of the Wiki article. The Troyer papers were completely wrong, the Katzgraber paper is about planar graphs, and the Vazirani quote was from 2007 (before even the D-Wave One) and it does not seem "noteworthy" enough to be on Wikipedia. Furthermore, Vazirani's paper was quite deceiving. He proved in 2001 that AQC with only Z's, and standard QC with gates, have exactly the same speed-up for a Grover search. So then what is his problem with D-Wave? Why is he saying that the company is founded on a "misunderstanding" of his paper? Seems to me that he himself proved that AQC with only Z's can be as powerful as a universal QC for Grover search, and to me all that matters is that it's useful for at least one algorithm .... I mean give me a break .... how far have "universal" QCs come in the 34 years since David Deutsch's 1985 paper? Call me when there's a "controversy" section for Google's 72-qubit QC presented at the APS conference but never shown to anybody, and for IBM's 16-qubit public access QC which is vastly outperformed by classical computers, not because of lack of qubits but because of too much noise. VPL Strathcona (talk) 03:12, 15 May 2019 (UTC)

@VPL Strathcona: I agree D-Wave is held to a higher standard than other companies and that a lot of the controversy section is completely outdated, particularly the quote from Vazirani. The landscape has changed a lot in the past 12 years. Much of the criticism comes from people who have an interest in other quantum computing efforts, of which there are now many. And much of it comes from academics who don't want anybody saying that a quantum annealing processor is a quantum computer. This is in spite of the underlying equivalence between circuit model and annealing quantum computers, cf. and . A critic might say that D-Wave's coherence times are too short and temperature is too high. However, decoherence (which affects circuit model and annealing in different ways anyway) and temperature are issues for all physical QC implementations, including IBM, Google, Rigetti, etc. Some manifestations such as Rydberg atoms are less prone to e.g. temperature, but they have other implementation challenges, often programmability. Another criticism is lack of non-stoquastic terms in the D-Wave Hamiltonian, but D-Wave recently demonstrated non-stoquastic coupling as well. So, in my opinion the page needs a significant overhaul, but it should be done by people without the conflict of interest that I have. Adking80 (talk) 15:43, 15 May 2019 (UTC)

@Adking80:. D-wave machines are programmable, therefore they are computers. Any problem can be compiled onto a D-Wave machine with enough qubits, just not always efficiently. We could even do Shor's algorithm on D-Wave machines, but not efficiently (the gadgets for converting universal problems onto Ising problems are not efficient ... but they still exist, so you can still do any problem you want on the D-Wave machine, the embedding might just be so bad that it would be worse than doing it on a classical computer). It's a "quantum computer" just not efficient for every possible problem. You talk about decoherence and temperature, but we don't really need very long coherence for quantum annealing compared to a deep circuit. As for temperature, I don't know of any company who's hardware has better quality qubits than D-Wave. Thanks for the paper on non-stoquastic terms in D-Wave Hamiltonian, I didn't know about this. Seems it's not a lot of qubits though, and MIT Lincoln Labs and Norhtrup Grumman and Google have been trying to do non-stoquastic terms with not much success. I'm not sure if it's practical to program for non-stoquastic machines either (when we leave the Ising model, I lose my intuition for how to turns a real-world problem into a minimization problem). You talk about equivalence between AQC and circuit model, but those proofs only work at 0-temperature and for closed systems not even interacting with the vacuum modes (so the proof would not even work for qubits in a real vacuum that has QED vacuum modes). But luckily we don't have to worry about any of that if we don't claim D-Wave machines can do universal QC. The main point is that they can solve some optimization problems faster than any classical algorithm on a classical computer, including Katzgraber's. There's not enough qubits to solve real-world problems, but they're getting there. A major criticism now would be: how does D-Wave's machines compare to Fujitsu or Hitachi annealers? Maybe Fujitsu can find the ground state just as fast as D-Wave, meaning there's not really any "quantum advantage" even though there's no doubt that there's quantum entanglement, quantum tunnelling, quantum coherence, etc... (how can there not be quantum effects? Impossible. .... But how relevant are the quantum effects ... it remains to be seen what the classical annealers can do). The classical annealers have all-to-all coupling? That seems like an advantage. As for conflict-of-interest, well I don't work for any company or any university ... I'm just an enthusiast. I just didn't want to remove the controversy section uni-laterally. In the title of this Talk section, I said I'll give people 2 months to provide some reason why anything from the controversy section should stay. I don't see anything "notable" enough to stay there. For historical purposes it might be neat to have information about how controversial D-Wave was in the early days, but are there any good BALANCED sources to cite there? Everything seems to be emotionally charged and unbalanced. For example Katzgraber keeps saying he's crushing D-Wave runtimes with classical computers, but then why in his figures the D-Wave curve is below the classical-algorithm curves when the y-axis is time-to-solution? Has Scott Aaronson even said one accurate thing about D-Wave controversy? Sure Vazirani made that quote, but is it really notable enough to be in an encyclopedia? Isn't keeping that quote there going to make people lose respect for him? Doesn't he probably want it to be taken down for the sake of his reputation? I will wait until end of May to see if anyone comes up with a really strong argument for why anything in the controversy section should really be there. Finally: what do you want me to do about the Pegasus? I don't think the arXiv paper can be "replaced" by some comments in a code on GitHub, which lacked the amount of detail given in the paper. How about something like: "D-Wave began development of a GitHub repository in 2018 which included a Pegasus generator and it should be possible to deduce a description of the graph by looking at the comments in the code [1] but later in 2019 a group of academics de-coded it themselves and made their findings available to the public [2]" where [1] is the GitHub repository, and [2] is the arXiv paper. I just don't know if "GitHub repository" is considered an appropriate "source" on Wikipedia because I'm somewhat new here. VPL Strathcona (talk) 03:33, 16 May 2019 (UTC)

