⚛️ quantum physics

Characterizing Photonic Partial Distinguishability Through Gram Matrix Tomography

This paper introduces a robust protocol called Gram-matrix tomography that characterizes the partial distinguishability of independently prepared photons by reconstructing their internal state overlaps from a polynomial number of measurable invariants, thereby enabling accurate prediction of multiphoton interference statistics without requiring direct access to the unknown internal degrees of freedom.

Original authors: Marcin Kotowski, Marco Robbio, Oliver Reardon-Smith, Benoit Seron, Leonardo Novo, Ernesto F. Galvão, Michał Oszmaniec

Published 2026-10-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Marcin Kotowski, Marco Robbio, Oliver Reardon-Smith, Benoit Seron, Leonardo Novo, Ernesto F. Galvão, Michał Oszmaniec

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Light is often treated as a stream of identical particles, but in the delicate world of quantum physics, even the slightest difference between two photons can ruin a calculation. When scientists build machines to process information using light, they rely on a phenomenon called interference, where waves of light combine to create patterns. This works perfectly only if the photons are truly indistinguishable, meaning they are identical in every way, including their color, timing, and polarization. However, in real-world experiments, photons are rarely perfect. They might arrive a fraction of a second apart or have slightly different frequencies. This "partial distinguishability" acts like a hidden flaw, introducing errors that can turn a powerful quantum computer into a device that classical computers can easily mimic. To fix these machines, researchers must first understand exactly how different the photons are, but this is a difficult task because the internal properties that make them unique are often invisible to the detectors used to measure them.

A team of physicists has now developed a new method to map these hidden differences without needing to see the internal details directly. Instead of trying to measure the internal state of each photon individually, which is often impossible, they focused on the relationships between the photons. They realized that all the information needed to predict how a group of photons will behave in an experiment is contained in a specific mathematical structure that describes how much each photon overlaps with every other one. The researchers created a protocol to reconstruct this structure, which they call a Gram matrix, by measuring how the photons interfere with one another in a controlled way. Their approach is efficient because it does not require knowing the complex internal nature of the photons, such as their exact frequency or polarization, and it works regardless of how many internal properties might be involved.

The team demonstrated that by measuring a specific set of interference patterns, they could build a complete picture of the photons' distinguishability. They proposed two different ways to gather this data. The first method uses a series of specialized optical devices, known as Fourier interferometers, to target specific relationships between pairs or small groups of photons. The second, more streamlined approach uses a single, randomly configured optical device to collect a vast amount of data at once. By analyzing the patterns of how many photons arrive at different detectors in this single shot, they can mathematically reverse-engineer the relationships between the photons. Both methods allow the researchers to calculate the Gram matrix, which serves as a blueprint for the photons' behavior.

Crucially, the researchers proved that knowing this blueprint is enough to predict the outcome of any future experiment involving these photons, even if that experiment cannot see their internal differences. They showed that if the reconstructed blueprint is accurate, the predictions made from it will match the actual experimental results with a high degree of certainty. They also provided strict mathematical guarantees on how many measurements are needed to achieve a certain level of accuracy. Their analysis revealed that the number of measurements required grows with the square of the number of photons, which is a manageable rate for current technology. This is a significant improvement over older methods that might have required an impossible number of measurements as the system grew larger.

The team also addressed the reality that photons are rarely perfectly pure; they often have a small amount of "mixedness" or noise. Their method accounts for this by separating the errors caused by the limited number of measurements from the errors caused by the photons not being perfectly identical. They showed that even with imperfect photons, their protocol can still provide a reliable description of the system, provided the photons are close enough to being pure. This makes the technique robust enough for real-world laboratory conditions where perfect control is impossible.

In their tests, the researchers compared the two data-gathering methods using computer simulations. They found that while the single-shot method is experimentally simpler because it requires only one setup, it tends to produce slightly larger errors for the same amount of data compared to the method that uses multiple specialized setups. However, the single-shot method remains a powerful tool, especially for certifying that a quantum experiment is working as intended without needing to reconfigure the equipment constantly. The simulations showed that both methods could successfully reconstruct the relationships between photons, with the errors decreasing as more data was collected.

This work fills a critical gap in the field of quantum optics. Before this, scientists could tell if photons were indistinguishable or not, but they could not fully reconstruct the specific way they differed to predict complex multi-photon events. By providing a way to map these relationships efficiently and with guaranteed accuracy, the new protocol offers a practical tool for improving quantum devices. It allows engineers to benchmark their light sources and understand exactly how much partial distinguishability is affecting their results. As the field moves toward larger and more complex quantum experiments, having a reliable way to characterize these subtle differences will be essential for building machines that can truly outperform classical computers.

Technical Summary: Characterizing Photonic Partial Distinguishability Through Gram-Matrix Tomography

Problem Statement
Partial distinguishability among photons is a critical factor in multiphoton interference, central to large-scale photonic experiments like Boson Sampling and various quantum technologies. Uncontrolled distinguishability introduces errors in linear-optical gates, reduces sensitivity in quantum-enhanced interferometry, and can degrade quantum computational advantages to efficient classical simulations. While the theoretical framework for partial distinguishability relies on Bargmann invariants (multivariate traces of internal states), characterizing these internal degrees of freedom is experimentally challenging. Often, the internal states (e.g., polarization, frequency, arrival time) cannot be directly accessed or their dimension is unknown, while experimental setups only manipulate and resolve spatial modes. Existing methods can certify indistinguishability but lack a rigorous protocol to fully reconstruct the relational information (including collective phases) of partially distinguishable photons and provide quantitative error bounds for subsequent predictions.

