⚛️ quantum physics

Adaptive Reconstruction of Bosonic Quantum States

This paper introduces and experimentally validates an adaptive reconstruction technique that combines physics-informed modeling with Bayesian inference and active learning to efficiently estimate the fidelity of bosonic quantum states (specifically Schrödinger cat states) across a family of phase-space transformations, enabling rapid, robust characterization and autonomous optimization on a circuit quantum electrodynamics platform.

Original authors: Vasilisa Usova, Phila Rembold, Ian Yang, Marco Rossignolo, Simone Montangero, Samuele Tosatto, Gerhard Kirchmair

Published 2026-08-04

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Vasilisa Usova, Phila Rembold, Ian Yang, Marco Rossignolo, Simone Montangero, Samuele Tosatto, Gerhard Kirchmair

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 ghost. This isn't just any ghost; it's a "quantum" ghost made of light and electricity, living inside a super-cooled metal box. In the world of quantum computing, these ghosts are called bosonic states. Unlike the tiny, simple switches (qubits) used in most computers today, these quantum ghosts are like complex, swirling clouds of energy that can hold a massive amount of information. Scientists love them because they are efficient, but they are incredibly hard to catch on camera.

The problem is that these clouds are huge and fragile. To see what they look like, scientists usually have to take thousands of photos from every possible angle, a process called "tomography." It's like trying to map a whole city by walking every single street; it takes forever, and by the time you finish, the city has changed. Furthermore, these ghosts are tricky: if you nudge them slightly or spin them around, they look different on the photo, even though they are essentially the same "ghost" doing the same thing. Traditional cameras get confused by these shifts, making it hard to tell if the ghost is perfect or broken.

This is where the story gets interesting. A team of researchers has invented a new, super-smart camera that doesn't just take random photos. Instead, it acts like a curious detective. It takes a few quick snapshots, figures out where it is most confused, and then zooms in specifically on those blurry spots to get a clearer picture. This "adaptive" approach lets them reconstruct the ghost's shape in just a few minutes instead of hours, even if the ghost has moved or spun around. They tested this on "Schrödinger's cat" states—quantum ghosts that are simultaneously alive and dead—and used this fast camera to automatically fix the camera settings to make the ghost even better. It's a leap forward for making quantum computers that can fix themselves.

The Detective Camera: How the Paper Works

The researchers, led by Vasilisa Usova and her team, tackled a major headache in quantum physics: how to quickly and accurately describe a quantum state without spending all day measuring it. They focused on bosonic systems, which are like a special type of quantum memory that can hold more data than standard bits but is notoriously difficult to measure.

The Old Way vs. The New Way
Traditionally, to figure out what a quantum state looks like, scientists use a method called Wigner function reconstruction. Imagine trying to draw a portrait of a person who is constantly moving and changing shape. The old method is like taking a grid of photos: you take a picture at every single point on a grid, whether the person is there or not. If the person moves (a "displacement" or "rotation" in the phase space), your grid might miss them entirely, or you might waste time taking pictures of empty space. This is slow and expensive.

The team's new method is adaptive reconstruction. Instead of a rigid grid, their algorithm is like a detective with a "gut feeling."

The Guess: It starts with a physics-based guess of what the state should look like (a "parametric model").
The Uncertainty: It calculates where it is most unsure. Is the left side of the cloud blurry? Is the right side wobbling?
The Zoom: It uses a technique called bootstrap (which is like asking 10 different detectives to draw the picture based on the same clues and seeing where they disagree) to find the areas of highest uncertainty.
The Action: It then chooses to take the next measurement exactly where the uncertainty is highest. It ignores the empty, boring spots and focuses only on the interesting, confusing parts.

The Results: Fast, Robust, and Self-Correcting
The team tested this on a circuit quantum electrodynamics (cQED) platform, which uses a superconducting cavity and a qubit. They created "Schrödinger cat states" with amplitudes ( $\alpha$ ) ranging from 1 to 3.

