And surprisingly, there are actually algorithms that your conventional Von Neuman laptop can do (or it's vectored processing GPU) that we simply haven't found a way to use a quantum effect we know of do. Most problems simply can't be done by a quantum computer at all yet. IBM is trying to push the envelope on a general purpose quantum computer but really a quantum computer as adaptable as our current computers are still not even on the horizon... This announcement is just that they figured out how to use a quantum effect to perform a certain type of calculation that would take conventional computers a very long time to do and just shows that we don't have a general purpose quantum computer yet.
It's actually the daily work of chemists. If they want the data of a so far not existing molecule, they synthesize it. It will then solve the exact Schrödinger equation in a couple of picoseconds. Then you can read out the result by NMR or so. The example of the spoon was just a tiny bit hyperbolic.
TL;DR version: D-Wave's quantum computer (which has been criticized by some in the scientific community for working more like a "quantum simulator" due to its low programability compared to IBM or Google's devices) solved magnetic systems of scales previously unsolvable/extremely time-consuming (an entire epoch, in fact) by conventional computing, they did this to prove that their device works for quantum computation involving quantum simulation and that this is a huge leap forward for their devices
Nice reference to Neo's "There is no spoon" D-Wave has a history of making such claims that they have done a computation that would take conventional computers excessive time, some of which have been subsequently debunked by example. In their early days, Scott Aaronson was skeptical. I hope he takes notice of their latest.and weighs in.
The funny thing about Scott and the other theorists who criticized D-Wave, is that they all got on the NISQ hype bandwagon when the field became lush with money. D-Wave pioneered the model of quantum overhype and then Google, IBM, etc picked up the torch around 2018. "oh but these are gate model machines, at least it can run Shor's algorithm one day!" (assuming a decades worth of developments that are not working yet). They criticized D-Waves baby steps and then turned around to expect applause for Google's baby steps.
If people are saying the Ising Model is a specific case that may not be relevant for real world computing, please note that the Ising Model is the basis of the Hopfield model which is a model of neural storage of associative memories in neuroscience. Quantum versions are predicted to have MUCH higher memory capacity than the classical version which only has capacity linear in the number of neural units (magnets).
Exactly again that idea of specialized purpose for memory. It’s why our brain separates cortexes based on function and purpose. Or why we keep multiple drives to separate our work on a computer. Essentially using a more organized memory structure to use memory space as optimally efficiently as possible. Also why we store memories that are associated in physical neuron proximity. Helps them fire off of each other
Actually, associative memories modeled after the Hopfield model have a capacity that scales less than linear in the number of units: O(N/log(N)), where N is the number of units, to be exact.
Solving the Ising model is an NP-complete problem. By definition, every NP-complete problem can be mapped to each other, so solving one solves every other. This is why it so relevant.
Even in conventional computing, annealing is a very interesting and cool way to solve problems. I've used simulated annealing myself, to optimise patterns in steganographic art. It's actually a lot of fun!
In classical computing, there is the analog computation, where for example a dial is twisted by one process and the machine spits out a result elsewhere after electrical signals have been manipulated. Not too different from the truly classical manual computer, where someone did calculations by hand, say, looking up as precise as possible a value on a graph. Because of these two historic parallels, i'm inclined to actually call this a calculation.
I agree but from another perspective: it takes numerical input (be it a configuration of qubits) and gives a numerical output, so it's a calculation. Simulation can also be, and often uses a lot of, calculations.
@@user-sl6gn1ss8p That is exactly what it seems to be, and truth be told, that can be extremely powerful given a question worth building such a system for
@@user-sl6gn1ss8pQuantum computing does not have to be general purpose in order to be disruptive. If you and I could make a quantum SHA reverse solver that can almost instantly defeat SHA and can only do that, that alone would make us billionaires. A device that instantly finds high primes would be another example of a non-general-purpose, yet very disruptive technology that could be achieved this way.
I think this is mostly just semantics. If it takes an input and produces useful and non-trivial output it is a computation. I think this is in the same class of computation as the DNA "computer" which can perform very NP search and mapping problems by "annealing" millions or billions of macromolecules in parallel at molecular speed. If the Everett interpretation is correct, this is pretty much how "Qbits" do it, just faster and parallelized over a chunk of a multiverse. Are both of these devices computers? Not if you definition of computer requires a Turing complete machine.
