⚛️ quantum physics

Benchmarking Gaussian and non-Gaussian input states with a hybrid sampling platform

The paper introduces the Paderborn Quantum Sampler (PaQS), a hybrid platform that enables direct, semi-device-independent benchmarking of Gaussian and non-Gaussian input states in a 12-mode interferometer, demonstrating that non-Gaussian resources provide clear performance gains essential for achieving quantum computational advantage.

Original authors: Michael Stefszky, Kai-Hong Luo, Jan-Lucas Eickmann, Simone Atzeni, Florian Lütkewitte, Cheeranjiv Pandey, Fabian Schlue, Jonas Lammers, Mikhail Roiz, Timon Schapeler, Laura Ares, Milad Yahyapour, Alex

Published 2026-04-01

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Michael Stefszky, Kai-Hong Luo, Jan-Lucas Eickmann, Simone Atzeni, Florian Lütkewitte, Cheeranjiv Pandey, Fabian Schlue, Jonas Lammers, Mikhail Roiz, Timon Schapeler, Laura Ares, Milad Yahyapour, Alexander Kastner, Joschua Martinek, Michael Mittermair, Carlos Sevilla-Gutiérrez, Marius Leyendecker, Oskar Kohout, Dmitriy Mitin, Ronald Holzwarth, Jan Sperling, Tim Bartley, Fabian Steinlechner, Benjamin Brecht, Christine Silberhorn

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

The Big Picture: The Quantum "Dice Roll" Race

Imagine you are trying to build a super-computer that can solve problems no regular computer can touch. One way to do this is with a game called Boson Sampling.

Think of Boson Sampling like a giant, complex pinball machine (the interferometer). You shoot little balls (photons) into the bottom, they bounce around through a maze of bumpers and flippers, and then land in different buckets at the top. The goal is to predict exactly which buckets the balls will land in.

For a long time, scientists thought the only way to win this game against a classical computer was to use perfect, single balls (single photons). But making these perfect balls is incredibly hard, expensive, and slow. It's like trying to build a pinball machine where you have to hand-deliver one perfect marble at a time, waiting for it to land before sending the next one.

To speed things up, researchers started using "fuzzy" clouds of light (called Gaussian states or squeezed light) instead of perfect marbles. These clouds are easier to make and much faster. But there's a catch: these clouds are "softer" and less "quantum" than the perfect marbles. Scientists worried that by using these easier clouds, they might lose the very special "quantum magic" that makes the computer unbeatable.

This paper is about a race. The team at Paderborn University built a special machine (called PaQS) that can run the game with both types of inputs (perfect marbles and fuzzy clouds) side-by-side to see which one actually wins.

The Machine: The "Swiss Army Knife" of Light

The researchers built the Paderborn Quantum Sampler (PaQS). Imagine a kitchen where you can instantly switch between baking a cake with perfect, fresh eggs or using a pre-mixed batter, all in the same oven without cleaning it in between.

The Source: They use a laser that creates pulses of light.
The Switch: They have a special switch (an electro-optic modulator) that can instantly change the light from "fuzzy clouds" (Gaussian) to "heralded marbles" (non-Gaussian).
The Maze: The light goes into a tiny, programmable chip (the interferometer) that acts as the pinball maze.
The Detectors: At the end, they use super-sensitive cameras (detectors) that can count exactly how many photons hit each bucket.

Because they can switch between the two modes so quickly, they can compare them fairly. It's like running two different cars on the exact same track, at the exact same time, to see which engine is truly better.

The Test: How Do We Know It's "Quantum"?

How do you know if the machine is doing something a normal computer can't do? You can't just look at the final bucket counts; you need to check the "vibe" of the data.

The team used a clever trick called a "Non-Classicality Check."

The Analogy: Imagine you have a bag of marbles.
- Classical Light (Thermal): If you shake the bag, the marbles move randomly, but they don't "talk" to each other. They are independent.
- Quantum Light: The marbles are holding hands. If one moves left, the other must move right. They are entangled.