Speed: While a standard high-resolution map takes about 4 hours to build, their adaptive method produced a nearly identical map in just 3.5 minutes.
Robustness: Even when they intentionally shifted or rotated the quantum state (making it look very different from their initial guess), the algorithm still found the correct shape. It didn't get confused by the movement; it just adapted its sampling to find the new location.
Sensitivity: The method was so precise that it could spot tiny imperfections, like slight "ellipticity" (squashing) in the shape of the state, which standard methods might miss.

The "Self-Driving" Quantum Computer
The most exciting part of the paper is how they used this new camera for closed-loop quantum optimal control (QOC).
Imagine you are trying to bake the perfect cake, but your oven temperature keeps drifting.

Open-loop (Old way): You set the oven to 350°F and hope for the best. If the oven is broken, the cake burns.
Closed-loop (New way): You taste the cake after every minute. If it's too dry, you adjust the heat immediately.

The researchers used their fast reconstruction method to "taste" the quantum state. They fed the results into an optimization algorithm (using a method called dCRAB). The algorithm would tweak the control pulses (the "recipe") to fix errors.

They found that by including a "penalty" for specific distortions (like the "Kerr effect," which bends the state's shape), the system could automatically correct itself.
In their tests, the system successfully improved the "fidelity" (how perfect the state is) and fixed the shape of the cat states, even starting from a zero-amplitude guess.

What They Didn't Do (and Why It Matters)
The paper is careful to note what this method isn't. It doesn't claim to solve every quantum problem instantly.

It is not a magic wand that fixes everything in one go. The improvements were "modest" because the optimization algorithms they used (like Nelder-Mead) are good at simple tasks but struggle with very complex, high-dimensional landscapes.
They explicitly ruled out the idea that a simple grid of measurements is sufficient for closed-loop control. They showed that grid sampling fails when the state moves, leading to biased or unphysical results.
They also noted that while their method works great for cat states, it relies on having a good mathematical model of the state. If you don't know what kind of state you are looking for, the "detective" needs a new model to work with.

The Bottom Line
This paper demonstrates that by combining a physics-based model with smart, adaptive sampling, we can measure complex quantum states in minutes rather than hours. This speed and accuracy make it possible to build "self-driving" quantum experiments that can fix their own errors in real-time. While the current results are a proof-of-concept, they lay the groundwork for a future where quantum computers can autonomously optimize their own performance, paving the way for more reliable and powerful quantum technologies.

Technical Summary: Adaptive Reconstruction of Bosonic Quantum States

Problem Statement

Bosonic quantum systems offer a hardware-efficient paradigm for quantum information processing due to their large Hilbert spaces, enabling resource-efficient quantum error correction. However, characterizing these states is inherently challenging. Traditional state tomography requires full reconstruction of the Wigner function, which is resource-intensive and demands a large number of measurements. Furthermore, existing fidelity estimation methods typically target a single fixed state. This is insufficient for practical applications where physically equivalent states differ by phase space translations, rotations, or other transformations. Additionally, realistic experimental setups suffer from unaccounted Hamiltonian contributions, calibration imperfections, and system drifts, which limit the effectiveness of open-loop (model-optimized) control strategies. There is a critical need for reconstruction protocols that can identify the optimal fidelity within a relevant family of states, remain robust to phase space displacements, and operate within the tight time constraints (minutes) required for closed-loop quantum optimal control (QOC).

Methodology

The authors introduce an adaptive reconstruction technique that combines a physics-informed parametric model with Bayesian inference, bootstrap resampling, and active learning. The core workflow involves:

