As vendors make steady progress toward quantum computing, there are a seemingly endless number of problems to solve and challenges to address. European researchers recently said they resolved a “decades-old debate” by running simulations on a GPU-powered supercomputer.
In a study published in the journal Nature earlier this month, scientists investigated a very special, complex, yet important puzzle about quantum annealing, a quantum computing technique that has been leveraged for over 20 years primarily by D-Wave Systems to develop the only commercially available quantum computer on the market today.
Nearly every other vendor working on quantum computing (Microsoft, Google, IBM, Honeywell, Atos, Amazon, etc.) is pursuing superconducting gate model systems. D-Wave has used annealing techniques since its inception, which company executives say will allow vendors to bring systems to market more quickly.
Although D-Wave announced in 2021 that it was building its own gate-model quantum system, its annealing system remains the foundation of the company’s business. This year the company unveiled its Advantage2 quantum processing unit (shown in the feature image above), now available on its Leap cloud service. In addition, D-Wave is equipping all of Leap’s QPUs with a new fast annealing protocol, and at its user conference last month, it unveiled its latest hybrid quantum-classical quantum solver to help solve increasingly complex nonlinear optimization problems.
While quantum annealing cannot be used to solve complex computing problems, it is particularly well suited to addressing optimization problems across the board, from workforce scheduling to supply chain management.
In search of a stable state
The problem the European researchers were trying to solve focused on how the properties of two-dimensionally arranged magnetic particles can suddenly change their behavior.
According to Esperanza Cuenca Gomez, relationship manager for quantum computing development at Nvidia, quantum annealing works by systematically decreasing the magnetic field applied to what she calls a “collection of magnetically sensitive particles.” “If strong enough, the applied magnetic field acts to align the magnetic orientation of the particles, similar to how iron filings become uniformly upright near a bar magnet.”
If scientists could slowly vary the strength of the magnetic field, the magnetic particles would arrange themselves in a way that minimizes the energy of their final configuration, Gomes wrote in a blog post.
“Finding this stable, minimum-energy state is crucial for a particularly complex, disordered magnetic system called spin glass, because quantum annealers can encode certain kinds of problems into a minimum-energy configuration of the spin glass,” she wrote. “Finding a stable configuration of the spin glass solves the problem.”
Knowing this allows scientists to create better algorithms to tackle extremely difficult problems. They can mimic the way nature deals with complexity and disorder. This is key to allowing quantum annealing and the applications it runs to solve computationally complex problems that classical computers cannot solve efficiently in a variety of fields, such as logistics and cryptography.
Gomez said the research also highlights an important difference in quantum computing approaches.
“Unlike gate-model quantum computers, which work by applying a series of quantum gates, quantum annealers allow quantum systems to freely evolve over time,” she wrote.
Understanding Phase Transitions
The researchers, led by Nobel laureate Giorgio Parisi, wrote that their work marks an important step in understanding how to build quantum annealing systems that can more quickly solve computational problems that would take significantly longer to solve using classical systems.
“Similar to metallurgical annealing, where ferrous metals are slowly cooled, quantum annealers seek a good solution by slowly removing the transverse magnetic field at the lowest possible temperature,” they write. “Removing the field reduces the quantum fluctuations, but forces the system to pass through a critical point that separates the disordered phase (large magnetic field) from the spin-glass phase (small magnetic field).”
What has been lacking until now is a more complete understanding of phase transitions.
“All hopes of achieving exponential speedups compared to classical computers are based on the assumption that the gap shrinks algebraically with the number of spins,” they write. “However, renormalization group calculations predict that a fixed point of infinite randomness exists instead.”
Researchers turn to GPUs
The debate among the scientists centered on questions about bridging the energy gap that separates the ground state and the first excited state. To resolve the debate, they used simulations running on Nvidia GPUs and say they found that there is some truth on both sides of the argument.
According to Nvidia, the research team used 2 million GPU computing hours at the Leonardo facility at the CINECA HPC center in Bologna, Italy, and about 160,000 GPU computing hours at the MeluXina GPU cluster in Luxembourg. An additional 10,000 GPU hours were run on the Spanish Supercomputing Network, a distributed computing environment of 12 interconnected supercomputers coordinated by the Barcelona Supercomputing Center, and the researchers also had access to the Dariah cluster in Lecce, Italy.
These systems were used to simulate the operation of quantum annealing systems.
The research comes as dozens of technology companies are developing quantum computing systems and components, much of which will be performed in the cloud as a service running in hybrid environments, with quantum systems expected to handle more complex computational workloads that classical systems cannot handle.
There will be room for both annealing and gate models in the growing field of quantum computing, according to Alan Baratz, president and CEO of D-Wave, which some forecasts say could grow the global market from $885.4 million last year to more than $12.6 billion by 2032.
“Optimization is a really important problem,” Baratz said at a D-Wave conference in June. “This is the low-hanging fruit for quantum applications today. D-Wave’s annealing quantum computer is great at solving these kinds of problems. Annealing quantum computers aren’t good at solving all problems — they’re not particularly good at quantum chemistry, for example, or designing long-lasting batteries. Gate-model systems are much better suited for those applications.”
