Murray Thom: Just to begin with, I think probably the best place to start is for you to tell us a little bit about how you got started in quantum computing.

00:02:08Andrew King: Well, I got started in quantum computing at D-Wave. So, I don't have an academic background in quantum computing or quantum information theory or anything like that or quantum physics. My background academically is as a graph theorist and doing algorithms and theoretical computer science. So, I was a postdoc and I saw the opportunity to come join D-Wave for working on a particular type of constraint satisfaction problem and I took it. There was somebody I knew who worked here, and so I was very interested in the opportunity to come and work on constraint satisfaction problems in a practical setting.

After I started working on that problem, I realized quite quickly that it wasn't a very good fit for the hardware. So, I started asking, "Well, what makes things a good fit or not so great for the QPUs that we have?" And following that thread of research has drawn me to trying to mitigate these issues that might come up or trying to play to the strengths of the QPUs. And part of that led to quantum simulation. So, what we have here is a programmable quantum material and so we can use this to do quantum simulation in a very natural way.

00:03:24Murray Thom: Yeah. I think that's an interesting point because what you're describing, I mean, I think it helps people to see a window into our lives developing quantum computers, which is that we have an idea of the machine we're going to build when we set out, but quantum mechanics is also something that there's so much that we can still learn about. And so, as we're going through developing and implementing the quantum processors themselves, we can measure them and deepen our own understanding of quantum mechanics and how we can use quantum mechanics in useful work.

Now, one of the things I want to make sure we get to right at the beginning is a scientific result that I describe as the most exciting development in quantum computing in the last two years. So, at a high level, can you tell us a little bit about the result that was achieved?

00:04:07Andrew King: Sure. So, we did a quantum simulation very fast. This quantum simulation was the simulation of the dynamics of programmable quantum magnets as they pass through a quantum phase transition. So, basically, we set up a race between the QPU, the quantum processing unit and a whole bunch of state-of-the-art classical methods that could potentially solve these problems and we raced them and the object of the game is to get equal or better solution quality to the QPU. And we found that for some of these simulations, the best classical approach would take almost a million years to solve the problems that we solved in just a few minutes using the QPU.

00:04:54Murray Thom: What a significant difference that is, right? Nearly a million years on the classical side, 20 minutes on the quantum computing side. I mean, when we're talking about making a comparison like that, what are the classical resources that we're discussing that would take nearly a million years to do these calculations?

00:05:09Andrew King: Oh, well, it's not a laptop. It's the world's fastest supercomputer and we can't do this calculation directly. We have to pretend that Oak Ridge National Lab is going to give us 100% access to their state-of-the-art supercomputer, which is obviously not going to happen. So, we got a little bit of the supercomputer and then divided the number by the proportion that we had. So, GPU-based supercomputer, which was at the time the fastest in the world, the Frontier supercomputer versus on little QPU.

00:05:39Murray Thom: And I think in the domain of benchmarking, it's important to recognize that we're talking about the very limits of computing, the absolute frontier of computing. So, it's like a demonstration that you do and you put it out there and then everyone else who wants to try to do better tries to compete with that. How long has this basically been standing?

00:06:02Andrew King: We demonstrated confidence in our result by posting the preprint on archive when we submitted to the journal and it can take six months to a year usually if everything goes well to get your paper published in a peer-reviewed journal at that level. And in this case, it took closer to a year and the result still stands and that was a year ago that it was published almost exactly actually.

People have been working at it and there have been improvement in the classical methods that have solved some of the problems up to the same sizes that we had demonstrated with classical methods, but not to the size that we solved on the QPU. But what's really interesting to me is the fact that these are three different modalities that have demonstrated beyond classical computation and they're all doing quantum computing. This is mind-blowing that you can just do quantum computing with different devices of that kind.

00:06:57Murray Thom: Yeah. No, it's true. It's remarkable. I mean, probably I think maybe the other thing is that there is an interplay between classical computing and quantum computing. In order for us to be able to make computers of this level of complexity and sophistication, we have to use classical computers in order to be able to operate them. But then there's also this context of running applications where we're sharing work between classical computers and quantum computers.

This result I think was significant because it was direct to the QPU rather than a hybrid context. So, I would say arguably the highest bar for a quantum computer to hit on its own. I mean, how would you articulate the division of effort between a classical computer and a quantum computer in this context?

00:07:42Andrew King: This is what I call a bare metal demonstration. So, you don't do any classical post-processing or post-selection or anything like that. So, when you're solving a problem for a customer, anything goes. It doesn't matter what compute resources you're using, what's important at the end of the day is the result that you get and the amount of money that you spend basically.

