⚛️ quantum physics

Heralded Non-Gaussian Squeezed-State Inputs for Parity-Detection SU(1,1) Interferometry

This paper demonstrates that while non-Gaussian heralding operations like photon subtraction, addition, and catalysis can locally enhance conditional phase information in SU(1,1) interferometry, their success-weighted metrological performance ultimately falls short of an optimized Gaussian benchmark when accounting for fixed resource constraints and measurement mismatches.

Original authors: Lifen Guo, Qingqian Kang, Teng Zhao, Cunjin Liu, Xin Su, Liyun Hu

Published 2026-08-06

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Lifen Guo, Qingqian Kang, Teng Zhao, Cunjin Liu, Xin Su, Liyun Hu

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

Imagine you are trying to measure something incredibly tiny, like the distance between two atoms or the faintest ripple in a pond. In the world of physics, we use a tool called an interferometer to do this. Think of it as a super-precise race track for light. You send two beams of light down different paths, mix them back together, and look at the pattern they make. If one path gets slightly longer or shorter, the pattern shifts. The trick is that light is made of particles called photons, and they are jittery and unpredictable. This "jitter" creates noise that hides the tiny signal you are trying to find.

To beat this noise, scientists use quantum mechanics. Instead of using ordinary, messy light, they use special "squeezed" light. Imagine a balloon that is usually round and bouncy; squeezing it makes it long and thin in one direction but fat in another. In quantum terms, this means we reduce the uncertainty (noise) in the property we care about, at the cost of making the other property noisier. This is like focusing all your attention on one thing to see it more clearly. But even with squeezed light, there's a limit to how clear we can get. Some scientists have tried to make the light even stranger by performing "non-Gaussian" operations—basically, snipping out a photon, adding one, or swapping one in a very specific way—to see if this makes the measurement even sharper. The big question is: Does this extra complexity actually help us measure better, or does the trouble of creating this special light cancel out the benefits?

This paper dives into that exact question using a high-tech version of an interferometer called an SU(1,1) interferometer. Unlike the standard kind, this one uses special amplifiers (like optical microphones) to boost the signal before measuring it. The researchers tested three different ways to make the light "stranger" before it enters the machine: Photon Subtraction (snipping a photon out), Photon Addition (tossing an extra photon in), and Photon Catalysis (swapping a photon in and getting the same one back, but with a twist). They wanted to see if these tricks could beat the standard "squeezed" light in measuring a phase shift.

The story they tell is a bit of a plot twist. When they looked at the light after a successful "heralding" event (meaning they only counted the times the experiment worked and they got their special light), the non-Gaussian tricks did seem to win. Specifically, Photon Subtraction and Photon Addition made the measurement much more sensitive when the equipment was very efficient (high transmissivity). Photon Catalysis showed some promise, but only when the equipment was very inefficient (low transmissivity), acting like a filter that only works in the dark.

However, the paper's main conclusion is a reality check. The researchers realized that counting only the "successful" times is like judging a runner only on their fastest lap while ignoring all the times they tripped and fell. When they accounted for the probability of success—how often the experiment actually produces the special light—and the total energy used, the picture changed. Once they optimized the setup fairly, comparing the total information gained per attempt, none of the non-Gaussian tricks beat the standard Gaussian (squeezed) light.

In fact, the paper argues that the "winning" non-Gaussian states were often a bit of a trick. For example, Photon Catalysis created a state that had a huge amount of hidden information (high Quantum Fisher Information), but the specific way they measured it (checking if the light was even or odd, called "parity detection") couldn't see most of it. It was like having a treasure chest full of gold, but using a key that only opens a tiny, empty drawer. The state wasn't bad; the measurement tool just didn't match the state.

So, what's the final verdict? If you have a perfect, high-tech setup and you only care about the moments when the special light is successfully created, the non-Gaussian tricks can look amazing. But in a real-world scenario where you have to pay for every attempt, including the failures, and you want the best overall precision, the standard squeezed light is still the champion. The paper suggests that while these fancy non-Gaussian operations are cool and create interesting quantum states, they don't currently offer a practical advantage over the simpler, more reliable Gaussian methods for this specific type of measurement. The "magic" of the extra complexity was mostly an illusion created by ignoring the cost of failure and the mismatch between the state and the measurement tool.

Technical Summary: Heralded Non-Gaussian Squeezed-State Inputs for Parity-Detection SU(1,1) Interferometry

Problem Statement
Quantum-enhanced interferometry seeks to improve phase estimation beyond the standard quantum limit using nonclassical states. While non-Gaussian operations (such as photon subtraction, addition, and catalysis) can reshape photon statistics to potentially enhance sensitivity, their metrological advantage is often ambiguous. Previous studies have utilized different operation positions (input vs. internal), measurement schemes, and resource conventions, making direct comparisons of their advantages difficult. Furthermore, the "heralding probability" (the success rate of generating the non-Gaussian state) and the photon-number resources consumed during failed attempts are frequently not accounted for consistently. This paper addresses the need for a unified framework to evaluate whether heralded non-Gaussian inputs genuinely offer a precision advantage over optimized Gaussian inputs in a balanced SU(1,1) interferometer with parity detection, under explicit resource constraints.

