Looking for a dose of reality *and* realistic optimism about quantum computing? Matthias Troyer, Microsoft distinguished scientist, plans to do just that in his ISC2021 keynote in two weeks – *Quantum Computing: From Academic Research to Real-world Applications*. He notes wryly that classical computers enjoy a roughly billion times advantage (op/s) over quantum systems at the moment. So why is Microsoft betting heavily on quantum computing, you ask?

“The first is quantum is disruptive. It will change computing, and it will enable computations that are just totally unimaginable classically. But quantum computers will not be general purpose. They will not solve all problems. They will be like purpose accelerators for problems for chemistry, material science, and we’ll find more [applications] in the future. It could be useful for machine learning or small data problems. We don’t know yet what else comes but it will be as special purpose accelerators,” said Troyer during a recent interview with *HPCwire.*

“One reason why I’m talking about this at ISC is because there is big skepticism from the HPC community on all of the claims on quantum. And one thing I want to do is to bridge between quantum to HPC because the people who are quantum experts are typically not the people who are HPC application experts. But as quantum will be an accelerator for HPC we need the HPC community to start to learn where quantum can help and we need to train people who will be affected to become experts in both classical HPC and quantum so they can drive quantum-inspired algorithms,” he said.

Troyer (and Microsoft’s) view of the evolving quantum landscape is nuanced and fascinating. On the one hand Microsoft has bet big on one of the least-well developed technologies for qubits (Majorana-based topological qubits). More near-term, it has planted a deep, broad stake in the race for quantum advantage with its Azure Quantum portal that provides access to different qubit technology platforms and quantum tools from several quantum technology suppliers.

In his ISC keynote (Tuesday, June 29) Troyer will explore quantum’s challenges and early successes. One item on his agenda is myth busting – here’s one, quantum computing will be great for big data problems. It won’t as he explains. Another is the unexpected pleasant exchange of learning between classical and quantum computing; it’s produced quantum-inspired algorithms that are already reaping benefits on classical systems, including a pilot application for improving MRI scans.

The ISC description captures Troyer rather ambitious talk: “I will describe the hardware and software architecture of quantum computers and discuss how they differ from conventional classical high performance computers. Based on this, I will also attempt to dispel myths and hype surrounding the field and present a realistic assessment of the potential of these devices and the specific application areas on which they are expected to have a large impact.”

Presented here is small portion of *HPCwire’s* discussion with Troyer. Don’t miss his full keynote at ISC. (A short bio of Troyer at the end of the article.)

*HPCwire:* Quantum computing isn’t new but has burst to the forefront in recent years. What is that’s attracted Microsoft and to the field?

**Matthias Troyer:** It’s the first time in millennia, basically, that we are learning to compute in a totally novel way. It will be a really disruptive, but we need to find out first, where is the biggest impact, where will it be much better than in classical computing? That’s what I’ve focused on for the last decade, to see what problems we can solve quantumly that we can’t solve classically. I came to quantum computing when I realized there are certain problems we will never solve classically, even if we build a machine the size of the planet, but which quantum computers can solve. I want to solve the classically intractable problems and build hardware for that.

There are three ingredients to do that. One is you need the quantum machine that we have to [still] build. The second is we need a quantum algorithm that can solve the problem better than a classical one. The third is the problem itself [which must be well-chosen and suitable]. So far, most of the research in quantum algorithms has focused just on the asymptotic speedup, but as I make the problem larger and larger, quantum speedup means the time to solve the problem on quantum hardware scales lower than on classical hardware. If you make the problem big enough at some point, the quantum computer will beat the classical one. That is the key foundation of quantum computing. As we make problems bigger – for certain problems – we can solve them better quantumly than classically.

*HPCwire:* That sounds like practical quantum computing remains distant?