But when you're doing a scientific demonstration, there are certain rules that you want to stick to. First of all, you don't want to do something that you say is quantum that is not actually quantum and you don't want to be using classical resources in a way that significantly improves the appearance of the results from the QPU. So, I find the bare metal demonstrations much more interesting because I'm really just geeking out on these processors and I'm really interested in seeing what they can do by themselves. So, this would fall into that category.

00:08:40Murray Thom: Well, and I think how this speaks to people as well. I mean, sometimes I try to make an analogy to the industrial revolution. And during periods of the industrial revolution, we have created engines that can basically do useful work for us. Take an actual, like an engine like a motor we would have in a car or something like that, that is an engine which takes fuel and it produces rotary motion with torque.

At the time that that engine might get developed, there might not be tons and tons of applications for rotary motion with torque, but because we can do it so effectively, the amount of RPMs that we can produce, the amount of torque we can produce is so high relative to what we can do before we can start to transduce that into other forms of work.

So, I think to your point about this being like a bare metal QPU demonstration of advantage is that now there's this opportunity for everyone approaching this field and entering this field to understand that capability and look at how they can leverage it for a variety of applications. And that's what's significant about this. This is application relevant. I mean, maybe I'll jump to this point to what some of the applications are that this work is relevant to.

00:09:44Andrew King: Yeah. So, not everybody knows why you should be interested in solving the dynamics of a quantum phase transition. So, a phase transition is where you change a parameter of the system and you get a macroscopic reordering of the structure. The classic example is melting water and in a quantum system, you don't have temperature because there's no coupling to the thermal bath in the systems that we're talking about.

So, you change a magnetic field instead. And so, you go from a system that is totally disordered in this case to one that is a different disorder. Anyway. But what's important is that you're tuning a programmable magnetic material at the edge of one of its phases. And so, a potential application is in the research cycle in materials design.

So, you have to design the material a priori, you have to fabricate it, you have to test it, figure out all these parameters like where is the phase transition, for example. And then you need to send it back to the drawing board. You need to change some doping parameter, refabricate it. And so, if we can shortcut that process, then this could be an enormous accelerant for scientific progress and materials production and research.

00:10:55Murray Thom: Yeah. I mean, that's a good point. I mean, if we think about it from the perspective of like you were talking about like melting ice or even going the other way. We've started with water, it's a liquid, we can swim in it, there's a lot of disorder in that. Then the parameter that we adjust is temperature and all of a sudden, the properties completely change at a transition point, it becomes solid. There's a longer-range ordering, but it's not completely ordered like a crystal would be. And then you can skate on it or you can now move equipment across it or something like that.

So, if we were going to put this into the magnetic space, we use lots of magnetic parts. There are pieces that we use that snap into place, plugs and things like that, whose properties are magnetic. And you can imagine a scenario where you would switch them to where the properties are non-magnetic. So, sometimes they actually connect together and sometimes they release each other. So, the most immediate application would be in materials, design, material simulation.

I want to maybe push pause on the applications of discussion to introduce a really important dimension of this calculation, which has to do with energy efficiency. So, we talked about the time it might take the world's largest supercomputer to do these calculations relative to the time on the quantum computer. What about from an energy consumption standpoint, what were those metrics like?

00:12:12Andrew King: So, obviously if you're talking about a problem that's going to take Frontier 900,000 years to solve or something like that, you have to think about the energy burn of that. And so, some of these problems could take the entire worldwide electricity consumption for one year to solve. You can just say this is totally unreasonable in terms of both the amount of time and the energy draw that is required to solve the problem. So, you've got a huge advantage in energy efficiency.

So, one situation where you're trying to solve a lot of problems and you're trying to do it energy efficiently is Bitcoin or blockchain problems. So, blockchain problems are predicated on the idea that you have a certain number of small computational problems to solve and they have to be solved at a certain rate and it should be profitable to mine the currency if we're just talking about a cryptocurrency, proof of work cryptocurrency. It should be profitable to mine the cryptocurrency by solving these problems.

So, instead of taking a classical hashing problem, you take a quantum simulation problem such as the one that we solved in this work, you can actually build a blockchain application around that. And so, if you want to do that, you have to demonstrate a few things. You have to show that you can solve these problems consistently enough and you have to be able to solve it across multiple QPUs, otherwise it's not going to be practical. So, we actually wrote a paper on that and it's been published on the archive and is soon to be published in a journal.

00:13:48Murray Thom: It's interesting because like I had a conversation with someone about this point that to do these calculations on a classical supercomputer might require as much electricity as the entire world uses every year. And their response to me was like, "Well, why do you even calculate it to that level of detail? You can just say it can't be done. That's so much electricity. That's never going to happen." But I think it's precisely to make the point that the quantum computer to do those calculations required less than $1 worth of electricity and it helped maybe make more understandable the magnitude of the difference between classical computing and quantum computing.

