⚛️ quantum physics

SSP-QST: Spectral Subspace Purification for Photonic Quantum State Tomography

This paper introduces SSP-QST, a lightweight, rank-adaptive post-processing method that enhances photonic quantum state tomography by purifying noise-induced eigenmodes through a Weyl-motivated noise floor, thereby significantly improving reconstruction fidelity and shot efficiency without requiring iterative optimization or prior rank assumptions.

Original authors: Anuvab Sen, Saibal Mukhopadhyay

Published 2026-07-27

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Anuvab Sen, Saibal Mukhopadhyay

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 single, perfect note played on a violin in a concert hall. In the world of quantum physics, scientists use special "notes" called entangled light particles (photons) to measure things with incredible precision—far better than anything we can do with ordinary tools. This is called quantum sensing. To know if their "violin" is playing the right note, they have to take a picture of the sound, a process called Quantum State Tomography (QST). Think of this like trying to reconstruct a song just by listening to a few seconds of it through a slightly broken radio.

The problem is that real-world radios aren't perfect. They hiss, crackle, and sometimes add fake notes that weren't there. In quantum terms, this "static" is called noise, and it comes from two places: the universe being a bit fuzzy (finite shots of measurement) and the hardware being imperfect. When scientists try to reconstruct the quantum state, this noise often makes the picture look messy and "full of static," hiding the true, clean signal. If they can't clean up the picture, their measurements become less precise, and they lose the super-power advantage that quantum physics promises. So, the big question is: How do we filter out the static without accidentally throwing away the real music?

This is where a new method called SSP-QST (Spectral Subspace Purification for Quantum State Tomography) comes in, proposed by researchers Anuvab Sen and Saibal Mukhopadhyay. They realized that the old ways of cleaning up these quantum pictures were like using a sledgehammer to fix a watch. One common method kept every note, even the fake static ones, making the picture too noisy. Another method tried to force the picture to be perfectly simple (only one note), which worked great if the signal was pure, but if the signal was actually a complex chord (a mix of a few notes), this method would smash the real notes and leave only the loudest one, destroying the information.

The authors propose a smarter, "Goldilocks" approach. Instead of guessing how complex the signal is, SSP-QST listens to the "spectrum" of the noisy picture and asks a simple question: "Is this note loud enough to be real, or is it just background hiss?" They use a mathematical rule (based on something called the Weyl perturbation theorem) to draw a line in the sand. Any "note" (eigenvalue) below that line is tossed out as noise; anything above it is kept and the volume is adjusted so the total sound is balanced again.

What makes this special is that it doesn't need a cheat sheet. It doesn't need to know in advance whether the signal is a single note or a complex chord. It figures it out on the fly. In their computer simulations using Qiskit Aer, this method proved to be the best at reconstructing the true state of the light particles. When the signal was a simple single note, it worked just as well as the old "force it to be simple" methods. But when the signal was a complex chord (ranks 2 through 7), SSP-QST kept all the real notes while still getting rid of the noise, beating every other non-iterative method they tested.

The results from these simulations show that SSP-QST can boost the "fidelity" (how close the reconstruction is to the truth) by up to 0.584 compared to some older methods. Perhaps even more exciting for practical use, it is incredibly efficient with "shots" (the number of times you have to measure). The simulations suggest it can achieve the same quality of result with at least 8 times fewer measurements than the standard method. This means sensors could work faster or with weaker light sources.

The researchers also showed that this method is robust. Even if the noise level changes or the signal is a bit messy, the method still correctly identifies how many real notes are in the chord. It's designed to be fast, requiring only one mathematical "unscrambling" step, which makes it perfect for being plugged into real-time feedback loops where a quantum sensor needs to adjust itself instantly. While these results are currently based on computer simulations and not yet tested on physical hardware, the paper suggests that SSP-QST offers a lightweight, reliable way to clean up quantum data, making quantum sensing more accurate and efficient without needing complex, slow calculations.

Technical Summary: Spectral Subspace Purification for Photonic Quantum State Tomography (SSP-QST)

Problem Statement
Photonic quantum sensing relies on low-rank entangled probes (e.g., GHZ, Bell, NOON states) to achieve precision scaling at the Heisenberg limit ( $1/n$ ). However, practical Quantum State Tomography (QST) faces a fundamental reconstruction dilemma. Standard Least-Squares QST (LS-QST) produces density matrix estimates that are full-rank due to finite-shot measurement fluctuations and physical noise. This "eigenvalue contamination" inflates the estimated entropy and reduces the Quantum Fisher Information (QFI), degrading downstream sensing precision.

Conversely, simple rank-1 purification (forcing the estimate to be a pure state) is structurally biased for real-world probes. Practical photonic sources often exhibit effective ranks $r > 1$ due to beam-splitter imbalances, multiphoton emissions, waveguide crosstalk, and source drift. A fixed rank-1 reconstruction discards these valid signal modes, introducing a systematic error independent of shot count. Existing solutions like Compressed Sensing or Iterative Maximum Likelihood Estimation (MLE) are either too computationally expensive for real-time feedback or require prior knowledge of the rank, which is often unavailable.