@VPL Strathcona: Here is a whitepaper on the D-Wave website describing Pegasus: found at . My personal opinion about the criticism section is that it is out of date, but was important at the time. The fact that it is difficult to find prominent scientists giving such pointed criticism is proof enough of that. In any case, maybe someone else will provide a good reason to keep a lot of this criticism section. Maybe not. Adking80 (talk) 15:26, 16 May 2019 (UTC)

@Adking80: Thank you for the white paper! Very interesting! I note it's form February 2019, so a bit newer than the references in the current Wiki article, but that doesn't mean we can't add it to the Wiki page. Figure 2 in the whitepaper actually looks a lot more like Figure 12b of the Dattani-Chancellor paper than the pictures they used to show in the talks (for which a PDF of Mark Johnson's is online and pictures from Trevor Lanting's are online) and GitHub program. Okay so there is the Pegasus section and the Criticism section: I get the sense that you're a bit vague about what you want me to do, or in any sense, you're not explicitly saying what you would like. I understand this is because you say you have a conflict of interest, which is very admirable of you (some company employees would just try to boost their company's reputation). But I do not have a conflict of interest since I am not working for any tech company or university, I am just an enthusiast. So I will make the changes. I am indeed seeking the opinions of other people here though (and I guess you're the only one engaging in the discussion, so your opinions are 100% welcome whether or not you have a "conflict-of-interest"), because I don't want to uni-laterally delete the criticism section (or change the Pegasus section which I didn't even write in the first place). My opinion is that the criticism section is unfairly biased *against* D-Wave, and has inaccuracies (some of them due to being "out of date" but some of them plain wrong) which can hurt the company's reputation, and hurt the entire field of adiabatic quantum computing or quantum annealing in general. I get the sense from people that adiabatic quantum computing "isn't real quantum computing" even though the papers you mentioned seem to indicate that it is quantum, and common sense tells me that it is computing. I personally think the criticism section should just be removed entirely, and someone else can start a new one if they really think it's necessary to criticize D-Wave. But you say that "it was important at the time" and that "maybe someone will provide a good reason to keep a lot of this criticism section", which makes me unsure of what you want me to do. Do you want to keep some parts? I'm thinking to remove the whole thing, but if there's parts you want to keep I can do that. I think you're too worried about having a "conflict-of-interest", and I'm too worried about making uni-lateral decisions. I guess it would be fair for me to delete the criticism section if no one objects though (I will still wait until the months are over, even though it feels like 2 months have already past, I want to give everyone a long enough time to object to what I'm considering). Do you object to what I'm doing? (i.e. removing the criticism section, and modifying the Pegasus section)?

