D-Wave Systems Previews 2000-Qubit Quantum System
dwavesys.com
dwavesys.com
Inside is also a link to the whitepaper on how to program the DWave, which has a lot more detail on "ok, but what does that mean I can do with it, and how?". More or less, you need to map your problem space to the functioning of the DWave (a series of weights), and then you need to map the answer space of the DWave (out-state of each qubit) back to your problem space. The DWave doesn't actually return a canonical answer, but rather a bundle of statistics for each qubit, from which you then determine your answer (say, by taking the average).
Some things to remember, aside from the debate about whether it's actually a QC and actually uses entanglement to produce answers:
1) All 2k qubits are NOT entangled with each other The qubits are grouped into cells, and the cells have a coupling between them, but each qubit does not (directly) interact with all other qubits. This is a large part of why it's not a "general" quantum computer; it's more like an ASIC.
2) You program in "similarity" and "dissimilarity" to neighboring qubits, and an initial weighting. Each qubit in the dwave has some programmed possibility of being 1 or 0, and of being the same or different from each neighbor. "Running" the calculation more or less applies all these weights, and then you look at the resulting state.
3) The "answer" is actually the statistics on multiple measurements. After programming the weights, you run the machine, and get out an answer. You do this 50, 100, whatever, times, and now you have statistics on the state of each qubit. From this you determine your answer; AFAIK, usually you just take the average.
But it's not as awesome as scifi QCs :/
Did you mean s/ASIC/FPGA/ ?
In the D-Wave case it looks like the topology of coupling between cells is completely fixed.
I get the impression it's a lot like how one uses an analog computer.
Edit: removed an incorrect remark about entanglement.
I'm going to need a citation on that.
Even if only some small fraction of the qubits in the 2^2000 search space are fully entangled, that's still a massive improvement over the current state of the art using conventional computers because of how absolutely huge the space is.
We're still waiting for one problem, no matter how arbitrary, where the D-Wave beats basic consumer computers, let alone a few million dollars worth of computing power. And if D-Wave (or anyone else) had found one for their 2000-qubit computer, it would be in this press release. So no, any gain in performance form the D-Wave hardware is still in the "theoretical but without any theoretical backing" stage.
This is like saying that NASA has never gone to space, because there is not ONE thing they have EVER done in space, no matter how arbitrary, (including going to the moon) that I care about more than literature.
Let's look at the parts of my analogy:
NASA has not gone to space BECAUSE it is not better than something irrelevant.
DWAVE does not have a quantum computer BECAUSE it is not better than something irrelevant.
What on Earth do classical computers have to do with whether DWAVE is using quantum effects?
This is very irrelevant.
I can't believe so many people use this to "prove" that it's not a quantum computer. This is a complete and utter non sequitur.
This is very basic logic. It's like saying that until the minute a machine beat Kasparov at chess, computers couldn't play chess.
In a very narrow sense of cheating, sure, a computer could fake playing chess if a person is hidden and playing for the computer. Maybe the comouter isn't even powered on. Likewise, in a narrow sense a classical computer could be hidden in the DWAVE's cabinet and producing the output with a python script. Maybe the DWAVE isn't even powered on. (and has a cheater inside, a macbook.)
But short of this scenario it is a completely inappropriate point to make. It doesn't say anything about how dwave works.
-> Are you accusing them of not even powering it on and hiding a macbook in the case, which produces all its results? (if not, how is your contention different.)
My only objection is on this suggestion - that they have to find something it's already better at than a desktop PC (with literally billions of transistors).
I quoted the part I objected to.
That's not why I'm saying it's probably not a quantum computer. It's not a quantum computer because in depth analysis of its behavior has consistently shown that above the ~20 qubit scale it's better modeled by a classical explanation than a quantum one. We only really have evidence of entanglement at the 8 qubit level[1] so far.
The fact that the processing power is abysmal just makes the hype even more ridiculous. The point of that fact is that it's not like the EmDrive, where it's doing something we can't yet explain under the assumption that nothing interesting is going on. If it weren't for the fact we've looked inside, it could be running a basic Intel CPU for all we know. There's no smoke at any level, so it's nonsensical to give D-Wave a pass and think they've got some fire.
Being not-general-purpose means it's not a computer. If it's not a computer, it can't be a quantum computer.
As far as I can tell, D-wave's machines do appear to use quantum effects, but that doesn't make them quantum computers. They're quantum calculators.
I actually interviewed for a field engineer position with D-Wave embedded at Lockheed in Fort Worth. Apparently I wasn't smart enough for them because they went radio silent after the phone interview.
I'm a CFD guy and I have absolutely no clue how anyone expects to get any aerodynamics simulation results out of that with anything resembling usefulness.
Really, even a super-simple 2D wing profile simulation that you can run in milliseconds even in javascript will require data storage that's absolutely massive compared to 2000 qubits.
Quantum computers have several algorithms for various applications, such ashort shors for factorization. Not aware of any algos for CFD.
Perhaps a rarefied gas dynamics, bolts man distribution sort of application could be created? At least in some cases quantum mechanical effects are significant...