Methodology: SSP-QST
The authors propose Spectral Subspace Purification for Quantum State Tomography (SSP-QST), a non-iterative, rank-adaptive post-processing layer applied to standard LS-QST outputs. The method operates in three steps:

Eigendecomposition: The LS-QST estimate ( $\rho_{LS}$ ) is decomposed into its eigenvalues and eigenvectors.
Weyl-Motivated Noise Floor Estimation: A threshold ( $\epsilon_{th}$ ) is computed to distinguish signal eigenmodes from noise. This threshold is derived from the Weyl perturbation theorem, which bounds the deviation of eigenvalues under noise. The formula is:
$\epsilon_{th} = \frac{\hat{p}}{d-1} + \frac{0.5}{\sqrt{N_s}}$
where $\hat{p} = \max(0, 1 - \lambda_d)$ is the estimated noise level derived from the dominant eigenvalue $\lambda_d$ , $d$ is the Hilbert space dimension, and $N_s$ is the number of shots. The first term accounts for depolarizing noise scaling, while the second term provides a finite-shot margin.
Subspace Selection and Renormalization: Eigenvalues below $\epsilon_{th}$ are discarded. The remaining eigenmodes are renormalized to ensure the trace is 1, producing a purified state $\rho_{SSP}$ .

The algorithm includes a "rank-1 override" logic: if the second-largest eigenvalue is below twice the threshold, the method defaults to a rank-1 reconstruction, ensuring safety against over-estimating the rank. The computational cost is dominated by a single eigendecomposition ( $O(d^3)$ ), making it suitable for real-time feedback loops.

Key Contributions

Data-Driven Rank Selection: SSP-QST introduces a closed-form post-processor that selects the reconstruction rank directly from the eigenspectrum without requiring a rank prior or iterative optimization.
Unified Handling of Pure and Mixed Probes: The method avoids the two failure modes of non-adaptive reconstruction: it does not retain noise-inflated full-rank structures (like LS-QST) nor does it discard valid signal modes when the probe rank $r > 1$ (like rank-1 purification).
Theoretical Fidelity Analysis: The paper provides a perturbative analysis showing that removing non-signal eigenmodes improves fidelity, with the gain scaling as $p(d-r)/d$ . This advantage is most pronounced in the low-rank regime ( $r \ll d$ ) typical of photonic probes.
Low-Latency Implementation: By requiring only one eigendecomposition and renormalization, SSP-QST serves as a lightweight primitive for feedback-oriented sensing pipelines, compatible with FPGA-based implementations.

Experimental Results
The authors evaluated SSP-QST using Qiskit Aer simulations with $n=4$ qubits ( $d=16$ ) under depolarizing, amplitude damping, and phase damping noise models.

Fidelity: Across probe ranks $r=1$ $r = 1$ to $r=7$ $r = 7$ , SSP-QST achieved the highest reconstruction fidelity among all tested non-iterative baselines (LS-QST, spectral squaring, and top-eigenvector extraction).
- For rank-1 probes, it matched the performance of top-eigenvector extraction ( $F=1.000$ ).
- For mixed states ( $r \ge 2$ ), it significantly outperformed rank-1 extraction, with fidelity gains ranging from $+0.204$ to $+0.584$ over top-eigenvector extraction.
- It also outperformed spectral squaring by $+0.008$ to $+0.051$ .
Shot Efficiency: SSP-QST demonstrated an $8\times$ improvement in shot efficiency. For a rank-3 probe, SSP-QST achieved a fidelity of $0.956$ with only $512$ shots, whereas LS-QST required $4096$ shots to reach only $0.890$.
Robustness: The method maintained positive fidelity gains across a wide range of depolarizing noise rates ( $p \in [0, 0.22]$ ) and demonstrated robustness to variations in the threshold constant.
Rank Identification: In simulations with noise levels $p \le 0.10$ , SSP-QST correctly identified the true probe rank in all 20 trials for ranks 2 through 6. At higher noise, it exhibited "graceful degradation" by slightly underestimating the rank, a safe failure mode that avoids adding unsupported signal modes.
Comparison to Iterative MLE: Even when given a computational budget equivalent to 400 iterations of Rho-Rho (standard MLE), SSP-QST outperformed the iterative method in fidelity, as the iterative method retained a full-rank noise pedestal that SSP-QST successfully purged.

Significance and Claims
The paper claims that SSP-QST provides a critical bridge between the theoretical promise of quantum sensing and practical implementation. By recovering the low-rank signal subspace without prior knowledge, it prevents the systematic underestimation of sensing precision caused by rank inflation in standard tomography.

The authors emphasize that SSP-QST is specifically designed for feedback-oriented quantum sensing pipelines. Its closed-form nature ( $O(d^3)$ ) allows it to function as a real-time reconstruction primitive, enabling parameter-shift feedback controllers to correct coherent source drift (e.g., wave-plate creep) using a high-fidelity estimate. The paper concludes that while current results are based on software simulations under controlled noise models, the method offers a lightweight, reliable reconstruction tool that can be deployed on reconfigurable hardware (FPGAs) to enhance the reliability of photonic quantum sensing under finite-shot constraints. Future work is identified as validating the method on integrated photonic platforms and extending the spectral purification principle to quantum process tomography.

Technical Summary: Spectral Subspace Purification for Photonic Quantum State Tomography (SSP-QST)

More like this