Just in Case You Want to Simulate an Ising Spin Glass Model

This is just a short update to include a link to the solver code developed by Matthias Troyer et al. that I discussed in my last blog post.

The advantage of conventional open source code is that everybody can play with it. To the extent that the code faithfully emulates quantum annealing of an Ising spin model, this will be a net benefit to D-Wave as it allows programmers to perform early testing of algorithmic concepts for their machine (similar to the emulator that is part of D-Wave's evolving coding environment).

Matthias Troyer attached the source code to this earlier paper, and for easier download I put it into this zip file. As I am hard pressed for time these days, I didn't get to work with it yet, but I confirmed that it compiles on my OS X machine ("make an_ss_ge_fi" produced a proper executable).

Meanwhile D-Wave continues to receive good press (h/t web 238) as the evidence for substantial quantum effects (entanglement and tunneling) keep mounting, indicating that they can successfully keep decoherence at bay.  (This older post on their corporate blog gives an excellent introduction to decoherence, illustrating how thermal noise gets in the way of quantum annealing).

On the other hand, despite enormous theoretical efforts, it is yet not decisively shown what quantum resources are required for the elusive quantum speed-up, but this recent paper in Nature (arxiv link) claims to have made some substantial headway in isolating some of the necessary ingredients (h/t Theo).

While this is all quite fascinating in the academic sense, at this time I nevertheless regard quantum speed-up as overrated from a practical point of view. Once I manage to set some time aside I will try to explain in my next blog post why I don't think that this has any bearing on D-Wave's value proposition. There's no place in the multiverse for the snarky, dark movie script that Robert Tucci offered up the other day 🙂

An Example for a two Qubits Contextuality Graph

16 thoughts on “Just in Case You Want to Simulate an Ising Spin Glass Model

  1. Hi Henning: Thanks for this update with its many interesting links, the most intriguing of which I found to be the recent paper in Nature about “Contextuality”!! If you click your link above, you will find to the right side of the abstract a “Related audio”, being a fascinating 7-minute interview by a Nature reporter with Dr. Joseph Emerson, trying to explain in everyday language this “magic of Contextuality”. I urge every reader of your blog to listen to it carefully. Thanks.

    1. Bob, I live up here in the continuous traffic jam that is sometimes referred to as the GTA, but I am just a permanent resident as this point. Have only one passport and it’s German 🙂 Have to admit though that at the Olympics I tend to cheer for the Canadian hockey team.

      At any rate, I think you are quite correct about this strange NMR QC obsession at Perimeter. Ever so often they make noise how, real soon, there will be some amazing commercial spin-offs and then time just moves on. Meanwhile D-Wave brings home the bacon. It is as if all sorts of other shiny quantum things constantly distract them.

      At this point I only skimmed the pink unicorns paper and I liked the pretty picture that I included with this post. Will have to give it an actual read to see if this is actually more than new wine in old wineskins.

      My maybe somewhat simplistic view is that QM requires a non-commutative generalization (QP) of classical probability theory (P) which is embedded in the former, hence only quantum effects that lie outside the latter set i.e. which require QP \ P will allow for what is considered ‘true’ quantum computation (a necessary but certainly not sufficient requirement for the elusive quantum speed-up).

      Incidentally your generalized Bayesian networks are a very nice down to earth computational approach to QP.

  2. Robert: For God’s sake, a little bit of chauvinism, once in a while, even for us “wimpy” Canadians doesn’t hurt, I hope!! After all, it’s about “magic”! Be good-humoured and stay well! Thanks.

  3. Taking a more careful view of the problem used in Troyer’s paper I am not sure whether this is really the best benchmark to be used. In general, on can map any such problem to the product of two objects C * B where the second is a universal logical mask that can be constructed beforehand for all the 2^N binary patterns. Practically one does not really need matrices to do that – just a linked list for the coefficients in C and a special pointer access to clip the addresses for the zero positions in the logical mask. But here is the catch. The problem has now been split into two parts of which only the first seem to be sped up if ever by D-Wave.
    As long as the Boolean part is universal it could in principle be precomputed for all possible input C matrices and most of all this set has a strange recursive structure, but for long strings (large N) one could at least assume the capacity of a quantum device for building all the combinations (and the B masks) “at once”.
    The second part is just summation and given what the company says this would still have to be performed classically if I understood them correctly. Unless you know more about speeding up summations in D-Wave.

    1. Interesting observation. My understanding is that every summation is masked as the chip physically implements the Ising model. That’s great for artificial neural networks but unfortunately not for cellular network kind of calculations.

  4. Henning: I forwarded the above link to Alex Selby and he said this:

    Sol Warda says:

    23 June 2014 at 15:07

    Hello Alex: I wonder if you have seen this latest study of benchmarking the DW2 against some optimised software, including yours, by Dr. Itay Hen of USC’s ISI, at a recent conference on Adiabatic Quantum Computation. Thanks:

    http://www.isi.edu/sites/default/files/top_level/events/aqc2014/day4-talk4-Hen.pdf

    Alex Selby says:

    24 June 2014 at 01:20

    Yes, I saw that and it looks very interesting. (By the way, the author informs me that we should disregard the slides after p.46 as these were not meant to be included in the upload and may possibly be erroneous.)

    1. Very interesting tidbit about the later slides. Will be interesting to see what will get publish in the end.

  5. I’ve just come across your blog for the first time and have read a few articles. It’s refreshing reading original thoughts on the current state and future of quantum computing rather than glancing at writers who merely regurgitate the same old stuff. I wonder if you’ve heard of this guy? https://www.goodreads.com/author/show/7224999.Mark_J_Solomon – I figure you would have some philosophical thoughts about the future of quantum computing in addition to technological/scientific ones.

    1. Just re-visited this comment thread and noticed that I never answered you.

      First of, always happy to hear that somebody enjoys my blog and thanks for noticing that I try to come up with my own material 🙂

      As to Mark Solomon, didn’t know about his work. He certainly was ahead of the curve in terms of the ever more popular ‘simulated universe’ meme.

      Of course at this point we can be pretty sure it would have to be simulated by a quantum computer and hence, at this point, I don’t find the evidence for this idea very compelling.

  6. Henning: I posted a comment on the above interview with Dr. M. Troyer, the link of which is in my above comment of June 19, 2014. If you click on it, you will see his response which I, Geordie and Dr. H. Neven disagree with!!. All three of us disagree on the number of cores they used to compare. Dr. Troyer says they used one CPU with eight cores. Also, Jean-Francois Puget of IBM (France) wants to compare his “optimised” CPLEX using his classical machine with 32 cores!. Dr. Neven rejects that and says “we typically compare to a one core CPU”. Hence, the ongoing bitching!!.

Comments are closed.