⚛️ quantum physics

Optimal multi-spectral squeezing via deterministic 2D-phase optimization

This paper introduces a deterministic, linearly-scaling sequential algorithm for optimizing 2D phase masks that outperforms black-box machine learning methods by increasing mode-matching efficiency and boosting multi-spectral squeezing from -2.08 dB to -2.64 dB in waveguide-based quantum systems.

Original authors: Bastien Oriot, Peter Namdar, Emilie Gillet, RL Rincon Celis, Valentina Parigi

Published 2026-06-19

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Bastien Oriot, Peter Namdar, Emilie Gillet, RL Rincon Celis, Valentina Parigi

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 listen to a very faint whisper (a quantum signal) in a noisy room. To hear it clearly, you need a "local listener" (a local oscillator) that matches the whisper perfectly in pitch, volume, and timing. If your listener is even slightly out of sync or shaped differently than the whisper, you miss parts of the message, and the signal gets lost in the static.

In the world of quantum computing, scientists use light to carry information. A major challenge is making sure the "listening beam" perfectly overlaps with the "speaking beam." If they don't match, the data is lost, much like trying to pour water from a square cup into a round hole—some of it spills out.

The Problem: The "Black Box" Approach
Previously, scientists tried to fix this mismatch using complex computer programs (machine learning) that acted like a "black box." They would throw random adjustments at the problem and hope the computer figured out the best shape for the light. While this sometimes worked, it was slow, unpredictable, and didn't always find the best possible solution. It was like trying to tune a radio by randomly spinning the dial without knowing where the stations are.

The Solution: A Deterministic "Step-by-Step" Guide
The authors of this paper developed a new, smarter way to tune the light. Instead of guessing, they created a logical, step-by-step recipe (a deterministic algorithm) that guarantees the best possible match.

Here is how their method works, using a simple analogy:

The Pixel Grid: Imagine the surface of the light beam is divided into a grid of tiny squares (pixels), like a mosaic.
The Anchor: They pick one central square as a "reference anchor." This square stays fixed.
The One-by-One Tune: They then look at every other square on the grid, one at a time. For each square, they ask: "If I twist the phase (the timing) of just this one square, how much does it help the whole picture match the anchor?"
The Perfect Sync: They find the exact twist that makes that specific square "sing in harmony" with the anchor. Once they find the perfect twist, they lock it in and move to the next square.

Because they do this one piece at a time, and because the math proves this method always finds the global maximum (the absolute best solution), they don't need to guess or run thousands of random trials. It's like tuning a choir by having the conductor fix one singer's pitch at a time until everyone is perfectly in harmony.

The Results: Clearer Signals
When they tested this method in their lab:

Better Overlap: They improved the "visibility" (how well the beams matched) from 76% to 84%. Think of this as increasing the clarity of a radio signal.
More Efficiency: This small percentage jump actually meant a 20% increase in the efficiency of catching the quantum data.
Quantum Proof: When they used this improved setup to measure "squeezed light" (a special quantum state), the quality of the measurement jumped from -2.08 dB to -2.64 dB. This confirmed that their better matching directly led to better quantum data.

Why This Matters
This paper shows that you don't need complex, unpredictable AI to solve optical problems. By understanding the physics of the light and using a logical, sequential approach, you can get a guaranteed, optimal result much faster. It turns a chaotic "search for the best" into a reliable, efficient process, making quantum experiments more robust and easier to repeat.

The authors note that while they fixed the timing (phase) issues perfectly, there is still a tiny bit of loss because the shape (intensity) of the light isn't a perfect match, which requires a different kind of fix. However, their new method has successfully solved the phase-matching puzzle, paving the way for more efficient quantum measurements.

Technical Summary: Optimal Multi-Spectral Squeezing via Deterministic 2D-Phase Optimization

Problem Statement
In continuous-variable (CV) quantum information, particularly for measurement-based quantum computing (MBQC), generating large cluster states requires high-quality multimode squeezed light. While waveguide-based systems offer broad bandwidth and compatibility with pulsed operation, they suffer from spatial mode irregularities caused by fabrication defects and inhomogeneous geometries. These irregularities degrade the mode overlap between the quantum signal and the classical local oscillator (LO) during homodyne detection. Since the observed squeezing is fundamentally limited by detection efficiency ( $\eta$ ), where $\Delta X^2_{obs} \approx 1 - \eta$ for strongly squeezed states, imperfect mode matching ( $\eta_{mod}$ ) imposes a severe constraint on measurable squeezing levels.