It’s also been found that growing slime moulds can solve travelling salesperson type problems as an alternative to simulated annealing in digital computers. But it’s much harder to raise venture capital funds to grow slime moulds, and also much less sexy for clients to look at than all the quantum plumbing. [edit: This was meant to be irony, focused on the fund raising activities of quantum startups. But evidently irony isn’t everyone’s cup of possums juice]
Except slime molds solve the problem by simultaneously traveling through all paths and then start optimizing the most successful paths. It's not really a novel solution that classical computers can't perform. The problem is that it doesn't scale well. Which is why it's a problem in the first place.
@@obsidianjane4413 Fair enough, just figured on a video about quantum processing, a little clarification about slime mold computation might be appreciated for those that don't know.
The definition of "computer" is an example of the imprecision of human languages. Computer was once a job description, of a person (usually a woman) who ran hundreds, to thousands, of math calculation during her work day. When applied to devices, it covered mechanical, and electronic analog computers, before being applied to digital computers. Some with the calculations "hard set" in their design. Others reconfigurable, or programmable. I've used analog computers back in college, which is "programmed" by wiring up a circuit. I've also made a living programming microcontrollers, which are "hard coded" most of the time to specific functions. Even though some (the old HC05) have the von Neumann architecture (most are Harvard), will you exclude them as computers, when their app is hard coded ? Language gets political, like that !
@@michaelmoorrees3585hard coded on a von neumann architecture is an oxymoron. Unless you are going to try the semantic dance of RAM vs ROM. A rather brief and futile exercise. The contemporary definition of computer is only sullied by early squabbles (across the pond) of who had the first one (that was Manchester). It's been a while, but I came to the conclusion that quantum machines cannot execute code... A requirement for a processor is to have NOT and AND (or gates to the equivalent). Quantum machines cannot do this. They cannot implement the recursive solution of Towers of Hanoi. What quantum machines *can* do is resolve. The "programming" is then just working out how to model a system in qbit terms.
Define a "computer". Your definition may not be the same as other people's, even if you were to consult a well respected dictionary or textbook. As someone pointed out, once upon a time "computer" was a job description for humans. It's a very imprecise term.
@michaelmoorrees3585 @@another3997 maybe you think you're disagreeing with me? There's a reason I put the word in scare quotation marks. The video is referring to how the word has come to be commonly used these days, which, as you note, isn't the "one true definition."
I'm glad you mentioned the spoon. Before you said it, I was thinking of "10 pebbles in your hand" where you throw them in a box and observe the exact positions down to 15 decimal places and thought "no quantum computer will ever be able to beat its accuracy in calculating/simulating the result of a throw"
My understanding of the discussion surrounding D-Wave being / not being a quantum computer has been that it is not so much the simulation vs computation part, but that the way they implemented the system. While a “proper” adiabatic QC tries to avoid an excited state, the D-Wave annealers happily leave the ground state and essentially rely on quantum tunnelling to bring them back down. Hence the debate whether a transition from QC -> non-QC -> QC is realistic. To date, none of the studies have shown that the D-Wave systems have a true quantum advantage, so will be interesting to see what the effect of this new one will be (at least after peer reviewed).
Yes, you are right that the quantum advantage has been questioned. But think of the spoon. The spoon also has a "quantum advantage". It's just not the sort of advantage that people usually discuss in terms of operations needed etc. I have little doubt that if you have a large quantum system like D-wave it will be faster than a conventional computer at *something* -- they just have to find that something. (And then there's the question whether that will be useful for anything, but that's a question you can also ask about the present-day universal quantum computers...)
@@SabineHossenfelder The "classical" quantum computers that can run Shor's algorithm are also good at something, namely running Shor's algorithm. They are as universal as D-Wave's quantum computer is. So much for one-trick ponies.
Objection. You say "you can't verify the solution because we don't have any computer that solves it". But... the Ising model is (according to Wikipedia, I wasn't aware of it before, but I trust it on this) equivalent to an NP-complete problem. NP-complete problems are well known for being hard to solve but very easy to check a claimed solution. This is one of their core features. A solution to an NP-complete problem is very easy to check by a simple computer. The difficulty is in coming up with it to begin with. So if the Ising model can be translated into an NP-complete problem, then just take the problem they solved, translate it, translate their claimed solution, then have a checker verify it. I didn't know about the Ising model and I know very little about the D-Wave quantum computer beyond what you say here, but as a computer scientist who knows about theoretical computer science, this claim that "because a problem is difficult to solve, it's difficult to check the solution" is outright wrong (and kinda basic in complexity theory), and I think it's wrong in this particular case too. Could you clarify??