00:14:22Andrew King: You have to blow their hair back a little bit with these numbers. I mean, there's a couple of reasons. First of all, if you're going to say something that is impossible, you need to quantify it in a scientific journal. That's one reason. And the way that you're going to justify this claim is by looking at your extrapolation curve, which needs its own set of justifications and say, "You are here and this has a number and the number is 900,000 years," or something like that. So, yeah, obviously nobody would spend 900,000 years trying to solve one of these problems.

00:14:52Murray Thom: Well, and it's helping people to understand what this new thing is. And the changes go deeper than that at the device level themselves. So, for instance, all of the computers and the chips that we're using in our phones and new chips from Intel and NVIDIA, as those have been developing and have been doing more work, they've also been consuming more electricity.

One of the interesting things about the quantum computers that we've built here at D-Wave is they sit in refrigerators that cool them down to really low temperatures, low temperatures that are more than a hundred times colder than interstellar space. Just pause on that for a moment. So, in our daily lives, in our laptops, the operating temperature of our chips is room temperature, but when we turn them on, they heat up. Whereas in our quantum computers, the operating temperature is low temperature, but when we turn them on, they don't heat up. And that's the significant part is that although the refrigeration system requires about 15,000 watts, what's arguably like 60 to 100 times lower electricity consumption than common supercomputers that are measured in the megawatts category, the actual quantum computer chips themselves consume less than one millionth of power of a mobile phone.

And that means that as we go through these multiple generations of quantum computers being developed, the power consumption of the systems are flat because the chips themselves are not contributing like a noticeable amount of electricity consumption. So, what's fascinating about that is the point that you're making. Blockchains for Bitcoin or other applications, that's a tool for allowing people to track a chain of events like a ledger or a series of transactions and that's something we can already do. It's just that when you look at what the quantum computer is, when you introduce it into that application, we estimate that can reduce the electricity consumption by a thousand times. That's the part that's new and novel that's getting introduced here by quantum computing.

00:16:44Andrew King: And it's also for better or worse, it's not just an energy consumption issue. It's an issue of being able to say the people who have access to these computers will be able to solve these problems and other people won't be able to.

00:16:56Murray Thom: Right. Yeah.

00:16:57Andrew King: So, you either restrict it or democratize it.

00:17:00Murray Thom: Now, the other thing is that those are both applications and material simulation, blockchain. Those are both things where we've actually demonstrated how these calculations can be used in those applications. I think the other thing that's interesting is that in artificial intelligence, like in 2024, John Hopfield and Geoffrey Hinton won the Nobel Prize in physics for Hopfield Networks. These are a certain type of neural network where the neurons are like affect each other bidirectionally. That was the Hopfield Network that John Hopfield proposed. And Geoffrey Hinton talked about the energy-based response of that neural network. That's a useful tool for artificial intelligence.

Well, that turns out to be exactly the machine instruction of the quantum computer, and it's like a quantum mechanical version of that. Now, that's an application that's currently being explored. How much involvement have you had in some of those explorations and the connections there?

00:17:49Andrew King: Not very deeply, I would say, but it's certainly a lot of AI practice is based on classical statistical mechanics and similar to blockchain applications is based on solving lots of not very hard problems classically. And so, the question is, is there a way that you can use quantum computing or specifically quantum annealing to either accelerate these applications or get them done with less cost and less energy cost? And so, we have some promising results, but you'll have to wait to find out more about that, I think.

00:18:27Murray Thom: Let's start to dive in a little bit more into this work here. We're talking about a demonstration of a calculation that a quantum computer can do that we don't think any classical computer could ever do in any reasonable period of time. No one would attempt it. What does it take to do a demonstration like that? How long, how many people did you involve, the organizations, how much classical computing time did you need? Can you tell us some of those?

00:18:47Andrew King: It was about a year and a half that we worked on this from when we started even exploring the paper to having it submitted. And very, very early in that process we figured out what we wanted to do and we demonstrated that you could do it with a QPU. So, that's not the hard part. The hard part is formalizing all the claims, figuring out what the classical methods are going to be and doing the work for the classical algorithms because you want to show that these problems are intractable. It's not a straightforward thing necessarily and also, it's not straightforward to do the classical calculations because they get so hard so fast and it's not very convincing to fit a line on two data points because who knows what this functional form is.