Existing strategies to improve spatial overlap, such as using spatial light modulators (SLMs), typically rely on machine-learning-based optimization. While effective, these "black-box" approaches often lack reproducibility, offer limited physical insight, and do not guarantee convergence to a global maximum. Furthermore, they can be computationally expensive and inefficient.

Methodology
The authors propose a deterministic, sequential algorithm to optimize the spatial phase profile of the LO using an SLM. The method is grounded in a physical description of the optical process rather than agnostic machine learning.

Pixel Basis Representation: The LO spatial mode is imaged onto the SLM and decomposed into a basis of square macropixels ( $p_i$ ). The visibility ( $V$ ) of the interference between the LO and a weak coherent seed beam (serving as a proxy for the quantum signal) is expressed as the magnitude of the sum of contributions from these pixels:
$V(\phi_0, \dots, \phi_n) = \left| \sum_i \langle f_{seed}, p_i \rangle \langle p_i, f_{LO} \rangle e^{i\phi_i} \right|$
Sequential Optimization: The algorithm exploits the fact that the visibility function is a sum of individual pixel contributions without cross-terms dependent on pairs of pixels. By selecting a reference pixel ( $p_0$ ) at the beam center, the algorithm optimizes the phase ( $\phi_i$ ) of each subsequent pixel $p_i$ sequentially.
Phase Alignment: For each pixel $p_i$ , the system scans the relative phase $\phi$ between $p_0$ and $p_i$ to find the value that maximizes the two-pixel visibility $V_i(\phi)$ . This aligns the phase of $p_i$ with the reference $p_0$ .
Experimental Implementation: To isolate specific pixels during the scan, a linear phase ramp is applied to the target pixels, deflecting their light into the Fourier plane where an iris blocks unwanted contributions from other regions. This allows the system to measure the contribution of a single pixel against the reference without interference from the rest of the mask.

Key Contributions

Deterministic Global Optimization: The paper introduces an algorithm that provably reaches the global maximum of visibility in a pixel basis. Unlike random search or gradient-based machine learning, this method guarantees convergence.
Linear Scalability: The computational complexity scales linearly with the number of pixels ( $n$ ), as the $n$ -parameter problem is reduced to $n$ single-parameter optimizations.
Physical Interpretability: The approach provides a clear physical understanding of the optimized mode, avoiding the "black-box" nature of machine-learning techniques.
Efficiency: The optimization process is rapid; for a $10 \times 10$ grid, the algorithm completes in under 5 minutes, with runtime dominated by signal acquisition rather than computation.

Results
The method was tested in a waveguide-based setup generating multi-spectral squeezed light at telecom wavelengths.

Visibility Improvement: The optimized phase mask increased the interference visibility between the seed and the LO from 76% to 84%.
Efficiency Gain: This visibility increase corresponds to a 20% gain in mode-matching efficiency ( $\eta_{mod}$ ).
Squeezing Enhancement: Multi-spectral squeezing measurements confirmed that the improved mode matching directly translated to quantum readout performance. For the fundamental Hermite-Gauss (HG0) spectral mode, the measured squeezing improved from $-2.08 \pm 0.03$ dB to $-2.64 \pm 0.02$ dB.
Consistency: The observed squeezing improvement aligns closely with the theoretical prediction derived from the visibility gain using the efficiency-loss model ( $\Delta X^2_{obs} = \eta \Delta X^2_{state} + (1-\eta)$ ). Across other modes, the mean efficiency rose by 21%.

Significance and Claims
The authors claim that deterministic spatial phase shaping offers an effective and interpretable route to enhanced multimode squeezing in waveguide platforms. The work demonstrates that by focusing on the specific physical constraints of the optical process, one can develop optimization routines that are not only more efficient and reproducible than machine-learning alternatives but also theoretically guaranteed to reach the optimal solution for phase-only shaping.

The paper notes that the remaining visibility gap (not reaching 100%) is likely due to intensity-profile mismatches that cannot be corrected by phase-only shaping, as well as potential parasitic reflections. However, the results establish that the proposed deterministic algorithm is a robust tool for maximizing the performance of homodyne detection in complex, multimode quantum systems.

Technical Summary: Optimal Multi-Spectral Squeezing via Deterministic 2D-Phase Optimization

More like this