⚛️ quantum physics

Simultaneous reconstruction of quantum process and noise via corrupted sensing

This paper proposes a framework for quantum process tomography that simultaneously reconstructs quantum processes and sparse corrupted noise using generalized restricted isometry properties in both Choi-state and process-matrix representations, significantly reducing the required experimental configurations.

Original authors: Mengru Ma, Jiangwei Shang

Published 2026-02-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Mengru Ma, Jiangwei Shang

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 take a perfect photograph of a very fast-moving, invisible object (a quantum process, like a logic gate in a quantum computer). In the real world, your camera isn't perfect. Sometimes the lens is smudged, the shutter sticks, or a sudden flash of light ruins the shot. In the scientific world, these are called noise and corruptions.

Usually, when scientists try to figure out what the object looks like, they have to assume the camera is working perfectly. If the photos are blurry or have weird spots, they just throw them away or try to guess what the "real" image was. This is slow, expensive, and often impossible for complex objects.

This paper proposes a smarter way to take these photos. It's like having a camera that can simultaneously figure out what the object looks like and what was wrong with the camera lens at the exact same time.

Here is a breakdown of how they did it, using simple analogies:

1. The Problem: The "Smudged Lens"

In quantum computing, we need to check if our "gates" (the switches that do the math) are working correctly. To do this, we send signals through them and measure the results.

The Old Way: If the measurement data is messy (due to a broken detector or a calibration error), the whole experiment fails. You have to do the experiment thousands of times to get a clear picture, which takes forever.
The New Idea: The authors realized that most of the time, the "messiness" isn't random everywhere; it's sparse. Think of it like a photo where only a few pixels are stuck or bright white, while the rest of the image is fine. Because the errors are rare (sparse), we can mathematically separate the "bad pixels" from the "real image."

2. The Two Methods: Two Different Cameras

The paper tests two different ways to take this "smart photo."

Method A: The "Choi-State" Camera (The Theoretical Proof)

Imagine you want to understand a machine by feeding it a special, perfectly entangled "test ball" and seeing what comes out.

How it works: They proved mathematically that if you take enough random snapshots of this test ball, you can use a special algorithm to reconstruct the machine's behavior and identify which snapshots were corrupted by noise.
The Catch: It works, but it's a bit inefficient. It's like taking 1,000 photos to get a clear picture, even though you only really needed 200. It proves the math works, but it's not the most practical tool yet.

Method B: The "Process-Matrix" Camera (The Practical Winner)

This is the main star of the show. Instead of using one special test ball, this method uses a variety of different "input states" (like sending different colored balls into the machine) and measures the output in different ways.

The Magic: By mixing and matching these inputs and outputs, they found a way to be much more efficient.
The Results:
- For a 2-qubit gate (a simple switch), they could get a clear, high-quality picture using only 64 settings instead of the usual 256.
- For a 3-qubit gate (a more complex switch), they needed only about 10% of the usual data to get a very accurate result.
- For a 4-qubit gate (a very complex switch), they needed only 3% of the usual data to get a near-perfect picture.

3. The "Super-Resolution" Analogy

Think of the quantum process as a complex puzzle.

Standard Tomography: You try to solve the puzzle by looking at every single piece, one by one. If a few pieces are painted over with black ink (noise), you can't solve it.
This Paper's Method: You look at the puzzle from many different angles. Even if some pieces are painted over, the algorithm looks at the pattern of the whole puzzle. It says, "Okay, these three pieces look weird, so they must be the 'painted over' ones. Let's ignore them and solve the rest."
The Benefit: You don't need to look at every single piece. You can solve the whole puzzle by looking at just a fraction of the pieces, and you can even tell which pieces were the "bad" ones.

4. Why This Matters (According to the Paper)

The authors show that this method allows scientists to:

Save Time and Resources: You need far fewer experimental setups (measurements) to get a reliable result.
Be Robust: You don't have to worry as much about occasional equipment glitches or "bad days" in the lab. The math can filter out the errors automatically.
Scale Up: As quantum computers get bigger (more qubits), the number of measurements usually explodes. This method keeps the number of measurements manageable, making it possible to test larger, more complex quantum systems.

Summary

In short, the authors built a mathematical framework that acts like a self-correcting camera. It doesn't just take a picture of a quantum process; it figures out what the process is while identifying and removing the "noise" that corrupted the data. This means we can characterize quantum computers faster, cheaper, and more accurately, even when our equipment isn't perfect.