So, you need as much data from the classical computers as you possibly can get and each time you add another data point it like quadruples the amount of work that you have to do. And a lot of this work is just dealing with the classical supercomputer, doing the applications for the granted hours and then dealing with the queue and all this stuff.

So, yeah, figuring out how to solve the problem with the QPU is easy, figuring out how to optimize that is a little bit less straightforward and doing the classical stuff is way harder than any of that. But maybe I'm speaking as somebody who spent a little bit too much time with the quantum computer.

00:20:14Murray Thom: Well, yeah. I'd heard that there was 11 organizations worldwide involved in this work and that it took like over a hundred thousand hours of time on some of the world's largest supercomputers.

00:20:27Andrew King: Yeah. Well, we certainly don't have all the expertise in house that you'd need. So, you need somebody who really understands matrix product state algorithms well, really understands scaling of entanglement in these systems, peps, neural nets, et cetera, et cetera. So, you need a broad base of expertise that we got from our suite of world-class global collaborators on this work.

00:20:54Murray Thom: Well, and I think it just makes you appreciate the level of effort and work that you put in, that those dozens of researchers across those 11 organizations, the effort that they put in, because oftentimes I think I run into folks who are like, "Hey, why aren't there beyond classical demonstrations on every single application you would use a quantum computer for?" We're trying to share a bit of an idea of like, it's a lot easier to do the calculations than it is to prove that nobody else could ever do them.

00:21:25Andrew King: Yeah. And if you do something before everybody else, it's a lot more difficult to convince people, and so the work is that much harder. So, doing it first is the most difficult, but doing it at all is a huge amount of work. And so, if you find that we can do things a little bit better and we have found ways to improve the quantum methods since doing this work, we're not going to redo this whole exercise again just because we think we can add another zero to the advantage. You have to move on to other things. Well, maybe it's a privilege to be able to say, "Oh, we're just going to move on to more interesting things because doing this beyond classical computing with quantum computers is boring now."

00:22:04Murray Thom: I love that, I love that. Yeah, you heard it here first. I mean, the other aspect to it is that the goal here is to do the calculations, demonstrate the work and give folks confidence in it. They can actually go and explore it and attempt it. And part of laying those foundations wasn't just this single piece of work. There were five papers written in nature and science portfolio journals leading up to this in order to construct the case in a body of work.

I think oftentimes people feel like, "Oh, science is like when someone crosses the tape at the finish line, it's like, well, this person was first," but it's actually a much larger body of evidence. It's all the steps they laid to get up there. Everyone is using all of those references to help understand it.

00:22:53Andrew King: Yeah. So, I mean, a few years ago there was no demonstration that we could do coherent quantum annealing even. And so, that was the first step. So, coherent quantum annealing means that you don't have any thermal effects. And this is where you get really into not only the quantum effects that are interesting, but the quantum effects that are easy to understand and explain in an experimental context.

So, if you mix up thermal and quantum effects, it becomes quite difficult. I mean, that's great if that's what you want to do because we can do that, but doing it one or the other is more difficult. So, we did this in a very easy system to solve, a one-dimensional system. And then we moved on to optimization problems because people were not totally convinced that experimentally you could see a quantum speed up in optimization.

And so, we demonstrated this in a nature paper in 2023 using thousands of qubits in a three-dimensional spin glass, which is what a physicist would say is like the most beautiful optimization problem. So, we showed that you get a quantum speed up. So, it's a polynomial speedup, which means that if you double the amount of time that you do with a quantum computer, it's going to quadruple the amount of time you have to do and you have to spend solving this problem with a classical computer to get the same benefit.

So, basically that was a longstanding open problem and we solved that. But even then, that was not the end of the story because like I said, everybody knows you can't solve quantum simulation on thousands of qubits unless you're in some parametric limit. And so, we had to demonstrate that we were not in that parametric limit and that was the hard part.

00:24:38Murray Thom: There are folks listening to you talk about like one dimensions and three dimensions. One of the things I want to articulate is like we're trying to describe the way the world is working. And when we use mathematics, we can make descriptions that are incredibly detailed. So, if you think about like a string on a violin that's vibrating, it's fixed at two ends. There is a lot of information we can describe about at different notes which parts of the string are moving up and down and which parts of the string are not moving up and down.

So, restricting ourselves to a one-dimensional system gives us a simpler system on which to describe what's happening in a material that you were doing that in a magnetic material. As you go from a bow string to let's say like vibrations on the surface of a drum head, which is in two dimensions, now we've got a plane, not just a string. So, those vibrations are moving like the surface of the ocean, let's say.

And then you imagine going to three dimensions. Three-dimensional vibrations would be like sonar or clicks where you're like, you're creating vibrations. Those vibrations move out i