Physics-Informed Parametric Model: Instead of non-parametric reconstruction, the Wigner function is modeled using a parametric form tailored for Schrödinger cat states. The model consists of three Gaussian components (left, center, and right) with amplitudes, means, and covariance matrices, plus a cosine term to account for fringe interference. This reduces the dimensionality of the problem and ensures physically plausible reconstructions.
Bayesian Inference with MAP and Bootstrap: The algorithm employs Maximum A Posteriori (MAP) estimation with a Wishart prior to constrain parameters to realistic regimes. To quantify epistemic uncertainty without the computational cost of full Markov Chain Monte Carlo (MCMC), the method uses a bootstrap approach. Multiple models are trained on datasets resampled with replacement from the original measurements.
Active Sampling via Boltzmann Distribution: The variance (disagreement) among the bootstrapped models at each phase space point serves as an uncertainty metric. New measurement points are selected adaptively from regions of highest uncertainty. Specifically, a Boltzmann distribution is constructed based on the estimated standard deviation, where the "temperature" (inverse temperature $\beta_B$ ) is adjusted to maintain a specific entropy range. This prevents the sampling from becoming too uniform (uninformative) or too sharp (risking rejection sampling failure).
Closed-Loop Integration: The reconstruction is integrated into a closed-loop QOC framework. Control pulses are iteratively refined based on a Figure of Merit (FoM) derived from the reconstructed state fidelity. The FoM includes a penalty term to account for state imperfections, such as Gaussian ellipticity caused by Kerr nonlinearities.

Key Contributions

Adaptive Reconstruction Framework: The paper presents a method that estimates fidelity with respect to a family of bosonic states (invariant under phase space translations and rotations) rather than a single fixed target.
Efficiency and Speed: The technique reconstructs high-fidelity Wigner maps within a few minutes using a small number of measurements, significantly faster than conventional high-resolution grid sampling.
Robustness to Mismatch: The method demonstrates robustness to substantial displacements and rotations in phase space, even when the prior used for reconstruction is mismatched with the actual state.
Sensitivity to Imperfections: Beyond fidelity, the reconstruction is sensitive to subtle state imperfections, such as the ellipticity of the Gaussian components, which are often missed by fidelity metrics alone.
Experimental Demonstration: The approach is implemented on a circuit quantum electrodynamics (cQED) platform using a superconducting cavity and a transmon qubit. It is benchmarked on Schrödinger cat states with amplitudes $\alpha \in [1, 3]$ .

Results

Reconstruction Quality: The algorithm produces reproducible Wigner maps that capture distinct features of high-resolution maps (which take hours to acquire) in approximately 3.5 minutes. The infidelity (1 - $F_2$ ) is minimized with specific hyperparameters: 100 averages per point, a batch size of 5, and an entropy threshold of 1.5.
Comparison with Sampling Strategies:
- Grid Sampling: While unbiased, grid sampling is inefficient for closed-loop applications due to long acquisition times and the risk of missing displaced states if the grid is not pre-defined correctly.
- Optimal Sampling ( $p_{sq}$ and $p_{abs}$ ): These strategies perform well for states close to the target but fail to accurately reconstruct states that are significantly displaced or rotated in phase space, as they lack sampling points in the relevant regions.
- Adaptive Sampling: The proposed method outperforms existing strategies in closed-loop scenarios, successfully predicting fidelity for rotated and displaced states where optimal sampling fails.
Closed-Loop Optimization: In proof-of-principle experiments, the adaptive reconstruction was used as the feedback mechanism for a dCRAB (dressed Chopped Random Basis) optimizer. The system successfully reduced Kerr-induced distortions (improving Gaussian ellipticity) and improved state preparation fidelity for various cat sizes and phases. The inclusion of a penalty term in the FoM was shown to be crucial for correcting specific distortions, though it introduced a trade-off with raw fidelity in some runs.

Significance and Claims

The paper claims that this work provides the first reconstruction technique that simultaneously satisfies the three critical requirements for closed-loop bosonic state optimization: minimal estimation error, high reproducibility across trials, and a total reconstruction time constrained to a few minutes.

The significance lies in enabling autonomous optimization of bosonic quantum states. By providing a fast, robust, and adaptive fidelity estimator that accounts for phase space transformations, the method allows for direct feedback-based control that compensates for experimental drifts and nonlinearities without requiring precise prior knowledge of the system Hamiltonian. The authors position this as a foundational step toward more general closed-loop protocols, suggesting that future integration with physics-informed reinforcement learning could further enhance the efficiency and capability of adaptive control for bosonic quantum systems. The authors remain modest regarding the current optimization gains, attributing limited improvements to the constraints of direct-search algorithms in high-dimensional landscapes, but emphasize the method's utility as a tool for a wide range of bosonic state experiments.