@@geordierose3242 irrelevant. Np complete means it's both np hard and np itself. According to Wikipedia, Ising problem without external fields is np complete, and therefore NP, and therefore solutions should be verifiable on polynomial time by a classical computer.
No. NP Hard problems are optimization problems that return real numbers. NP Complete problems are decision problems (yes/no). The former can't be checked (you don't know if you've got a global optimum) but the latter can.
@@geordierose3242 I'm sorry but that's just not the definition of what NP hard means. See en.m.wikipedia.org/wiki/NP-hardness Now regardless of the name, you're right that is you're trying to find an optimal solution then it can both be harder and more importantly harder to verify. I do get a bit lost with the complexity properties of verifying optimality, but given how it's phrased in the video, and especially the argument "we can't check the solution because we can't find the solution" is just very misleading and potentially incorrect in the context of complexity theory. Maybe in this case it is because checking optimality of the solution is inherently tricky, but it should have been said with more care. And your definition of np hard is just wrong.
I think the combination of analog computers with digital is going to be the most powerful option in the near term. Eventually combining digital, analog, and quantum might be feasible - maybe in a few hundred years?
Optimization problems, the kind that D-Wave's approach solves, are some of the most common needs for high performance computing and get used for many things. Nobody has yet proven there is even a quantum advantage for any other useful algorithm. So far, it's basically Shor's algorithm which, outside of academia, is mainly of interest to thieves and national intelligence agencies and I expect both are behind the funding for many of those projects. Given that, I'd say D-Wave's strategy is currently the only real quantum computing. Google can compute arbitrary algorithms on their platform but most, if not all, will be no faster than their classical counterpart and far less efficient.
😂 Spoon analogy is at point! They just built physical model of exact process and measure results. Don't need quantum things to achieve "billions years worth of simulation" - we already have many lab simulators with sensors.
After a teacher or someone says "I'll come back to that later" I'm waiting for that the entire time, and I cannot focus/concentrate on the things the teacher "wants to say first".
it absolutely crazy, i have been trying to learn renormalization yesterday and they start using ising model as example, and now you are also talking ising model
The spoon analogy is not so silly. In the 1960s finite element methods were just beginning to be used in continuum mechanics (stress/strain analysis). Those methods were found to have limitations, as they still do today - especially in vibration analysis. Experimental stress analysis comes to the rescue. Basically, you build a device and test it. If you want to know how a spoon works, it's easiest to just make one.
I feel like it would make sense to call this an analog computer, similar to other types of analog computers where you create a circuit that obeys the same equations as the system you're trying to predict.
Super interesting! Question, is the "Salesman problem" suitable for quantum computers or simulations? The funny thing with science is that it sometimes find applications, after a new "tool" has been made , oder?
You mean the travelling salesman problem? Yes, it's one of the problems you can put on a quantum computer. I believe you can also do it by quantum simulation, at least in principle. However, it's a rather academic problem. There is a class of related problems in logistics that people want to look at with quantum computers. I think the reason everyone is starting with the Ising model is that it's particularly easy to match onto qubits.
@@SabineHossenfelder The travelling salesman is often used as an example of an NP-complete (and NP-hard) problem because it's easy to explain and less abstract than a lot of other NP-complete problems. However, NP-complete and NP-hard problems come up in practical situations all the time though, and if you've come up with a fast solution to an NP-hard problem, you've unlocked many others. In fact, if you find a polynomial time solution for an *NP-complete* problem, you've solved all of them because they can be converted to one-another. The problem is that nobody's been able to demonstrate that quantum computers are any better at NP-complete problems, and it's unlikely to be true. At best, it's speculation, but mostly it's marketing. If it were likely to be true, a lot more computer scientists and software engineers would actually care about quantum computers. Importantly, factoring integers into primes is probably NOT an NP-Hard problem. That's the big one that will break cryptography and everybody gets excited about it. Getting back to the OP's question, "is the "Salesman problem" suitable for quantum computers or simulations?" The answer is almost certainly "no". Like most problems, you can solve it with a quantum computer, but probably not significantly faster than a conventional computer. It's also worth noting that >99% of the time you don't even need an exact solution to an NP-Hard problem and a heuristic (fast but inexact solution) will do. So even if quantum computers were good at NP-Hard problems, they'd still be almost useless if they're at all more expensive or difficult to use. Technically speaking, classical computers might also be able to solve NP-complete problems in polynomial time - it hasn't been proven they can't. If you manage a proof that they can or can't, you will become very famous and win a million dollars because it's a millennium prize, and probably the one with the most practical applications.
