Adaptive Reconstruction of Bosonic Quantum States

AI summary

Overview

• Research area: Quantum physics — continuous-variable (bosonic) quantum state characterisation and closed-loop quantum optimal control, implemented experimentally on a circuit quantum electrodynamics (cQED) platform.
• Technical level: Advanced. The paper assumes familiarity with Wigner functions, phase space, Bayesian inference, and superconducting circuit hardware.
• Scope (1 sentence): The paper introduces an adaptive, Bayesian state-reconstruction method that estimates the fidelity of bosonic states with respect to a whole family of target states from a small number of Wigner-function measurements, benchmarks it against existing sampling protocols on Schrödinger cat states, and plugs the reconstructed fidelity into a closed-loop optimal-control loop.

What This Paper Is About

Bosonic quantum states are attractive for quantum information processing, but they live in a very large Hilbert space, so characterising them normally requires reconstructing a full Wigner function — a process that conventionally takes hours of measurement. Existing "optimal sampling" methods get around this cost only by assuming a single fixed target state, which fails when the physically relevant states differ by trivial operations such as displacements or rotations in phase space. The authors build a reconstruction protocol that instead estimates fidelity to a family of bosonic states, produces a reproducible Wigner map within minutes, and is fast and stable enough to serve as the feedback signal in closed-loop quantum optimal control.

Key Contributions

1. An adaptive reconstruction algorithm that combines a physics-informed parametric model of the Wigner function (three Gaussians plus an interference fringe, with the thermal purity 𝒫 kept as a free parameter rather than assumed to be 1) with Maximum A Posteriori (MAP) estimation, a Wishart prior, bootstrap-based uncertainty estimation, and active learning that selects the most informative phase-space points.
2. Experimental implementation on cQED hardware — a high-Q three-dimensional niobium cavity coupled to a transmon qubit in the dispersive regime — benchmarked on qcMAP-generated Schrödinger cat states with amplitudes α ∈ [1, 3].
3. An experimental head-to-head comparison of the adaptive strategy against existing Wigner sampling protocols (uniform grid sampling and the optimal distributions p^sq ∝ W_target² and p^abs ∝ |W_target|) for both a target cat state and a rotated-and-displaced family of cat states.
4. A proof-of-principle closed-loop quantum optimal control experiment in which the reconstructed fidelity is folded into the figure of merit, with the optimiser correcting the phase of the first pulse in the qcMAP sequence and the qubit I and Q quadratures during the entangling operation.

Main Findings

• Large speed-up at comparable structure: For an α = 3 cat state, the algorithm produces a reconstructed Wigner map from about 1000 measured points in roughly 3.5 minutes, whereas the high-resolution reference Wigner map used for comparison required about 4 hours of measurement.
• Reproducibility across runs: Statistics were gathered from 100 independent experimental reconstruction runs per hyperparameter setting. The 3σ spread indicates that 99.7% of runs yield high-fidelity reconstruction, which the authors identify as a prerequisite for stable closed-loop feedback.
• Batch size dominates performance: The optimal configuration found was 100 averages per point, a batch size of 5, and an entropy threshold of 1.5. Increasing the batch size to 15 gave the highest observed infidelity and produced outlier runs with substantially worse reconstruction — visible only in the 3σ spread, not in the 1σ or 2σ spreads.
• Robustness to prior mismatch: Displaced and rotated α = 3 cat states were reconstructed using the same prior as the non-rotated benchmark state and still yielded reliable, high-fidelity reconstructions despite the significant prior–target mismatch.
• Known model limitation: Residuals between measured and reconstructed data show a small, persistent discrepancy in the fringe region, attributed to the three-Gaussian parametrisation's inability to account for fringe bending (curvature).
• A working early-termination metric: The integrated standard deviation σ_int across the 10 bootstrapped models tracks infidelity and ellipticity convergence, and reconstruction is stopped once σ_int drops below a user-set threshold.
• Compared to existing samplers: Both p^sq and adaptive sampling converge to the same target fidelity, while p^abs converges to a slightly different value, which the authors attribute to calibration drift in the Wigner measurement scaling over that strategy's longer measurement time. Reference values quoted for the |C_√2^π⟩ comparison are F = 0.94 for the measured high-resolution map versus the target and F = 0.93 for the fit versus the target.
• Grid sampling is biased at low resolution: Coarse grids produce biased fidelity estimates and, in extreme cases, fidelities above one — unphysical states arising from insufficiently resolved Wigner maps.
• Theoretical uncertainty bound: A Cramér–Rao bound analysis (Appendix F) indicates that the optimal and adaptive sampling strategies achieve approximately 2 times lower fidelity uncertainty than grid sampling. The experimentally observed larger standard deviation for adaptive sampling is partly ascribed to the different uncertainty-estimation methods (100 real experimental runs versus bootstrap on data).
• State-size dependence: Convergence curves for α ∈ {1, 1.5, 2, 2.5} show consistent behaviour, but smaller cat states depend more strongly on sampling stochasticity and need higher entropy thresholds to reach comparable fidelity.
• Closed-loop control: The abstract reports that the reconstructed fidelity was successfully used as the figure of merit in a proof-of-principle closed-loop optimal control experiment; the quantitative fidelity improvement from that experiment is not reported in the provided (truncated) content.