The one at USC definitely isn't being used for anything past the research level.
From my understanding, these things do beat classical computers, but no one cares because the problems they are solving is not useful for anything.
Remember your CPU requires quantum mechanics to explain its operation (above and beyond "that's why the atoms don't implode"), so "it does quantum things" does not a quantum computer make.
> D-Wave’s quantum system runs a quantum annealing algorithm to find the lowest points in a virtual energy landscape representing a computational problem to be solved.
In statistics and machine learning, this is great for finding optima in cost functions.
The ideas in this paper are related: https://arxiv.org/abs/1412.3489v2
Simulated annealing they do is not even the best algorithm to search nonlinear solution spaces.
Regarding the second line of your comment: I don't have comprehensive knowledge about the possible alternatives, could you direct me towards some sources?
http://news.mit.edu/2015/3q-scott-aaronson-google-quantum-co...
"In the current model of the D-Wave chip, there are 1,000 or so qubits [quantum bits], but they’re organized into clusters of eight qubits each."
"what the Google paper finds is that Selby’s algorithm, which runs on a classical computer, totally outperforms the D-Wave machine on all the instances they tested."
Now they've just made a machine of twice as much qubits but is there any performance advantage compared to the code run on the classical computer?
It might be the same kind of thing though ... I'm curious if it is.
Edit: What's the beef?
The search for good problems for the D-Wave (irrespective of whether anyone cares about their answers) is still on.
If you really think the problems you work on are that novel and up the D-Wave's alley, I suggest you email one of the academic teams looking for problems that the D-Wave is good at. They've yet to find any.
I've never heard anyone claim it'd be exponential. In fact, there are proofs showing it can't be.
This is interesting research in an interesting area; it's certainly not a scam. But it's also not what you seem to imagine, doesn't work like you're assuming, isn't useful for what you hope, and shows no signs of being headed in that direction. Yes, 2000 entangled qubits that could be used to solve arbitrary problems would be amazing, but D-Wave isn't making that, isn't claiming to be making that, and doesn't seem to be working on a way to make that.
http://www.digitaltrends.com/features/dt10-quantum-computing...
To avoid public disclosure that P=NP (and losing out on any way to monetize their discovery) they are hiding their work behind a "quantum computer" which is a vague and sophisticated enough cover (nobody really understands quantum physics) to dupe some customers while the P equations run in ring-zero of a normal CPU. What would be interesting is to read the sales contracts for these machines. My guess is they are written in such a way that D-Wave makes no promises or guarantee that they are actually using qubits to solve customer's problems, just that they promise to solve customer's NP problems with D-Wave computers, regardless of the method. Moreover, I'd bet that the language states that the machines sold are "equivalent" to a 2000 qubit computer and not necessarily a 2KQbit processors. In this way D-wave is off the legal hook if/when the NP=P solution is revealed by D-Wave or others. My second guess is that these sales contracts are protected by NDA's.
1. In practice classical computers (as in "Intel CPUs") are better at solving the specific problem D-wave is trying to solve (timewise and pricewise).
2. In theory quantum computers can not solve NP problems exponentially faster than classical computers (a strongly believed conjecture).
The essence of my argument still stands and you can easily recast my argument, if you cared to think about it, in a form that is consistent with your vastly superior knowledge of QM, QC, P=NP, etc. If your point #1 is true then how can DW sell a single computer? That makes no sense on the face of it. Do you propose that their customers are buying them in an act of charity or that they are too stupid to notice its a scam? Don't forget these machines sell for millions of Thalers. Clearly, to make a sale DW must have demonstrated some thing superior to the Intel or standard CPU in solving some kind of problem. My guesses at "thing" and "kind" are wrong but my argument stands.
The gist of the issue for me is that DW is claiming to cohere 2000 qbits when the "state of the art" is a handful, at best. So I say, let's see sales contract and check out what are they really selling. That point stands despite your objections too.
1. The "algorithms" are input by the users, not dwave. Moreover, you don't even have a way to input an NP-complete problem. You can only input the setup for a very specific type of optimization problem.
2. You're not comparing like with like. Even Dwave doesn't claim to have a general, gate based quantum computer like the sub-100 gate quantum computers you're talking about. The also don't claim coherent entanglement of all 2k qubits; they've still failed to convince anybody more than a handful get entangled together coherently during the operation of the device.
3. The dwave has yet to beat a laptop at any problem, let alone the NP-complete problems of their customers. The only people who buy D-Wave systems are those doing pure research.
4. The sales contracts aren't protected by NDAs. I've talked to people who work on Lockheed's D-Wave (at USC), and all their investigations are freely available on the Arxiv.
Thanks for point #4, I will investigate.
Now look at the cost of the D-Wave. A few million dollars, compared to the trillions you could make just in licensing fees if it pans out and you were the ones to crack the code. Considering how deep these companies' pockets are, wouldn't you take that bet? I would, and I side with the evaluations of some of the biggest D-Wave skeptics in academia.
They're better at prime factorization, but in a way that's well understood (quantum Fourier transform) and possible to simulate on classical computers.