You can put it on a quantum computer, just like you can with a conventional computer, but it won't help. The traveling salesman problem is NP-complete, and quantum computers cannot efficiently solve NP-complete problems. Quantum computers, at least the fully capable quantum computers that no one is actually building, solve a category of problems called BQP. BQP is a superset of P, the category of problems that can be solved by a classical computer, but a subset of NP, which includes many famous problems including traveling salesman. The most well-known problem in BQP is integer factorization.
Please, Explain time. Why do we say time passes slower or faster depending on your location and/or movement within the universe? Could it be it's just the measurement of time is different? Scenario: Calculate a future spot in earth's movement through the solar system and pinpoint it. Now pick three spots within the universe to observe the earth's movement until it reaches the calculated spot. Observation location 1 is NIST laboratory in Boulder Colorado. Observation location 2 is our moon. Observation location 3 is an orbit around the sun at near light speed. Now have all 3 observation locations start recording the passing of time at exactly the same time. All three observation loctions would stop the time recording when the earth landed on it's previously calculated spot. Would it be illogical to say, the physical amount of time passed for each observation location would be the same? Could it be that the recording device used to record the passing of time showed differently, but the physical amount of time would be the same for all 3 observation locations? Does it seem reasonable that what we need is a way to record time with a device that can account for its location and movement within the universe, so that the recording of time passing is equal on all accounts? The earth will move through the solar system and reach its calculated spot in the future in its due time. It won't be early and it won't be late. It will be right on time according to the calculation. Why would we say the passing of time is different for the 3 observation locations?
Enjoyed the report. Sabine is the best at illuminating progress in physics. Like DWave, A company in San Diego named Mem Computing has been doing quantum simulations and has shown incredibly abbreviated compute times compared to the linear digital computer. I believe there is a standards body that develops standard tests so that these companies that do q simulations can compete. For example there are tests for optimization and for finding primes. I am not an expert in this area but will try to find out more to add to this list of comments.
I did Sabines course. It is really nice but it is short and just the beginnings, finished it in one day. I wish they added more advanced quantum topics in the same style.
Thanks to you I checked out Brilliant and quite enjoyed it. Getting up to the end of the trial period though and I see that it's €100 a year, that's insane! There are so many good free resources already that I can't justify such a spend. I wouldn't mind if I could pay monthly for that but the monthly cost is something like the equivalent of $25 which is far too much.
It seems funny to me that some people do not think of a specialist machine as a computer. I can't imagine anyone saying that the machines onboard the Apollo rockets we're not computers - yet they could not do general computations of, for example, bookkeeping. Nor could they have played chess because they were hardwired to handle the specific calculations needed fr the mission.
Wow, this is incredible! I look forward to the day where we all have such a machine in our homes that can compute a very, very, very, very specific thing and nothing else.
You must walk before you can run, but I decided to become impressed by a quantum “thing” when it decrypts a now unbreakable message within a minute or a second or so, or when that thing finds the provable best move at every position in a game of go.
'Simulation' kind of implies 'computation', but hair-splitting aside, I think that (specialized) computation via setting up initial configuration of the system, and then letting that system settle into some kind of ground state is a really interesting approach. I may be wrong (and I'm happy if I'm right 50% of the time, especially about things that I have only glancing understanding of), but do I think that I can already see how such approach could be used for extremely fast factorization of public keys (used in public-key cryptography). What one would need is a (dynamically resizable) 2-dimensional grid occupied by fermions (only fermions will do, because of Pauli's exclusion principle), and with that grid occupying vertical position in a gravitational field. Start with the grid size of NxM (N horizontal positions [along x axis], and M vertical positions [along y axis], where N
It is a kind of analog computer, such computers were common long time ago until they were superseded by digital computers. Analog machines were very effective for solving differential equations. The difficulty with this quantum machine lies in finding the proper analog problem to mimic the problem you are trying to solve. I would love to see them solve the Travelling Salesman Problem as a demonstration of a useful application.
