⚛️ quantum physics

Orthogonalised Self-Guided Quantum Tomography: Insights from Single-Pixel Imaging

This paper introduces orthogonalised self-guided quantum tomography by establishing its mathematical equivalence to single-pixel imaging and demonstrating that this approach significantly improves convergence speed and fidelity without requiring additional experimental overhead.

Original authors: Kiki Dekkers, Alice Ruget, Fazilah Nothlawala, Sabrina Henry, Stirling Scholes, Miles Padgett, Andrew Forbes, Isaac Nape, Jonathan Leach

Published 2026-04-10

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Kiki Dekkers, Alice Ruget, Fazilah Nothlawala, Sabrina Henry, Stirling Scholes, Miles Padgett, Andrew Forbes, Isaac Nape, Jonathan Leach

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 Idea: Finding a Needle in a Haystack (Without Looking at the Whole Haystack)

Imagine you are trying to find a specific, hidden object in a giant, dark room. You can't see the object directly. Instead, you have a flashlight that can only shine in one tiny spot at a time, and a sensor that tells you how much light bounces back.

This is the challenge scientists face in two very different fields:

Single-Pixel Imaging (SPI): Trying to reconstruct a picture of an object using a camera that only has one "pixel" (one sensor).
Quantum Tomography (QT): Trying to figure out the exact state of a tiny quantum particle (like a photon) without destroying it.

Usually, to solve these puzzles, you have to take thousands of measurements, write them all down, and then use a supercomputer to solve a massive math problem to reconstruct the image or state. It's slow and expensive.

The "Self-Guided" Shortcut

The authors of this paper introduce a clever trick called Self-Guided Imaging (SGI) and Self-Guided Quantum Tomography (SGQT).

The Analogy: The Blind Hiker
Imagine you are a blind hiker trying to find the top of a hill (the "true state" or "correct image").

The Old Way: You take 1,000 steps in random directions, write down the elevation at every single spot, go back to base camp, and use a map to calculate where the peak is.
The Self-Guided Way: You take two small steps: one slightly left, one slightly right. You ask your GPS, "Which way is higher?"
- If the right step was higher, you take a big step toward the right.
- If the left step was higher, you go left.
- You keep doing this, zig-zagging your way up the hill until you reach the peak.

You never need to see the whole hill or write down a map. You just use the immediate feedback from your two steps to guide your next move. This is what "Self-Guided" means: the experiment guides itself to the answer.

The Breakthrough: Connecting the Dots

The researchers made a surprising discovery: The math used for the "Blind Hiker" in quantum physics is exactly the same as the math used for the "One-Pixel Camera."

They realized that Single-Pixel Imaging (taking photos with one sensor) is just a special, simpler version of Quantum Tomography. Because they are mathematically twins, they could borrow a trick from one field to fix a problem in the other.

The Problem: Getting Stuck in a Rut

In the "Self-Guided" method, the hiker sometimes takes steps that don't give new information.

The Analogy: Imagine you are walking up a hill, but you keep stepping on the same patch of grass you've already measured. You keep asking, "Is this higher?" and the answer is "No, it's the same." You waste time and energy.
In the quantum world, the random "steps" (measurements) sometimes overlap too much with previous guesses. The algorithm gets confused, moves slowly, and might never reach the perfect peak.

The Solution: The "Orthogonal" Compass

To fix this, the authors borrowed a technique called Orthogonalised Ghost Imaging from the photography world.

The Analogy: The Grid vs. The Mess

Old Way: You are exploring a city by walking in random directions. You might walk North, then North again, then North again. You are wasting time.
New Way (Orthogonalised): You decide to walk in a perfect grid. North, then East, then South, then West. You ensure every step you take gives you brand new information that you haven't seen before.

By adding a "correction term" to their math, the researchers made the quantum hiker walk in a perfect grid. They subtracted out the "old information" from their steps so that every new measurement counted.

The Results: Faster and Sharper

They tested this new method, which they call Orthogonalised Self-Guided Quantum Tomography (OSGQT), in two ways:

Computer Simulations: They ran the math on a computer.
Real Experiments: They used actual lasers and crystals in a lab to manipulate light particles.

The Outcome:

Speed: The new method reached the solution much faster.
Accuracy: The final picture was much clearer.
- Analogy: If the old method gave you a blurry photo that was 92% clear, the new method gave you a sharp, HD photo that was 95% clear.
- In the computer tests, it went from 95% clear to 99% clear.

Why This Matters

This paper is like finding a universal remote control. It shows that the tools we use to take pictures with one pixel are the same tools we use to understand the quantum world. By mixing these tools together, we can:

Save Time: We don't need to take as many measurements.
Save Money: We don't need as much computing power.
See Better: We get more accurate results in both imaging and quantum physics.

In short, the authors taught quantum physics how to walk in a straight line instead of wandering in circles, making the journey to discovery much faster and more precise.

1. Problem Statement

The efficient reconstruction of unknown systems, such as quantum states or images, is a central challenge in modern physics.