**Matthias Troyer:** For me as an application scientist, I want to take one step further beyond that. Namely, I want to solve my problem in a reasonable time. Thus, I’m looking for problems where that crossover – where the quantum computer solves it faster than a classic one and where the time is short enough that it has an impact. So it’s exciting academically to know if we make the problem big enough, when it takes a billion years on the classic compute, it will only take a million years quantumly. That is beautiful theoretically. But to have an impact we need to solve the problem within days or weeks. We’re looking at the problems where quantum computer can solve it in a few days, a few weeks, and outperform a classical one. That will be one of my talk’s themes.

So where is this practical quantum advantage, where I can solve an interesting problem in at least a comparable time or beat the classical computer. The reason why this is important is that classical computing has made tremendous progress. We can put billions of transistors on a chip, we build the big machines, and clock them fast. Quantum bits and quantum computers will be much more complex than a transistor. In fact, we will need thousands of transistors to control a quantum bit. If I compare a classical chip that we have now – like an Nvidia GPU – to a quantum wafer that I could imagine building in the next decade, I [still] see the classical chip has a big advantage because we can pack more transistors on it, the clock speed is faster, and the cost is lower. The quantum computer will have fewer cubits than the classical one has got transistors, and they are slower.

There’s this constant disadvantage that we must make up with a quantum speedup. I’ll show the numbers (in my keynote) and it comes out to be a factor of about a billion in constant difference. We can do about a billion more operations [per unit time] on the classical chip that we have today then on the future quantum computer we plan to build. But the quantum computer now has to do fewer operations because of quantum speedup.

*HPCwire:* What does that mean, practically speaking?

**Matthias Troyer:** There are two immediate lessons. First of all, being constantly slower [means] that loading data will be slower. Reading in data is slow. That’s why one of the myths that I want to bust is that quantum computing will solve the big data problem. The answer is [it won’t] because that machines will run at a slower clock speeds and will have fewer qubits than the classical one will have transistors, and loading data is even more of a problem. The focus of quantum problems has to be on small data problems that are big compute problems.

The second thing is, [since] we have this factor of about a billion more operations on the classical chip than the quantum chip, we need to look at quantum algorithms that have a quantum speedup that is way beyond that factor of a billion. When you look at many of the applications proposed, they are based on a quantum algorithm called the Quantum search or Grover search which has a quadratic speedup. This means you’re doing quadratically more operations on the quantum computer than on the classical one. If you have to do quadratically more operations than it takes a while to overcome that factor of roughly a billion advantage the classic computer has. The applications that will have an impact early on are ones with more than quadratic speedup and ideally ones with exponential[i] speedup.

**HPCwire: OK, what are those applications? In its ‘enthusiasm’ the quantum community has labelled just about every application as a candidate for quantum computing.**

**Matthias Troyer:** The impact early on, meaning the next decade, will be on the small data problems where quantum computers have attempted exponential quantum speedup. With that we can really whittle the range of applications, that that stand out. One is cryptanalysis using Shor’s algorithm for code breaking. But the interesting application will be in the simulation of quantum systems for problems in material science and chemistry, inventing new catalysts for organic fertilizer, production for carbon fixation, new materials, so really in changing material science. That’s where the big impact will be.

When you hear [quantum computing] will be useful for big data, I mentioned why I don’t see that as [an appropriate] application because loading data will be harder. Similarly, often mentioned problems like protein folding or drug design, weather forecasting, climate modeling, these are all applications where the algorithms proposed so far are based on Grover search, and that’s where the quadratic speedup means the cost of times for quantum to beat classical systems are too long, years or decades. So that’s what will not be [good] applications unless we focus on the chemistry and material science to address the real application.

*HPCwire:* What about the adapting of quantum methods for use on classical systems? There seems to be a lot of activity there.

**Matthias Troyer:** Yes, that’s really the big impact for HPC. As we’ve developed quantum algorithm theory in the past decade, like in quantum machine learning and in quantum optimization, there were cases where we had a glimpse of an exponential quantum speedup. Then researchers thinking about those efforts started to realize they could de-quantize these algorithms; so they could invent a classical algorithm modeled after the quantum one that could run on classical hardware, and have almost the full advantage of the quantum one.