From the picture of D-Wave devices there are at least two of them. To check the accuracy of one you can run the same task on the other and compare the results.
In the paper at conclusions they claim: "We simulated square, cubic, diamond, and biclique topologies that are relevant to applications in materials science and AI, and are amenable to scaling analysis through the area law of entanglement and universal quantum critical scaling. More challenging and irregular topologies-for example those corresponding to deep neural networks". So if real world application exists and their results are applicable then we can test the quality of those results on the applications. Until this is done we should stay in a quantum state between amazed and anoyed.
It is like saying that an air tunnel used to test airplane wing designs is a computer. It is just a physical system very similar to the targeted one, so we can extrapolate some results with large accuracy.
If something gives you answers to one or more questions, based on input data, it has been computed. A wind tunnel simply moves "air" in a specific direction, at a speed which may or may not be controllable. It is analogous to reading parameter data from a storage device or keyboard. The tunnel itself doesn't recognise the presence of a wing, and it doesn't react its presence. That is different to the simulation within a quantum computer.
@@another3997 I meant the wind tunnel plus the particula wings, and the computation would be the lift, etc. the qbits just simulate quantum interactions, and the configurations need to change to emulate different systems.
Hello Sabine, I do my best to follow your channel, but often get lost in the labyrinth. Imagine yourself as a teacher of a class of 7/8 year olds, and me at the back sticking my hand up and asking, "Ms Hossenfelder, Ms Hossenfelder, it is not how that concerns me the most, but it is WHY?"
I wonder if the class of problems "solvable" this way is restricted to quantum phenomena, such as a large number of magnets, a large number of ions, etc. In other words, would quantum annealing be suitable for simulating systems where quantum laws are not the main actors, such as gravitational interaction of a billion objects for example? Also, how can we get verifiable results for these problems? We have some mathematical estimation methods for the magnets problem here, but imagine if we didn't, how would we be able to trust the results? Perhaps through experiment or observation of some natural phenomena.
Well, "quantum phenomena" is not as narrow a property as you might think. You see, all chemical, optical and electric properties of materials are strictly speaking quantum phenomena. This is why material science and chemistry is one of the major sectors of application for quantum computers. Physicists also have several algorithms that generally allow you to find solutions to some mathematical problems that could be either quantum or classical. Code-cracking is an often named example, but also some logistical and financial issues.
For many computational problems , checking the answer is correct is usually a faster process than finding the answer. Does this not apply to the Ising Model; could they not check the results from D Wave satisfy the problem on a classical computer?
Going back a few years I vaguely recall D-Wave teamed up with Boeing (or maybe Airbus) to do some fluid dynamics modelling but my memory may be faulty.
I'm really curious if the phrase "There is no spoon" would have got Sabine into some sort of copyright thing or if she just didn't feel like taking the time to google the exact quote and decided to paraphrase. One of life's little mysteries, I guess. FWIW.
What will these machines from D Wave, IonQ, and others actually be used for? Calculations for missiles, space flight, ingredient measurements for scientists trying to come up with cures? What might they actually be used for because I don't understand how trying to solve the positions of magnets makes it anymore valuable than a regular computer.
nice! ising model was my master project at university in late 90th, with a 25x25x25 grid. It's cool to see the problem is still there 😅! go Ising, crash the new gen computers!
I was making a molecule in Caligary true space Legacy Edition, it slowed down horribly at 200 meta balls, The project requires nanobots to go in there and create a nano lithography machine, to make progressively smaller nano machine's. Then you can hold a tesla giga fab in the palm of your hand. Send them to mars to produce 32oz diamond cut acrylic tumblers, just keep stacking the cups to make a little shelter for more rare and interesting materials, color the glass so you are amazed by the artifacts that have been actively arranged and collated.
"Quantum annealing" sounds like a type of optimization using, in essence, an analog computer. The biggest problem I've run into trying to use optimization for solving otherwise intractable problems is getting stuck in local minima. Often there appears to be an infinity of local minima that get infinitely close together, although I suppose the quantum aspect might resolve the infinitely-close-together problem. This can work if you are looking for a 'close enough' result, but if you need an exact optimal result, optimization can also prove intractable. I'm still waiting to be impressed by quantum computing.