Methodology
The authors develop a unified theoretical framework to analyze three specific heralded non-Gaussian operations applied to the input side of an SU(1,1) interferometer:

Photon Subtraction (PS): Mixing a squeezed vacuum with a vacuum ancilla and detecting photons.
Photon Addition (PA): Mixing a squeezed vacuum with an $m$ -photon Fock state ancilla and projecting onto vacuum.
Photon Catalysis (PC): Mixing a squeezed vacuum with an $m$ -photon Fock state ancilla and detecting the same number of photons.

Key Technical Components:

Unified Finite-Transmissivity Map: The authors derive a single operator map $\hat{K}_{\mu,\nu}(T)$ that describes all three operations based on the beam-splitter transmissivity $T$ and ancilla photon numbers. This allows for the calculation of closed-form conditional moments (mean photon number, variance, and pair coherence) at arbitrary operation orders without Fock-space truncation.
Quantum Fisher Information (QFI) and Parity Detection: The study calculates the conditional QFI for pure parity-eigenstate inputs and derives the classical Fisher information (CFI) for parity detection at the dark point ( $\phi=0$ ).
Internal Loss Modeling: Instead of simulating loss as a separate noise channel, the authors use the Integration Within an Ordered Product (IWOP) technique to "pull back" the internal loss through the interferometer. This results in a single effective parity observable $\Omega_\phi$ that acts on the prepared input state, allowing for exact analytic treatment of loss.
Resource Constraints and Benchmarks: The paper evaluates performance under three distinct protocols:
- Protocol I: Fixed preparation parameters (fixed seed), comparing the insertion of the heralding module.
- Protocol II: Fixed total two-mode energy of the successful conditional probe.
- Protocol III: Fixed sensing-arm photon exposure.
  Crucially, the authors distinguish between conditional information (given a successful herald) and success-weighted information (information per ideal module attempt, $P_j F_j$ ), which accounts for the probability of failure.

Key Results

Conditional vs. Success-Weighted Performance:
- Under fixed preparation parameters (Protocol I), single-photon subtraction (PS) and addition (PA) improve the conditional QFI and dark-point parity sensitivity over the Gaussian reference across most of the high-transmissivity regime ( $T \to 1$ ).
- However, when success probability is included (success-weighted), the information per attempt for PS and PA drops significantly because their success probabilities vanish as $T \to 1$ .
- Photon catalysis (PC) shows no conditional advantage for single-photon operations but opens useful low-transmissivity windows ( $T < 0.3$ ) for multi-photon operations ( $m \ge 2$ ).
Optimized Resource Comparison:
- When the coherent-squeezed allocation is independently optimized at a fixed conditional-probe energy and fixed interferometer gain ( $g=0.75$ ), the success-weighted Fisher information of all three non-Gaussian operations (PS, PA, PC) remains below the optimized Gaussian benchmark.
- The apparent advantages seen in fixed-seed comparisons disappear once the Gaussian reference is allowed to optimize its parameters (squeezing and coherent amplitude) to match the resource constraints.
Measurement Mismatch in Photon Catalysis:
- An expanded scan of PC reveals a "high-squeezing branch" with very high local QFI. However, this branch is dominated by photon-number variance (a parity-even component of the generator).
- Since parity detection is sensitive only to parity-odd components, it extracts very little of this encoded information. The authors identify this not as a state-preparation failure, but as a measurement mismatch: the state contains phase information that the specific binary readout (dark-point parity) cannot access.
Effect of Internal Loss:
- Internal loss deforms the effective parity observable. While it reduces the useful transmissivity windows, PS and PA retain a conditional parity advantage over the Gaussian input in the high-transmissivity regime even with loss ( $\eta=0.95$ ). Single-photon PC remains below the Gaussian benchmark under these conditions.

Significance and Claims
The paper draws a sharp boundary between "conditional non-Gaussian enhancement" (improvement given a successful event) and "practically available precision" (improvement per attempt under resource constraints).

Distinct Roles: The authors conclude that PS and PA serve effectively as conditional filters for dark-point operation when high-transmissivity modules are available, while multi-photon PC acts as a low-transmissivity filter.
Gaussian Benchmark: Under the tested constraints (fixed gain, coherent-plus-squeezed-vacuum family, parity readout, and explicit accounting for heralding probability), the optimized Gaussian input remains the preferred choice for per-attempt phase information.
Methodological Contribution: The work provides an exact analytic framework (unified map, pulled-back parity observable) that avoids numerical truncation and allows for consistent comparison of different non-Gaussian strategies. It highlights that claims of non-Gaussian advantage must be carefully qualified by the specific resource accounting and measurement scheme used.

The paper explicitly limits its scope to input-side operations, ideal herald detection, and specific loss points, noting that a full optimization over gain, general Gaussian probes, and ancilla generation costs is a separate problem. The results are presented as a conservative assessment within the defined constraints, showing that non-Gaussian filtering does not universally outperform optimized Gaussian strategies when resource costs are fully accounted for.

More like this