Technical Summary: Simultaneous Reconstruction of Quantum Process and Noise via Corrupted Sensing

Problem Statement
Quantum process tomography (QPT) is essential for characterizing quantum gates and channels, yet standard QPT suffers from exponential scaling, requiring $O(d^4)$ measurement configurations for a $d$ -dimensional system. While compressed sensing (CS) techniques have been developed to exploit structural priors (such as sparsity or low rank) to reduce this resource cost, existing methods often assume ideal measurement conditions. In realistic experimental settings, measurement outcomes are frequently corrupted by sparse, non-Gaussian errors arising from detector imperfections, calibration drifts, or occasional measurement failures. Conventional noise models often fail to capture these sparse corruptions, leading to inaccurate process reconstruction. This work addresses the challenge of simultaneously reconstructing a quantum process and the underlying sparse measurement noise without prior knowledge of the noise locations.

Methodology
The authors propose a corrupted sensing framework for QPT that integrates the principles of compressed sensing with the recovery of sparse corruptions. The study is conducted through two distinct representations:

Choi-State Representation:
- The quantum process $E$ is mapped to a Choi state $\rho_E$ via the Choi-Jamiołkowski isomorphism.
- The measurement model is formulated as $y = \Lambda(\rho_E) + v + z$ , where $y$ represents noisy data, $\Lambda$ is a linear map based on random Pauli observables, $v$ is a sparse corruption vector, and $z$ represents unstructured background noise.
- The authors derive a Generalized Restricted Isometry Property (GRIP) for the extended measurement matrix $M = [\Lambda, I]$ . They prove that if the number of measurements $m$ scales as $O(rd^2 \log^6 d)$ (where $r$ is the rank of the Choi state) and the corruption magnitude is bounded, the matrix $M$ satisfies GRIP with high probability. This condition ensures the stable recovery of both the low-rank Choi state and the sparse noise vector.
- Reconstruction is achieved by solving a convex optimization problem minimizing the $\ell_1$ -norm of the noise and the trace norm (nuclear norm) of the Choi state, subject to positive semidefinite and trace-preserving constraints.
Process-Matrix Representation:
- The process is represented directly by its process matrix $\chi$ in a known basis (e.g., Pauli basis), where $\chi$ is sparse.
- The measurement model is $y = \Phi\chi + v + z$ , where $\Phi$ encodes the input states and measurement observables.
- A similar convex optimization problem is solved to recover $\chi$ and $v$ simultaneously, enforcing the positive semidefinite and trace-preserving constraints on $\chi$ .
- This approach utilizes specific input-output configurations rather than purely random Pauli sampling, aiming for higher efficiency.

Key Results
Numerical simulations were performed on various quantum gates to validate the framework:

Two-Qubit Gates (CNOT, CZ, SWAP):
- In the Choi-state scheme, high fidelity ( $F \approx 0.95$ ) was achieved with significant noise ( $\eta=0.1$ ), though it required a relatively large number of configurations (e.g., $m=192$ for CNOT).
- The process-matrix scheme demonstrated superior efficiency. For the CNOT gate, a fidelity of $F \approx 0.962$ was achieved with only $m=64$ configurations (25% of the standard 256 settings). Similar improvements were observed for CZ ( $F \approx 0.971$ at $m=64$ ) and SWAP ( $F \approx 0.978$ at $m=48$ ).
Three-Qubit Gates (Toffoli, Fredkin):
- The process-matrix scheme showed enhanced scalability. For the Toffoli gate, a fidelity of $F \approx 0.9615$ was reached with $m=384$ configurations (9.375% sampling rate), and $F \approx 0.99$ at 40.625%.
- The Fredkin gate achieved $F \approx 0.9553$ at a 6.25% sampling rate ( $m=256$ ).
Four-Qubit Gate (CCCZ):
- The method scaled to four qubits with the CCCZ gate. A fidelity of $F \approx 0.971$ was achieved with a sampling rate of only 3.125% ( $m=2048$ ), and $F \approx 0.9945$ at 25% sampling.
Noise Recovery: In all cases, the framework successfully reconstructed the sparse Gaussian noise vectors with low mean squared error (MSE), confirming the simultaneous recovery capability.

Significance and Claims
The paper claims that the proposed corrupted sensing framework enables the simultaneous reconstruction of quantum processes and sparse measurement noise, a capability lacking in standard QPT and many existing CS-QPT protocols.

Efficiency: The results demonstrate that significant reductions in experimental configurations are achievable even under corrupted noise. The process-matrix representation is highlighted as being more efficient than the Choi-state scheme for a broader range of gates, utilizing input-output configurations more effectively.
Scalability: The method exhibits strong scalability, with the required sampling rate decreasing as the system size increases (e.g., the four-qubit CCCZ gate required the lowest sampling rate to reach high fidelity).
Robustness: By explicitly modeling and recovering sparse corruptions, the framework offers a robust alternative for realistic experimental scenarios where conventional noise models fail.

The authors conclude that while the current work provides a probabilistic condition (GRIP) and numerical validation, future work should focus on deriving explicit error bounds and exploring alternative analytical tools to provide stronger theoretical guarantees. They also suggest extending the corrupted sensing setting to measurement tomography.