D-Wave Insights: What our latest simulation research tells us and what it doesn’t
Recently, a team of researchers from D-Wave and Google, including myself, set out to measure the speed of quantum simulation in a D-Wave quantum annealing processor. Today, we published a peer-reviewed paper on our experiments in Nature Communications. In it, we reported a first for quantum computing: in an application scientists are interested in independent of quantum computing, the quantum simulation vastly outperformed the industry standard classical method. Quantum computing holds the promise to solve some of society’s most important problems, but that promise is often clouded by hype. I’d like to break down what our results tell us, what they don’t, and what it means for the future of computing.
Insight #1: We observed what actually goes on under the hood in the processor for the first time
Quantum annealing — the approach adopted by D-Wave from the beginning — involves setting up a simple but purely quantum initial state, and gradually reducing the “quantumness” until the system is purely classical. This takes on the order of a microsecond. If you do it right, the classical system represents a hard (NP-complete) computational problem, and the state has evolved to an optimal, or at least near-optimal, solution to that problem.
The D-Wave processor was used to simulate a programmable quantum magnetic system.
What happens at the beginning and end of the computation are about as simple as quantum computing gets. But the action in the middle is hard to get a handle on, both theoretically and experimentally. That’s one reason these experiments are so important: they provide high-fidelity measurements of the physical processes at the core of quantum annealing. Our 2018 Nature article introduced the same simulation, but without measuring computation time. To benchmark the experiment this time around, we needed lower-noise hardware (in this case, we used the D-Wave 2000Q lower noise quantum computer), and we needed, strangely, to slow the simulation down. Since the quantum simulation happens so fast, we actually had to make things harder. And we had to find a way to slow down both quantum and classical simulation in an equitable way. The solution? Topological obstruction.
Insight #2: To achieve a speedup, we had to slow the quantum escape act down
Imagine a Möbius strip — it has half a twist in it. Now imagine a Möbius strip with a full twist in it, but the twist is… imaginary. That’s how we start the quantum simulation, which then naturally untwists itself by reconfiguring the entire lattice. This takes a while, allowing us to measure the timescales accurately.
An example of a Moebius strip. The research team made the simulation task slower and harder by “twisting” the lattice topologically. This made the quantum simulation slow enough to measure accurately.
The standard classical algorithm for this kind of simulation is “path-integral Monte Carlo” (PIMC). We additionally programmed PIMC with four-qubit updates that make it faster in this particular simulation. Even with this extra tuning of the classical system, the D-Wave chip was much faster — up to 3 million times compared to a CPU. More importantly to physicists, the advantage was biggest for the coldest (i.e., slowest, hardest, most quantum) simulations. More importantly for computer scientists, the advantage was biggest for the largest simulations. This means that we not only saw a speedup, but we also observed it on some of the hardest and largest problems. And perhaps most importantly for quantum computing in general, this was all done on a real application.
What does that mean exactly? We simulated an exotic phase of matter, which has already been studied by computational physicists using PIMC on extremely similar lattices. J. Michael Kosterlitz and David Thouless were awarded the 2016 Nobel Prize in physics for developing the surrounding theory back in the 60s and 70s. This puts the result in stark contrast to results on engineered or random problems of little practical relevance.
Insight #3: It’s not quantum supremacy (but it’s an important step on a long journey)
These experiments are an important advance in the field, providing the best look yet at the inner workings of D-Wave computers, and showing a scaling advantage over its chief classical competition. All quantum computing platforms will have to pass this kind of checkpoint on the way to widespread adoption. But let’s be clear about the claims, this is not a quantum supremacy experiment, and it is not a demonstration of superiority over all possible classical competition — it’s possible to make highly specialized algorithms to simulate the lattice after you know the lattice’s properties. This could benefit D-Wave’s customers and the broader quantum annealing ecosystem, too, in the form of faster hybrid algorithms. But ironically, these hybrid algorithms would be too fast to measure, so wouldn’t tell us much about the underlying physical processes at play.
The future is hybrid, but to move forward we need to continue to understand the advantage conferred by D-Wave’s quantum processor. These experiments fill a crucial gap in our knowledge, and contribute to a growing body of work positioning quantum simulation as an early success story for quantum computing, just as Feynman envisioned. In this work we have seen several unexpected properties of the quantum simulation that turned out to be correct once we checked against the classical simulation. This is encouraging as it means that we’re peering into a complex system and have begun to observe and verify the intricacies of quantum simulations.
One tiny step at a time, we are learning new details of quantum magnets by simulating them on quantum computers. And one tiny step at a time, we are also getting closer to realizing the full potential of quantum systems in practical applications. In the meantime, it’s critical we continue driving the science of quantum computing forward through quantum research and application development.
If you’re a developer or researcher who wants to learn more about our work or explore the D-Wave Advantage quantum system yourself, you can sign up for free on our Leap quantum cloud service. If you are a business ready to get started in quantum computing and want extra help to explore potential quantum use cases that would benefit your business, you can sign up for the D-Wave Launch quick start program here.