Methodology in Plain English

The team starts from the known analytical Wigner function of an ideal qcMAP cat state and rewrites it as a flexible model: two Gaussian blobs (the two "lobes" of the cat) plus a central Gaussian multiplied by a cosine interference term, with adjustable positions, widths, amplitudes, a rotation angle for the fringes, and a fringe phase. This model is fitted to measured Wigner values using MAP estimation, where a prior keeps the Gaussian positions and variances in physically realistic ranges and prevents overfitting when data are scarce.

To decide where to measure next, they reuse the same data many times: they draw 10 resampled datasets from the measurements already collected, fit 10 independent models, and treat the disagreement between those 10 Wigner maps as the model's uncertainty at each phase-space point. A Boltzmann distribution built from that uncertainty then concentrates the next batch of measurements (batch size 5 in the optimal setting) on the most uncertain regions, with an entropy constraint that stops the sampler from drifting into near-zero, uninformative regions or collapsing too sharply. The loop repeats until the integrated uncertainty falls below a threshold. Fidelity is computed as a normalised overlap integral between Wigner maps on a phase-space grid (the F₂ fidelity of Eq. 11).

Experimentally, cat states are prepared with the qcMAP protocol from displaced coherent states, and Wigner values are obtained with the cavity–qubit dispersive interaction. For the control experiment, the fixed waiting time in the protocol is replaced by a qubit control pulse, and the pulse phases and qubit quadratures are tuned based on the fidelity fed back from the reconstruction.

Why This Matters

The bottleneck the paper attacks is not state preparation but state verification: full Wigner tomography is slow, and single-target fidelity estimators are brittle once a state is displaced or rotated. By making family-relative fidelity estimation fast and repeatable, the method turns tomography from an offline diagnostic into something that can sit inside a feedback loop.

• Automated calibration and tuning of quantum hardware: a minutes-scale, reproducible fidelity estimate can drive autonomous optimisation instead of manual sweeps.
• Verification of bosonic-code state preparation: cat states, and by extension GKP states and rotation-symmetric codes, need routine, low-overhead state verification.
• Benchmarking quantum memories and error-correction primitives that are encoded in a single harmonic oscillator rather than many qubits.
• Reducing measurement time in labs running repeated tomography, freeing machine time for experiments rather than characterisation.

Industry relevance: The author list includes an industrial co-author (Qruise GmbH), and the closed-loop framing points directly at commercial quantum-control software and superconducting-quantum-computing hardware teams that need automated, measurement-efficient calibration of cavity-based qubits.

Future Directions

• Generalising the parametrisation to other bosonic families — GKP states, rotation-symmetric codes, and general superpositions of arbitrary squeezed states — by substituting a different model in place of the three-Gaussian form.
• Non-Wigner representations: the authors note the model need not be a Wigner function and could instead be a characteristic function, which matters for platforms without number-resolved detection.
• Scaling the closed-loop protocol beyond the proof of principle to more control parameters and broader classes of bosonic states, as sketched in the paper's outlook section.
• Improving the model and the uncertainty estimate: accounting for fringe curvature, which currently limits residual agreement, and replacing bootstrap with fuller posterior sampling (e.g., MCMC) where computational budget allows; the paper also flags generalised Husimi Q functions as a less practical alternative in this framework.

Target Audience

Experimental superconducting-qubit and cQED groups; continuous-variable quantum information researchers working with cat, GKP, or rotation-symmetric codes; quantum-control and machine-learning-for-quantum-hardware researchers interested in measurement-driven feedback; and graduate students or advanced readers who want a concrete, experimentally demonstrated example of Bayesian experimental design applied to quantum state tomography.

Authors’ abstract

Bosonic quantum systems provide a hardware-efficient platform for quantum information processing but remain challenging to characterise due to their large Hilbert space and the high measurement cost of state tomography. Existing approaches estimate the fidelity with respect to a single target state, making them unsuitable for applications in which physically equivalent states differ by phase space translations, rotations, or other transformations. Here, we introduce an adaptive reconstruction technique that estimates the fidelity with respect to a family of bosonic states while reconstructing the underlying Wigner function from a small number of measurements. The method combines a physics-informed parametric model with Bayesian inference, bootstrap, and active learning to iteratively select the most informative phase space sampling points. We implement the approach on a circuit quantum electrodynamics platform and benchmark it on Schrödinger cat states with amplitudes $α\in[1,3]$. The reconstruction yields reproducible fidelity estimates within a few minutes, remains robust to substantial displacements and rotations in phase space despite using a mismatched prior, and is sensitive to subtle state imperfections. We further compare the adaptive strategy with existing Wigner function sampling protocols experimentally, demonstrating the advantage of adaptive sampling for measurement-efficient fidelity estimation with respect to a family of cat states. Finally, we incorporate the reconstructed fidelity into the figure of merit used in a proof-of-principle closed-loop quantum optimal control experiment, demonstrating the applicability of the method to autonomous optimisation of bosonic quantum states.