To Reach Quantum Supremacy We Need to Cross the Entanglement Firepoint

You can get ignition at the flash point, but it won't last.

There’s been a lot of buzz recently about Quantum Computing. Heads of state are talking about it, and lots of money is being poured into research. You may think the field is truly on fire, but could it still fizzle out? When you are dealing with fire, what makes the critical difference between just a flash in the pan, and reaching the firepoint when that fire is self-sustaining?

Finding an efficient quantum algorithm is all about quantum speed-up, which has, understandably, mesmerized theoretical computer scientists.  Their field had been boxed into the Turing machine mold, and now, for the first time, there was something that demonstratively went beyond what was possible with this classical, discrete model.

Quantum speed-up is all about scaling behaviour.  It’s about establishing that a quantum algorithm’s resource requirements will grow more slowly with the amount of computational resources than the next best classical algorithm.

While this is a profound theoretical insight, it doesn’t necessarily immediately translate into practice, because this scaling  behaviour may come into play at a resource threshold far beyond anything technically realizable for the foreseeable future.

For instance, Shor’s algorithm requires tens of thousands of pristine, entangled qubits in order to become useful.  While not Sci Fi anymore, this kind of gate based QC is still far off. On the other hand, Matthias Troyer et al. demonstrated that you can expect to perform quantum chemical calculations that will outmatch any classical supercomputer with much more modest resources (qubits numbered in the hundreds not thousands).

The condition of having a quantum computing device performing tasks outside the reach of any classical technology is what I’d like to define as quantum supremacy (a term invented by John Preskill that I first heard used by DK Matai).

Quantum speed-up virtually guarantees that you eventually will reach quantum supremacy for the posed problem (i.e. factoring in Shore’s algorithm case) but it doesn’t tell you anything about how quickly you will get there. Also, while quantum speed-up is a useful criteria for eventually reaching quantum supremacy, it is not a necessary one for outperforming conventional super-computers.

We are just now entering a stage where we see the kind of quantum annealing chips that can tackle commercially interesting problems.  The integration density of these chips is still minute in comparison to that of the established silicon based ones (for quantum chips there is still lots of room at the bottom).

D-Wave just announced the availability of a 2000 qubit chip for early next year (h/t Ramsey and Rolf).  If the chip’s integration density can continue to double every 16 months, then quantum algorithms that don’t scale better (or only modestly so) than classical ones may at some point still end up outperforming all classical alternatives, assuming that we are indeed living in the end times of Moore’s law.

From a practical (and hence commercial) perspective, these algorithms won’t be any less lucrative.

Yet, the degree to which quantum correlations can be technologically controlled is still the key to go beyond what the current crop of “wild qubits” on a non-error corrected adiabatic chip can accomplish.  That is why we see Google invest in its own R&D, hiring Prof. Martinis from the UCSB, and the work has already resulted in a nine qubit prototype chip that combines “digital” error correction (ECC) with quantum annealing (AQC).

D-Wave is currently also re-architecting its chip, and it is a pretty safe bet that they will also incorporate some form of error correction in the new design. More intriguing, the company now also talks about a road map towards universal quantum computing (i.e. see the second to last paragraph in this article).

It is safe to say that before we get undeniable quantum supremacy, we will have to achieve a level of decoherence control that allows for essentially unlimited qubit scale-out. For instance, IBM researchers are optimistic that they’ll get there as soon as they incorporate a third layer of error correction into their quantum chip design.

D-Wave ignited the commercial quantum computing field.  And with the efforts underway to build EEC into QC hardware, I am more optimistic than ever that we are very close to the ultimate firepoint where this technological revolution will become unstoppable. Firepoint Entanglement is drawing near, and when these devices enter the market, you will need software that will bring Quantum Supremacy to bear on the hardest challenges that humanity faces.

This is why I teamed up with Robert (Bob) Tucci, who pioneered an inspired way to describe quantum algorithms (and arbitrary quantum systems) with a framework that extends Bayesian Networks (B-nets, sometimes also referred to as Belief Networks) into the quantum realm. He did this in such a manner that an  IT professional who knows this modelling approach, and is comfortable with complex numbers, can pick up on it without having to go through a quantum physics boot camp. It was this reasoning on a higher abstraction level that enabled Bob to come up with the concept of CMI entanglement (sometimes also referred to as Squashed Entanglement).

An IDE built on this paradigm will allow us to tap into the new quantum resources as they become available, and to develop intuition for this new frontier in information science with a visualization that goes far beyond a simple circuit model. The latter also suffers from the fact that in the quantum realm some classical logic gates (such as OR and AND) are not allowed, which can be rather confusing for a beginner.  QB-nets, on the other hand, fully embed and encompass the classical networks, so any software that implements QB-nets, can also be used to implement standard Bayesian network use cases, and the two can be freely mixed in hybrid nets. (This corresponds to density matrices that include classical thermodynamic probabilities.)

So far, the back-end for the QB-net software is almost completed, as well as a stand-alone compiler/gate-synthesizer. Our end goal is to build an environment every bit as complete as Microsoft’s Liqui|>.  They make this software available for free, and ironically distribute it via Github, although the product is entirely closed source (but at least they are not asking for your firstborn if you develop with Liqui|>).  Microsoft also stepped up their patent activity in this space, in all likelihood in order to allow for a similar shake-down business model as the one that allows them to derive a huge amount of revenue from the Linux based (Google developed) Android platform.  We don’t want want the future of computing to be held in a stranglehold by Microsoft, which is why our software is Open Source, and we are looking to build a community of QC enthusiasts within and outside of academia to carry the torch of software freedom.  If you are interested, please head over to our github repository. Any little bit will help, feedback, testing, documentations, and of course coding. Come and join the quantum computing revolution!




3 thoughts on “To Reach Quantum Supremacy We Need to Cross the Entanglement Firepoint

  1. Some people incorrectly believe quantum annealing isn’t quantum computing. It’s more accurate to say that a quantum annealing QC isn’t the same as a gate model QC.

    D-Wave has an impressive list of customers/users, so they must be doing something right: Google, NASA, Los Alamos National Labs, Lockheed Martin, University of Southern California, US Government…

    1. If academia had trademarked the term “Quantum Computing” maybe they’d have a leg to stand on 🙂

      As it is, the machine uses qubits, entanglement has been demonstrated to be present and it clearly computes. If it walks like a duck …

      Quantum supremacy the way academia defines it, via the scaling behaviour is a tall order, to my knowledge it is still an open question if quantum adiabatic machines can clear that hurdle for certain problem classes. Nevertheless, I expect that this approach will definitely deliver a practical speed-up for many real world problems.

      IMHO the approach has also been strongly validated by Google first going for an adiabatic approach with their own chip. After all, copying is the most sincere form of flatery. It is simply the most pragmatic choice to do something useful with qubits, as it requires much less fidelity and control than gate based QC. Frankly, I am surprised that others, such as IBM for instance, just take a pass on this

Comments are closed.