Methodology
The authors propose Gram-matrix tomography, a protocol to reconstruct the Gram matrix $G$ of independently prepared photons with pure or nearly pure internal states. The Gram matrix elements are defined as $G_{ij} = \langle \psi_i | \psi_j \rangle$ . For pure states, $G$ contains all relational information required to predict the statistics of any experiment insensitive to internal degrees of freedom.

The protocol operates in two main stages:

Data Acquisition: The protocol requires measuring a specific set of $O(n^2)$ Bargmann invariants. The paper introduces two complementary experimental strategies to access these invariants using passive linear optics and photon-number resolving detectors:
- Fourier Method: Uses targeted Fourier interferometers to directly measure specific cyclic invariants (e.g., pairwise overlaps and higher-order cycle traces). This approach measures each invariant separately.
- One-Shot Method: Uses a single generic (Haar-random) interferometer. It collects output photon-number correlations (2-body and 3-body) and reconstructs the invariants via linear inversion of the measured correlators. This approach avoids reconfiguring the interferometer for each invariant.
Reconstruction and Error Analysis:
- The protocol reconstructs the Gram matrix $G$ up to physically irrelevant phase choices (diagonal unitary conjugation) by fixing a gauge using a spanning tree of the overlap graph.
- It explicitly separates errors into sampling error (finite data) and systematic error (deviations from purity).
- The authors derive rigorous bounds linking the estimation error of the Gram matrix (Frobenius norm) to the trace distance between the true and reconstructed external states (the state accessible to the experiment). This trace distance bound directly controls the total variation distance (TVD) of outcome distributions for any subsequent linear-optical measurement.

Key Contributions

Reconstruction Guarantees: The paper establishes that reconstructing the Gram matrix $G$ (up to phase) is sufficient to reconstruct the external state $\Omega_{ext}$ . It provides explicit error bounds showing that the TVD between predicted and actual outcome distributions is bounded by $\frac{\sqrt{n}}{2} \|G - \hat{G}\|_F$ .
Sample Complexity: The total sample complexity is shown to be independent of the internal Hilbert-space dimension. Under mild assumptions (pairwise overlaps bounded below by a constant), the protocol requires $O(n^5/\delta^2)$ samples for the Fourier method to achieve trace-distance accuracy $\delta$ . A lower bound of $\Omega(n^2/\delta^2)$ is proven, indicating that quadratic growth with photon number is unavoidable for general external-state reconstruction.
Robustness to Impurity: The framework is extended to nearly pure product states ( $\rho = \bigotimes \rho_i$ with purity $> 1-\epsilon$ ). The protocol approximates the dominant pure components, with the total error bounded by a combination of sampling error and a systematic term proportional to the impurity $\epsilon$ .
Experimental Protocols: Two distinct, operationally grounded protocols are detailed. The Fourier method offers higher precision per sample but requires multiple interferometer configurations. The one-shot method offers experimental simplicity (single configuration) but involves linear inversion which can amplify variance, requiring more samples for equivalent precision.

Results

Theoretical Bounds: The authors prove that for pure product states, the external state can be estimated with high probability to accuracy $\delta$ using $O(n^5/\delta^2)$ samples (Fourier method) or $O(n^3/\delta^2)$ under specific conditions. For mixed states, the error scales with the impurity parameter $\epsilon$ .
Numerical Simulations: Simulations comparing the two methods on systems of up to 7 photons show that while the one-shot method is theoretically viable and well-conditioned for random interferometers, the Fourier method generally achieves lower reconstruction errors for the same number of experimental shots.
Gap in Bounds: The numerical results indicate that the theoretical upper bounds on error are somewhat pessimistic; in practice, the observed errors are significantly lower than the bounds predict, particularly for the one-shot method where the variance behavior differs between second- and third-order invariants.
Scalability: The method successfully reconstructs external states for systems where full relational information is not recoverable by lower-order measurements (e.g., "circle dance" states), though the efficiency degrades as the number of particles increases due to the need to estimate exponentially decaying higher-order invariants.

Significance
The paper claims to fill a gap in the characterization of photonic sources by providing a rigorous, model-independent method to reconstruct the relational information of partially distinguishable photons without resolving their internal degrees of freedom. By linking the reconstruction of the Gram matrix to operational guarantees on the trace distance of the external state, the protocol provides a quantitative tool for benchmarking photonic sources. The authors position Gram-matrix tomography as a practical tool for assessing the quality of multiphoton sources and for protocols that rely on photonic indistinguishability, offering a robust alternative to methods that only certify indistinguishability without fully characterizing the underlying state. The work emphasizes that these guarantees hold even with finite sampling and deviations from purity, making it applicable to realistic experimental scenarios.

Technical Summary: Characterizing Photonic Partial Distinguishability Through Gram-Matrix Tomography

More like this