With that, the exponential advantage of the quantum algorithm one over the new classic one has been shrinking. For the quantum information scientists that’s a disappointment, but for the application scientist who wants to solve a problem that is actually great news because it reflects that thinking about quantum and developing quantum algorithms. If new classical algorithms that are exponentially faster than the classical state of the art, then implementing those algorithms on high performance computers will give us an impact from quantum computing today. So in that sense, we don’t need any qubits to have a big advantage from quantum research.

*HPCwire:* Are there a couple of examples you can cite of that?

**Matthias Troyer:** Yes. One example is optimization that came from the idea of quantum annealing. To solve optimizations problems, we realized we could implement that effectively, meaning as well as on the quantum hardware, [and] we developed quantum-inspired optimizations, as we call them. We have used them for the few problems in healthcare and finance, logistics and traffic. There’s one example of work with the Case Western University in Cleveland for optimizing pulse sequences for magnetic resonance imaging. That was a field where the experts had told us that if one can get 30 percent improvement, that would be a breakthrough.

The question was, how can we optimize the pulse sequences to make the scans faster or better. Many people are skeptical and told us this can’t be done; the problem is too hard. Now of course, that’s just the challenge I like. The team used the quantum inspired algorithm and found pulse sequences that could do scans of the same quality in 1/3 of the time. That means it takes 15 minutes instead of the 40-to-45 minutes. You can make the scans faster and more easily. With a cancer scan [for example] it’s much easier for the children to lay still for 10 minutes than for one hour. It’s a big breakthrough in healthcare. You can do more in less time or in the same time, you can make the scan 30% more accurate and see more details and find some smaller artifacts.

*HPCwire*: Switching gears for a moment, could you comment about the various competing qubit technologies?

**Matthias Troyer:** It’s great to see the focus and enthusiasm advancing in many qubit technologies. My goal is, as I mentioned, to solve classically intractable problems, like in chemistry. When you work through an example, as in chemistry, something that is beyond what a classical computer that can do now, we realize that in order to get there we need a million or more fast and high quality cubits. The challenge for all technologies will be how do you build a device with a million cubits that are good and fast? That the runtimes are short enough? And how do you use scale to the million cubits?

That’s why we focus on the full stack, the software, the control and finding a qubit that can scale. With all the qubit technologies, there are challenges to overcome and it is too early to tell [which will be best]. We bet on the new topological qubit that we are developing something that is small, fast, good, and will let us scale. Every qubit technology has to look at the challenges to scale and what needs to be solved there. Since we’re nowhere close to the fundamental limits for any of the [qubit] technologies, we first have to understand better where they can go before one can make a statement on which the technology can take us to a million[qubits].

__Troyer Bio__

Matthias Troyer is a Distinguished Scientist at Microsoft and affiliate faculty at the University of Washington. He is a Fellow of the American Physical Society and Vice President of the Aspen Center for Physics. Troyer is a recipient of the Rahman Prize for Computational Physics of the American Physical Society for “pioneering numerical work in many seemingly intractable areas of quantum many body physics and for providing efficient sophisticated computer codes to the community.” He is also a recipient of the Hamburg Prize for Theoretical Physics. He received his PhD in 1994 from ETH Zurich in Switzerland and spent three years as a postdoctoral researcher at the University of Tokyo. Later, Troyer was professor of Computational Physics at ETH Zurich until joining Microsoft’s quantum computing program at the beginning of 2017. At Microsoft he works on quantum architecture and leads the development of applications for quantum computers. His broader research interests span high performance computing, and quantum computing, as well as simulations of quantum devices and island ecosystems.

[i] Algebraically, linear functions are polynomial functions with a highest exponent of one, exponential functions have a variable in the exponent, and quadratic functions are polynomial functions with a highest exponent of two.