Researchers at the Technical University of Denmark (DTU) and international partners have demonstrated that entangled light can cut the number of measurements needed to learn the behaviour of a complex, noisy quantum system by an enormous factor.
"This is the first proven quantum advantage for a photonic system," says corresponding author Ulrik Lund Andersen, a professor at DTU Physics.
"Knowing that such an advantage is possible with a straightforward optical setup should help others look for areas where this approach would pay off, such as sensing and machine learning."
The work appears in Science under the title 'Quantum learning advantage on a scalable photonic platform'. It is carried out in collaboration with colleagues from the US, Canada, and South Korea.
Entanglement is key
At the heart of the study is a problem that shows up across science and engineering: When you want to understand or characterise a physical system, such as a device, you do repeated measurements and, based on those, for instance, work out the "noise fingerprint" of the device.
In quantum devices, however, it is not as straightforward. For one, quantum noise is part of the measurements. Also, the number of experiments required for complex systems can scale exponentially with the system's size, so it quickly becomes impractical or even impossible. The researchers set out to find another way using entangled light.
Entanglement is a key concept in quantum mechanics where two particles or light beams are so strongly linked that measuring one instantly tells you something about the other.
"We built a process we could control and asked a simple question: Does entanglement reduce the number of measurements you need to learn such a system? And the answer is yes, by a lot. We learned the behaviour of our system in 15 minutes, while a comparable classical approach would take around 20 million years," says Ulrik Lund Andersen.
Something no classical system can do
After laying the theoretical groundwork in the 2024 paper 'Entanglement-Enabled Advantage for Learning a Bosonic Random Displacement Channel', the researchers knew that entangled light would likely solve the issue.
The experiment was set up in the basement at DTU Physics and runs at telecom wavelengths with well-known optical parts. It works even with ordinary losses in the setup. That matters, the researchers say, because it shows that the gain comes from how you measure, not a perfect measuring device.
In more detail, the system consisted of an optical channel in which multiple light pulses shared the same noise pattern. Two beams of light were prepared – or more precisely, squeezed - so they became entangled. One beam is used to probe the system; the other is there for reference. A joint measurement compares them in one shot, and that comparison cancels much of the measurement fuzz and pulls out more information per trial than looking at the probe alone.
Jonas Schou Neergaard Nielsen, an associate professor at DTU Physics and co-author of the paper, stresses that the researchers have not targeted a concrete real-world system yet:
"Even though a lot of people are talking about quantum technology and how they outperform classical computers, the fact remains that today, they don't. So, what satisfies us is primarily that we have finally found a quantum mechanical system that does something no classical system will ever be able to do."
