29 comments

[ 3.0 ms ] story [ 68.1 ms ] thread
The summary is two case studies:

- Searching for molecules with specific pharmaceutical effects.

- Solving ordering problems for cargo ship offloading (which feels like a traveling salesman sort of problem?)

All in all, I liked reading this article because it helps show places where quantum computing adds value by solving problems. I spend a lot of time in cryptography where the fear of quantum computing is that it will force a new class of slower cryptographic algorithms into use, so seeing positive uses of the technology is really quite nice.

> I liked reading this article because it helps show places where quantum computing adds value by solving problems.

The D-Wave is not a quantum computer. Quantum computers have quite different usage cases (but on the macro level, should be very useful to the pharma industry too).

Technically there are a number of architectures that qualify to be called Quantum Computers- you're probably just from the camp that does not admit the D-wave design as being one. Not all quantum requires require entanglement.
I'm also of the camp that does not consider “throwing confetti” to be anti-gravity, “dousing cancer cells in gasoline and lighting it on fire” to be cancer-treatment, and “swinging a laser-pointer between two stars” to be ftl-communication.
I like to troll and call our silicon base chips "quantum computers" because they work via a very quantum process, that is, the photoelectric effect.

They are not of course, and hopefully the smart people here can clarify what mechanically a "quantum computer" is doing, the best I can figure out is that it can reduce the complex valued quantum state down to a binary state and due to quantum interaction you can compute some things with this. If you asked me, it sounds like massive parallel computation on the atomic level, that is, similar to the way soap bubbles can solve minimal surfaces, but people keep assuring me this is not the case.

Did you mean the photoelectric effect? I think transistors really exploit our understanding of conduction and valence bands of electrons (https://en.wikipedia.org/wiki/Electronic_band_structure, how to move electrons to the conduction bands, and then bias them so a large current can flow (https://en.wikipedia.org/wiki/Field-effect_transistor#Effect...). IIRC part of the nobel for the transistor was because they (Shockley) had built a theoretical model based on QM

There's no (visible) photons involved in silicon computers, although very similar solid state and vacuum tube devices are used to exploit the photoelectric effect, mainly for scientific measurement purposes.

This is a very uncritical article that basically relies on marketing material of two companies (POLARISqb and SavantX). Most people in the QC field are extremely sceptical about D-Wave, since there is no conclusive demonstration that their system has any advantage over optimized classical algorithms.
I'm an outsider but still sceptical as NASA and Google turned to Dwave (circa 2013) and we saw pretty much nothing since.
If Dwave managed to sell snake oil to NASA and Google, they have the best sales in the world.
This is a rumor I've heard at the time. Take it with a massive grain of salt, it could be complete bullshit.

Google purchased the D-Wave machines as a retention attempt of a few employees (probably former physicists) that were threatening to leave. The understanding was that they could play with the machines in their "20%".

I remember a couple of years ago there was a kerfuffle about how Ars chose articles to write. I certainly don't remember the specifics, but it seemed like advertisers could commission puff pieces on a topic as a sort of a "rising tide lifts all boats" for whoever contracted the advertiser. This was how they claimed to avoid ethical concerns for writing the puff pieces.
To my knowledge, it is an open challenge to find real-world problem instances where existing quantum annealers outperform state-of-the-art classical alternatives. It's even hard to find _extremely contrived families of instances_ where practical computational benefits can be observed with quantum annealing. I recently contributed to one such benchmarking manuscript [1], which may indicate future promise as the number of qubits in quantum annealers continues to increase.

Anyway, as a result of that experience, I'm skeptical of the benchmarking efforts that led these two companies to the conclusion that quantum annealing is more cost-effective than dropping in a classical alternative as the inner optimization solver. D-Wave even has some reasonably efficient, open-source algorithms that can be used as a point of comparison (e.g., [2]). I'd be interested in reading more about the companies' benchmarks, as this article is very light on details.

