Harnessing dimensional leakage for continuous-variable randomness certification

In the quest to build secure communication networks and powerful computers, scientists rely on a fundamental property of nature: true randomness. Unlike the predictable patterns of a clock or the rolling of a die, which are determined by physical laws and initial conditions, quantum randomness is genuinely unpredictable. This unpredictability is the bedrock of modern cryptography, ensuring that secret codes cannot be cracked. To generate this randomness, researchers often use "prepare-and-measure" setups. In these experiments, a device prepares a tiny quantum system, such as a particle of light, and sends it to a second device that measures it. The outcome of this measurement is the random number.

For these systems to be trustworthy, scientists must be certain that the quantum systems they are using behave exactly as their theories predict. A common assumption in these experiments is that the system is confined to a specific, limited size, much like a message being sent in a box that can only hold a certain number of items. If the box is too small, the message cannot be faked by an eavesdropper. However, in the real world, especially when using light, this assumption is rarely perfect. Light exists in an infinite variety of states, and even when scientists try to restrict it to a simple, two-state system, tiny amounts of "leakage" occur. The system occasionally slips into higher, unintended states. For years, this leakage was seen as a fatal flaw, capable of destroying the security of the randomness and allowing a clever adversary to predict the outcome.

A team of researchers has now shown that this perceived weakness can actually be turned into a strength. By changing how they look at the data, they demonstrated that it is possible to certify genuine randomness even when the system leaks out of its intended boundaries. Instead of focusing on a single, specific setting for the experiment, they analyzed the results across all possible settings simultaneously. This approach proved far more robust, allowing them to certify randomness using simple, low-power laser setups that were previously thought to be too imperfect for the task. Their work suggests that the messy, high-dimensional reality of physical systems does not have to be a barrier to security, but can instead be a resource for generating secure random numbers in practical, everyday devices.

The core of the problem lies in how scientists traditionally verify randomness. In a standard test, researchers pick one specific way to prepare a particle and one specific way to measure it. They then check if the results are random. If the particle leaks even a tiny bit of information into higher energy states, an adversary could theoretically use that extra space to mimic the results while knowing the outcome in advance. This makes the randomness appear fake. The researchers found that this vulnerability is most severe when looking at just one pair of settings. It is like trying to spot a thief by watching a single door; if the thief knows which door you are watching, they can easily slip through a back entrance you aren't monitoring.

To solve this, the team introduced a new method of evaluation that looks at the entire ensemble of inputs. Instead of checking a single door, they monitored every possible door the thief could use. By averaging the results across all the different ways the particles were prepared and measured, they found that the adversary's ability to predict the outcome was severely limited. Even if the thief could predict the outcome for one specific setting by exploiting the leakage, they could not do so for all settings at once without introducing detectable errors. This averaging process smoothed out the vulnerabilities, allowing the researchers to certify randomness in situations where the old method would have failed completely.

This shift in perspective opened the door to using continuous-variable systems, which are based on the smooth, infinite properties of light waves rather than discrete, countable particles. Specifically, the team tested a setup using single-mode Gaussian states, which are essentially simple, squeezed pulses of light, measured by a technique called homodyne detection. In the past, such setups were considered too "leaky" to certify randomness because the light naturally occupies an infinite number of states. However, by applying their new averaging method, the researchers showed that these simple, low-squeezing states could indeed generate certified randomness. They found that with a leakage parameter of just 0.075, they could certify an average of 0.0313 bits of randomness per round. While this number seems small, it is a non-zero amount of certified randomness from a system that was previously thought to be useless for this purpose.

The researchers did not stop at simple light pulses. They explored more complex scenarios to see how much randomness could be squeezed out of these systems. When they replaced the standard light detectors with a more sophisticated method involving displaced photodetection, the amount of certified randomness increased significantly, reaching 0.082 bits. They also tested non-Gaussian states, which are more complex shapes of light that are harder to create but offer more potential. Using a specific type of non-Gaussian state known as a squeezed cat state, they achieved even higher levels of randomness, certifying 0.119 bits per round. Remarkably, these structured, experimentally feasible states performed nearly as well as the theoretical maximum possible within a truncated mathematical model, capturing about 95% of the optimal randomness.

The findings challenge the long-held view that dimensional leakage is purely a defect to be eliminated. In the context of their new framework, the higher-dimensional nature of physical systems is not just tolerated; it is managed in a way that allows for practical implementation. The researchers demonstrated that by accepting a small amount of leakage and averaging over all inputs, one can build a randomness generator that is both robust and experimentally accessible. This is a significant step forward for quantum technologies, as it moves the field away from the need for perfect, idealized components and toward systems that can be built with current, imperfect technology.

The work also highlights the importance of how we evaluate security. The study showed that the choice of estimator—the mathematical tool used to calculate randomness—is critical. A method that focuses on a single input pair is fragile and easily broken by leakage. In contrast, the average estimator is resilient because it forces any potential adversary to be unpredictable across the entire board, a much harder task. This insight suggests that future quantum devices might not need to be perfect to be secure; they just need to be evaluated correctly. The researchers conclude that their results provide a clear path for experimentalists to build continuous-variable randomness generators that are both practical and secure, turning a potential weakness of the physical world into a practical advantage.