@VPL Strathcona: I think that the deletion of the Controversy section is unwarranted - I would recommend a rework of the section to remove outdated criticisms but to summarize the still-relevant, previously existing criticisms and accompany them with paragraphs explaining new progress they have made to address these problems. I personally think in particular that the Nature 2011 and Science 2014 paragraphs are worth keeping, as well as a reworked version of the paragraph of criticisms from Scott Aaronson, given that he is perhaps the most outspoken, and well-known critic of their work in the QIS field. Perhaps these could all be combined and shortened. The claim that a Wikipedia section of an article might "hurt the company's reputation, and hurt the entire field of adiabatic quantum computing or quantum annealing in general" as you state in the previous reply, seems rather like hyperbole to me, and is in any case not the concern of editors aiming to be objective only. Nonetheless, your concerns about bias in the previous version seem to me to be entirely warranted. As for supporting studies, keeping the PRX 2014 paragraph and adding new paragraphs about their work to implement nonstoquastic Hamiltonians in and the PRX 2016 paper would balance it out (again, perhaps shortening and combining these paragraphs). Whether or not it seems unfair that D-Wave faces a lot of criticism in contrast to others as noted in a previous reply, it is certainly a significant component of the public conversation about their work, largely due to serious concerns from experts in the field voiced repeatedly over several years. As for the comment at the top of the thread which claims that all criticism seems to be pre-2015, there is certainly a wealth of criticism levied in recent years such as those mentioned elsewhere in this thread and ,,, to name a few, which could be added if the timeliness of the content is a concern. The disagreement of one editor with the content of some critical articles does not mean that the criticism ceases to exist and that the criticisms then aren't notable enough to warrant a section on the Wikipedia page. The topic is still a rather heated, popular, and contentious one among those working in QIS (indeed, as shown by this Talk page alone) and I would advocate re-balancing the controversy section and putting it back up, rather than outright deletion. A restructuring of the section may also be nice, since the previous version bounces back and forth between criticisms and support. I think having the controversy section organized with two subsections of Criticism and Support or something of the like may be more clear to readers. If there are no serious objections, I think the section should be reinstated and a rework should be conducted as soon as possible thereafter. Tyttcfm (talk) 07:26, 24 April 2026 UTC [refresh]