Quantum Tomography (QT): Traditional QT requires collecting a large dataset (scaling as $O(d^{2n})$ for $n$ qudits) and solving a computationally intensive inverse problem.
Self-Guided Quantum Tomography (SGQT): An alternative introduced by Ferrie that uses iterative feedback to guide an estimate toward the true state without explicit reconstruction. However, standard SGQT can suffer from slow convergence and suboptimal final fidelity, particularly when measurement masks are not orthogonal.
Single-Pixel Imaging (SPI): A technique for reconstructing images using a single-pixel detector and structured illumination. While SPI and SGQT operate in different domains (classical intensity vs. quantum states), both rely on iterative optimization.
The Gap: There was no established mathematical link between the algorithms used in SPI and SGQT, nor was there a method to apply recent SPI advancements (like orthogonalization) to improve quantum state estimation.

2. Methodology

The authors developed a unified framework connecting these fields and introduced a new algorithmic improvement.

A. Mathematical Equivalence: Self-Guided Imaging (SGI)

The authors demonstrated that Self-Guided Imaging (SGI) is mathematically equivalent to Single-Pixel Imaging (SPI) when a linear distance measure is used.

SGQT Mechanism: Uses the Simultaneous Perturbation Stochastic Approximation (SPSA) to estimate gradients. It updates a guess state $|\sigma\rangle$ based on the difference in fidelity measurements of perturbed states $|\sigma \pm \beta\Delta\rangle$ .
Linearization: By applying the SGQT algorithm to real-valued images (where the distance measure is linear, i.e., $\langle O|\Delta\rangle$ ) rather than complex quantum states (where the measure is non-linear, i.e., $|\langle \psi|\sigma\rangle|^2$ ), the update rule simplifies exactly to the standard SPI update rule.
Conclusion: SPI is a specific subset of iterative self-guided tomography where the cost function is linear.

B. Orthogonalised Self-Guided Quantum Tomography (OSGQT)

Inspired by Orthogonalised Ghost Imaging (OGI), which uses the Kaczmarz algorithm to correct for non-orthogonal measurement masks in SPI, the authors introduced OSGQT.

The Problem with Standard SGQT: The random perturbation vectors ( $\Delta_k$ ) in SGQT are not mutually orthogonal. This leads to redundant sampling and potential "overshooting" in the cost function landscape, slowing convergence.
The OSGQT Solution: The authors incorporated a computational correction term into the SGQT update rule. The new update equation is:
$|\sigma_{k+1}\rangle = |\sigma_k + \alpha_k \left( \frac{f(\psi, \sigma^+_k) - f(\psi, \sigma^-_k)}{2\beta_k} - \frac{f(\sigma_k, \sigma^+_k) - f(\sigma_k, \sigma^-_k)}{2\beta_k} \right) \Delta_k \rangle$
- Term 1: The standard experimental measurement (gradient estimate).
- Term 2: A computational correction calculated numerically based on the current estimate $|\sigma_k\rangle$ . This term quantifies how much new information the current mask provides relative to the current estimate, suppressing updates when the mask offers little new information.
Overhead: This method requires no additional experimental overhead. The correction term is purely numerical.

3. Key Contributions

Conceptual Unification: Established that SGI (linear distance measure) is mathematically equivalent to SPI, bridging the gap between quantum tomography and classical imaging algorithms.
Algorithmic Innovation: Introduced OSGQT, adapting the Kaczmarz/OGI correction strategy to quantum state estimation.
Experimental Validation: Demonstrated the superiority of OSGQT over standard SGQT in both numerical simulations and physical experiments involving high-dimensional quantum states.

4. Results

The performance of OSGQT was compared against standard SGQT using heralded single-photon states encoded in high-dimensional Orbital Angular Momentum (OAM) modes ( $d=5$ ).

Numerical Simulations (Noise-free):
- Convergence Speed: OSGQT reached sub-0.1 infidelity significantly faster than SGQT.
- Final Fidelity: OSGQT achieved a final fidelity of 99.17% (±0.09%), whereas SGQT saturated at 95.2% (±0.2%).
Experimental Results:
- Convergence Speed: OSGQT converged faster in experimental runs.
- Final Fidelity: OSGQT achieved 95.3% (±0.8%), outperforming SGQT which reached 92.1% (±1.1%).
Robustness: The improvements held true across various scaling parameters ( $\alpha, \beta$ ) and different noise regimes (shot noise and experimental imperfections).
Visualization: Phase profiles of the estimated states showed that OSGQT estimates closely resembled the true state phase much earlier in the iteration process compared to SGQT.

5. Significance

Optimization without Hardware Cost: OSGQT improves measurement efficiency and final accuracy without requiring new hardware or additional experimental measurements, relying solely on a computational post-processing step.
Cross-Disciplinary Insight: The work highlights that advanced routines from classical imaging (like orthogonalization in SPI) can be directly transferred to quantum information science to solve scaling and convergence issues.
Scalability: The method is particularly relevant for high-dimensional quantum systems where traditional tomography is infeasible due to the $O(d^4)$ scaling of process tomography or the data volume required for state tomography.
Future Directions: The paper suggests a new paradigm where algorithms from SPI and SGQT can be interchanged and optimized, potentially leading to faster and more robust quantum sensing and imaging technologies.