[1] https://arxiv.org/pdf/2210.04291.pdf

[2] https://github.com/dwavesystems/dwave-neal

An interesting analogy that comes to mind is the electronic technologies that have tried to supplant silicon. I remember, perhaps as long as 40 years ago, reading that gallium arsenide could make faster transistors than silicon, and that we would soon have GaAs computers. Similar for other things like optical computing.

But every time GaAs cleared another hurdle, silicon also moved forward, kind of like the tortoise and the hare. There are certainly uses for GaAs, such as microwave amplifiers, but no GaAs computers yet that I'm aware of.

On the other hand we're looking at near-linear differences in speed with GaA, and it's easy to measure a pretty precise difference.

It's hard to even demonstrate how quantum annealing scales at all.

Gallium nitride transistors used in compact power transformers.
Or cryotrons or Josephson Junctions vs ultimately CMOS
These days lots of Startups use phrases like "we are doing...", "we are making...", when in reality they mean to say, "we are researching how to be able to do/make...".

It is not quite clear whether they have a working deployable product/service or whether it is R&D they are doing.

Minor trivia: The first major public demonstration of D-Wave used a modified version of my PerfectTablePlan software to show how D-Wave could solve a seating plan optimization problem. Seating plans are a difficult combinatorial problem with N! possible solutions for N guests in N seats.
Looks like it only solved 16 guest problem though? This sounds like something which is trivial to solve on modern PCs.
The D-Wave demo was just a demonstration that the technology works.

The standard version of PerfectTablePlan running on a standard PC can solve seating for 100 guests in a few seconds and 1000 guests is no problem.

To try to avoid confusion:

Most of what DWave have done is making a special-purpose analog computer that happens to use some quantum effects (maybe). It is an interesting analog computer, but not necessary a useful one. There is some hope that what they are doing would lead to what everyone else calls "scalable quantum computer", but for years they have been quite misleading in what they say. They have used scientific terms like "quantum computer" in ways that have little to do with what the rest of the scientific community means. They have lost a ton of good will to the point that I personally am extremely mistrustful of any claims they make. What they have done has significantly damaged the trust that the public had in my field of research.

I’ve never understood how D-Wave is more than a “bathtub accelerator”:

- fill your tub with water

- sit a computer at the edge that:

- plays music, vibrating the walls

- shines a light at the surface

- records the surface in HD

What does D-Wave actually do that I don’t, with my accelerator which solves the Bath Tub Problem faster than any classical computer?

At least mine is demonstrably faster: a bathtub can in real time solve diffraction problems that would take minutes to hours on a high end cluster using ray-tracing.

Are you intentionally referring to the use of vibrating water baths to emulate quantum behavior? https://news.mit.edu/2014/fluid-systems-quantum-mechanics-09... because if so that's a pretty funny joke.

Mentally I picture a quantum annealer being a lot more like spontaneously finding the ising model that optimizes the objective function through the use of quantum fluctuations.

I mostly agree, but I think there is a risk for a reader of your comment to completely confuse analog computers and scalable digital quantum computers.

Analog computers, like your bathtub accelerator, wind tunnels, ballistic aiming computers, scale models of flood plains, etc, are amazing at one single task (or maybe a few very related tasks), are not Turing complete, and, *by far most importantly*, can not scale as there is a hard insurmountable upper bound on the number of variables they can deal with before noise completely ruins the simulation.

Digital computers might be slower (they used to be in the vast majority of cases), but they can scale arbitrarily large, because digital computation permits error correction. That is true about classical digital computers (which have existed for 60ish years after being conceptually predicted 200 years ago) and about digital quantum computers (which do not exist just yet after being predicted 30 years ago). Digital quantum computers also happen to be able to solve efficiently a few important but idiosyncratic problems that classical digital computers can not solve efficiently.

I am writing all of this because your complaint, while fairly reasonable when made against any analog computer, is not correct when made against the (proposed) digital quantum computers (like the ones still in development at IBM, Google, PsiQuantum, or many research universities).

Lastly, it seems there is some limited resurgence in analog computation lately (classical or quantum, that does not matter), thanks to new technologies (mainly photonics) which provide a significant constant factor improvements in suppressing noise.