@Tyttcfm:. Thank you so much for writing to me rather than just quickly reverting the edit! Honestly I did originally have the intention of doing a complete re-write of the criticism section or making it much more compact while still having all the original (now historical) criticism such as the 2014 Science paper and all. However one conclusion I made for sure was that the section should be taken down, as no one in 9 weeks gave any reason to keep anything up. I disagree that the Science 2014 paper should be mentioned at all unless we are to have a "support" section and put it there. I very much like your idea of having a "support" section if there is a criticism section. The Science 2014 paper showed that the D-Wave machine beat Troyer's groups best classical-computer algorithm at the time, by several orders of magnitude, for many problem instances. His laptop also beat D-Wave for many instances, but that should NOT be considered criticism at all! I've been studying quantum computing in multiple decades and I can tell you first hand that it was a very "academic" and "theoretical" subject for so long, and some colleagues were saying it will be 70 years by the time we actually have hardware. The fact that in 2014 there was a device of any sort, that was BEATING classical computers by orders of magnitude, for ANY problem, was absolutely amazing to me. It means that for THOSE specific problem instances, I would certainly want to use the quantum annealer, and for the other instances, go ahead and use Troyer's laptop.... but do not "criticize" D-Wave at all for being beaten by a laptop on some very specific instances of problems, when D-Wave is beating the laptop by orders of magnitude on other specific instances of problems! You wrote "largely due to serious concerns from experts in the field voiced repeatedly over several years" but Scott Aarsonson is NOT an expert at all!!! He knows almost nothing about AQC or quantum annealing!! He is a "theoretical computer scientist" specializing in computational complexity theory and computational complexity classes for quantum algorithms. AQC is not his field. Look at the speaker lists for the annual AQC conference, the repeated speakers are Nick Chancellor (who never criticized D-Wave), Paul Warbourton (who never criticized D-Wave, Mohammad Amin (who works for D-Wave), Andrew King (who works for D-Wave), Daniel Lidar (who convinced his university USC to spend $10 million on a D-Wave, and published a Proof that AQC = Standard QC), Nike Dattani (who never criticized D-Wave), Helmut Katzgraber (who has shown repeatedly that D-Wave crushes his state-of-the-art classical algorithms in terms of runtimes), Elisabeth Crosson (who never criticized D-Wave). Now I'm looking at the 4 links you sent me. First, I never want to see that Katzgraber paper about "planar" graphs ever again. The D-Wave 2000Q has 2048 qubits, and he treated every 8 qubits as 1 logical qubit, so that D-Wave would only have 256 qubits instead of 2048, and these 256 qubit would be in a PLANAR graph. For a planar graph you can use the "blossom algorithm" which is polynomially scaling, and this was known since the 1960s when Jack Blossom invented it. D-Wave's graphs are NON-PLANAR. That's why if you use all 2048 qubits and don't treat each non-planar subset as just ONE spin, you CANNOT use the Blossom algorithm anymore. Therefore nothing Katzgraber did in that paper is new since 1960s, and it is an extremely unfair comparison to D-Wave. There is no sense in confusing readers by putting those misleading papers in a Wiki article! Unfortunately Katzgraber and Troyer are authors on two of the other 4 papers you gave me, and Troyer is the one bragging about a laptop beating D-Wave when that same laptop was orders of magnitude slower for several problems in the exact same paper. So I unfortunately do not think I should spend anymore time reading their papers unless someone says something specific that I should look into (such as the specific things I wrote for others here: planar graphs, D-Wave beating laptop for several instances of the problem, etc.). Thank you for the paper by Tameem Alabsh. His "review" on AQC was extremely biased... it talked about tonnes of AQC algorithms and had a whole section on AQC algorithms for which there's not yet any proof that it's faster than classical computers, and yet, why is it that there's not one single reference to one of the most famous and most frequently published AQC algorithms: the one for factoring integers? Anyway, while I cannot trust papers by Albash to be objective (just like I can't trust Katzgraber and Troyer), I see from the abstract that he is saying that the temperature has to be lowered as the number of qubits goes up. This I think everyone already knew, but perhaps it is a fair criticism. I can read more into it when I have time. However is it Wikipedia worthy? Are we going to list in an encylopedia, every single paper that says something like, in order for D-Wave to work, we need low temperatures? The paper has only 26 citations on google scholar. It is not in the media or newspapers or covered by journalists, which are often the requirements for things to be Wiki worthy. Where is the citation to the paper that AQC = Standard QC ? There's one by Aharanov et al., and also by Lidar & Menzel or something like that. Where's the papers that show that D-Wave factored numbers larger than anything Shor's algorithm ever did? I agree that there's no proof yet that those factoring algorithms will be scalable, but should we include those papers in a wiki article about D-Wave? I currently would say no. It's not encyclopedia worthy (and in fact, those article HAVE indeed been covered by the media in dozens and dozens of places, so in some sense I would push to include those types of articles to be in the D-Wave article if Albash's "quantum annealing requires very low temperatures" should be in it!). So I do like your idea of having "criticism" and a "support" section, but I don't know how those should look. What articles should we cite and not cite? How should we construct the paragraphs? and the overall structure? I believe Wiki is not a "forum" so it's not a place to dump everyone's opinions about a company's article. Right now I think it's much better to have "no criticism section at all" rather than having what we used to have. What we used to have, is something I'm glad was there originally, but I think now is the time we can move on from that era. FINALLY, maybe we can point out these things like Albash's result that we need low temperatures to do quantum annealing, but does it have to be on the Wiki page for "D-Wave Systems" the company? perhaps we could have a Wiki article called "Qauntum Annealers" or "D-Wave Quantum Annealers" or even "Criticism of D-Wave" ... is it criticism of D-Wave the company? Or of (some of) their products? I don't see why we need to have Albash's paper cited anywhere in the D-Wave COMPANY wiki page. I am open to discussion from everyone though! VPL Strathcona (talk) 20:26, 7 June 2019 (UTC)

This list still/again consists of people already mentioned, people who don't rate as among the most notable collaborators, and people who have never worked with D-Wave. As such I think it is best removed. Adking80 (talk) 20:38, 23 January 2020 (UTC)

Ladizinsky goes around calling himself a co-founder but the article doesn't reflect this. Is he more of a "founder" in name because they gave him extra stock, or what? Where does this information come from and what is his own claim based on? 12.231.67.211 (talk) 00:32, 26 February 2021 (UTC)

It is a bit confusing. The article currently states, "D-Wave was founded by Haig Farris (former chair of board), Geordie Rose (former CEO/CTO), Bob Wiens (former CFO), and Alexandre Zagoskin (former VP Research and Chief Scientist)." However, the D-Wave Systems website describes Eric Ladizinsky as "Co-Founder and Chief Scientist", and it also describes Geordie Rose as "The Founder" and Haig Farris as "co-founder". I suggest that the article be updated to read, "D-Wave Systems was founded by Geordie Rose (former CEO/CTO), along with Haig Farris (former chair of board), Eric Ladizinsky (current Chief Scientist), Bob Wiens (former CFO) and Alexandre Zagoskin (former VP Research and Chief Scientist)." nagualdesign 01:23, 26 February 2021 (UTC)