The researchers looked at the data to see if the photons were "holding hands" in a way that is mathematically impossible for normal, classical light. They did this by calculating a specific number (an eigenvalue).

If the number is positive, it's just normal light (boring).
If the number is negative, it's proof of Quantumness (the magic is happening).

The Results: The Surprise Winner

Here is where the story gets interesting. You might think, "The perfect marbles (single photons) should always win because they are the most 'quantum'."

But the results were surprising:

The Fuzzy Clouds (Gaussian Boson Sampling - GBS):
- When the light was dim (low energy), the fuzzy clouds showed strong quantum magic.
- However, as they made the clouds brighter (more energy), the magic disappeared. The data started looking like normal, classical light. It was as if the "quantumness" got diluted by the sheer volume of light.
The Heralded Marbles (Scattershot Boson Sampling - SBS):
- This method uses the "fuzzy clouds" but filters them to create specific "marble" events.
- The Winner: As they made the input brighter, the quantum magic in this mode got stronger and stronger. The more light they put in, the more "unbeatable" the computer became.

The Takeaway:
The paper shows that simply having "more light" isn't always better. In the "fuzzy cloud" mode, too much light kills the quantum advantage. But in the "heralded marble" mode, more light actually supercharges the quantum advantage.

Why This Matters

This research is a crucial step for the future of quantum computing. It tells engineers:

Don't just turn up the volume: If you use Gaussian states (the easy-to-make clouds), there is a "sweet spot" for brightness. If you go too bright, you lose your quantum advantage.
The Hybrid Approach is Key: By building a machine that can test different strategies on the same day, we can figure out exactly how to build the best quantum computers for real-world problems, like designing new drugs or solving complex logistics.

In short: The team built a fair referee to watch two different quantum strategies race. They found that while the "easy" method works well at low speeds, the "heralded" method is the one that can handle the high speeds needed for a true quantum super-computer.

1. Problem Statement

Boson sampling (BS) was originally proposed as a pathway to demonstrate quantum computational advantage using single-photon (Fock) inputs and click detectors. However, scaling these experiments is hindered by the difficulty of generating high-quality, deterministic single-photon sources. To overcome this, the field shifted toward Gaussian Boson Sampling (GBS), which uses deterministic single-mode squeezed vacuum (SMSV) states. While GBS enabled larger experiments, it relies on Gaussian resources, and the computational complexity advantage is sensitive to noise and loss.

Recent research has explored alternative architectures, such as Scattershot Boson Sampling (SBS), which uses two-mode squeezed vacuum (TMSV) states and heralding to effectively inject non-Gaussian resources (Fock states) into the interferometer. A critical open question remains: How do different input state resources (Gaussian SMSV vs. non-Gaussian heralded TMSV) compare in terms of performance and "quantumness" (non-classicality) as system brightness increases? Existing verification methods (like Total Variation Distance) are often exponentially expensive or sensitive to experimental imperfections, making fair, side-by-side benchmarking difficult.

2. Methodology

The authors introduce the Paderborn Quantum Sampler (PaQS), a hybrid experimental platform designed to perform sampling experiments with different input states within a single run to ensure commensurability.

Experimental Platform (PaQS)

Architecture: A 12-mode integrated photonic interferometer (silicon nitride, low loss) fed by a temporally multiplexed squeezed-light source.
Source: A waveguided PPKTP crystal generates picosecond squeezed pulses via Type-II Parametric Down-Conversion (PDC).
Dynamic Switching: An electro-optic modulator (EOM) and polarizing beam splitter (PBS) allow rapid switching (4 ns rise/fall time) between:
- GBS Mode: Two independent SMSV states interfere to create independent single-mode squeezed inputs.
- SBS Mode: The two modes of a TMSV state are separated; one is detected (heralded) to project the other into a Fock state input.
- Thermal Boson Sampling (TBS): The heralded mode is discarded, leaving thermal inputs (used as a classical baseline).
Detection: Eight Superconducting Nanowire Single-Photon Detectors (SNSPDs) with intrinsic Photon-Number Resolution (PNR) capability (up to 3 photons) via high-speed time-tagging (~2 ps jitter).
Control: A path-length compensation unit (PLCU) ensures temporal overlap of pulses from different time bins.

Benchmarking Framework

Instead of relying on Total Variation Distance, the authors employ a semi-device-independent certification framework based on normally ordered moments of the photon-number operator.

Metric: They construct a matrix of photon-number covariances, $C_{jk} = \text{Cov}(\hat{n}_j, \hat{n}_k)$ .
Criterion: For classical light, a specific matrix derived from these moments must be positive semidefinite. The observation of a negative eigenvalue in this matrix serves as an unambiguous witness of non-classicality (quantumness).
Advantage: This method is robust against experimental imperfections (like loss and phase drift) and does not require prior knowledge of the interferometer's unitary transformation.

3. Key Contributions

Hybrid Platform (PaQS): The first experimental platform capable of switching between GBS, SBS, and TBS configurations within a single experimental run (every 20 seconds), eliminating systematic errors associated with comparing separate experiments.
Semi-Device-Independent Certification: Application of a moment-based non-classicality witness that verifies the presence of quantum resources without needing full tomography or knowledge of the device's internal parameters.
High-Order SBS Demonstration: Implementation of SBS with PNR heralding, allowing for the generation of higher-order Fock states (beyond single photons) in a scattershot configuration.
Comparative Analysis: A direct, quantitative comparison of the "quantumness" of GBS and SBS datasets across varying mean photon numbers.

4. Key Results

System Characterization: The system achieved an average heralding efficiency of 38.4% and a signal path efficiency of 8.7%. The interferometer demonstrated a similarity of ~94-97% to target Haar-random unitaries.
Non-Classicality Verification:
- TBS (Thermal): As expected, TBS data showed no violation of classical bounds (positive eigenvalues), confirming it behaves classically.
- SBS (Scattershot): The SBS data exhibited a steady increase in non-classicality (more negative minimum eigenvalues) as the mean photon number increased. At high brightness ( $\langle n \rangle \approx 2.15$ ), the violation of the classical bound exceeded 36 $\sigma$ .
- GBS (Gaussian): The GBS data showed strong non-classical signatures at low mean photon numbers ( $\langle n \rangle \approx 0.17$ , >26 $\sigma$ violation). However, as brightness increased, the quantumness degraded significantly. At $\langle n \rangle \approx 2.15$ , the data failed to show quantum signatures for over 90% of input phase configurations, meaning it could be reproduced by classical resources.
Phase Sensitivity: GBS results showed strong temporal dependence due to phase-space asymmetry of SMSV states, whereas SBS results were stable, indicating robustness against phase drifts.

5. Significance and Implications

Resource Efficiency: The results challenge the intuition that "more squeezing" (higher brightness) always leads to better quantum performance in GBS. Instead, there appears to be an optimal squeezing level for GBS, beyond which the quantum advantage diminishes.
Superiority of Non-Gaussian Inputs: The study demonstrates that SBS (using non-Gaussian resources via heralding) outperforms GBS in maintaining quantum correlations at higher brightness levels. This suggests that for scaling up boson sampling applications, incorporating non-Gaussian resources (like heralded Fock states) may be more effective than simply increasing the brightness of Gaussian inputs.
Benchmarking Standard: The proposed semi-device-independent framework provides a practical, robust tool for the community to certify quantum advantage in sampling experiments without the prohibitive cost of full state tomography.
Future Directions: The findings highlight the need to explore hybrid architectures that combine the scalability of Gaussian states with the robustness of non-Gaussian resources, and suggest that future quantum advantage demonstrations may require moving beyond pure GBS protocols.

In summary, the paper provides a rigorous, side-by-side comparison proving that while Gaussian Boson Sampling is scalable, its quantum advantage is fragile at high intensities, whereas Scattershot Boson Sampling maintains and even enhances its quantum character as the system scales up.