Precision-guaranteed quantum tomography

Abstract: Quantum state tomography is the standard tool in current experiments for verifying that a state prepared in the lab is close to an ideal target state, but up to now there were no rigorous methods for evaluating the precision of the state preparation in tomographic experiments. We propose a new estimator for quantum state tomography, and prove that the (always physical) estimates will be close to the true prepared state with high probability. We derive an explicit formula for evaluating how high the probability is for an arbitrary finite-dimensional system and explicitly give the one- and two-qubit cases as examples. This formula applies for any informationally complete sets of measurements, arbitrary finite number of data sets, and general loss functions including the infidelity, the Hilbert-Schmidt, and the trace distances. Using the formula, we can evaluate not only the difference between the estimated and prepared states, but also the difference between the prepared and target states. This is the first result directly applicable to the problem of evaluating the precision of estimation and preparation in quantum tomographic experiments.

• The paper introduces the ENM estimator, achieving rigorous non-asymptotic precision guarantees for quantum state estimation.
• It projects linear least-squares solutions onto physical density matrices, ensuring valid and meaningful error bounds.
• The method is applicable to any finite-dimensional system and measurement protocol, guiding sample size choices and confidence trade-offs.

Precision-Guaranteed Quantum Tomography: An Authoritative Summary

Motivation and Background

@@@@1@@@@ (@@@@2@@@@) serves as the canonical methodology for characterizing quantum states in experimental scenarios. However, prior to this work, there lacked a rigorously justified framework for assigning objective, quantitative bounds (in terms of probability or "confidence level") to the precision of state estimators obtained via finite datasets. Typical approaches—including point estimators such as maximum likelihood (ML) and linear estimators, as well as the construction of confidence regions—fail to couple physically meaningful confidence guarantees with generality in terms of system dimension, measurement protocols, and loss functions.

This paper addresses the gap by introducing a new "extended norm-minimization" (ENM) estimator for quantum state tomography that, for arbitrary experimental settings (finite $d$, arbitrary informationally complete (IC) measurements, finite $N$), admits explicit, computable bounds on the probability that the estimated state is further from the true prepared state than a user-defined error threshold.

Problem Statement

In any QST experiment, an experimenter attempts to prepare a target state $\rho_*$, but due to practical imperfections, the actual state prepared is $\rho$. Tomography is performed to estimate this unknown $\rho$ from $N$ independent samples. The core question is: Given only the acquired data and the tomography protocol, what can be said (with high probability) about the distance between the point estimate and the true $\rho$? Moreover, can one directly bound the deviation between the experimental preparation and the ideal target state $\rho_*$, in terms of established quantum loss functions, without knowledge of $\rho$ itself?

The ENM Estimator and the Main Theoretical Result

The paper introduces the ENM estimator, which projects the standard (possibly unphysical) linear least-squares (LLS) solution onto the set of physical density matrices. Formally, for data $f_N$ the ENM estimate is

$$\rho_N^{\mathrm{ENM}} := \operatorname*{argmin}_{\rho' \in \mathcal{S}(\mathcal{H})} \| \rho' - \rho_N^{\mathrm{LLS}} \|_2,$$

where the norm denotes the Hilbert-Schmidt metric and $\mathcal{S}(\mathcal{H})$ is the set of physical density matrices (positive semidefinite, unit trace). Unlike the LLS estimator, $\rho_N^{\mathrm{ENM}}$ is always physical by construction.

The paper proves a non-asymptotic confidence-level bound for ENM estimation: for loss function $\Delta$ in $$\{\text{Hilbert-Schmidt}, \text{trace distance}, \text{infidelity}\}$$, for any true $\rho$, any finite $N$, any IC measurement, and any $\delta > 0$, the probability that $\Delta(\rho_N^{\mathrm{ENM}}, \rho) > \delta$ is upper-bounded by a computable function:

$$CL := 1 - 2 \sum_{\alpha=1}^{d^2 - 1} \exp\left[ -\frac{b}{c_\alpha \delta^2 N} \right],$$

where the constants $b$ and $c_\alpha$ depend only on the chosen distance measure and the measurement configuration. Critically, this confidence level is independent of the unknown prepared state $\rho$ and applies for arbitrary loss.

Technical Innovations

• General Applicability: The result covers any finite-dimensional quantum system, any IC measurement (including over-complete), arbitrary finite statistics, and loss functions relevant in quantum information (trace distance, infidelity, Hilbert-Schmidt).
• Explicit Formula: All quantities entering the confidence-level formula are computable from experimental settings, without further assumptions (no prior, no need for an explicit noise model, and no knowledge of the true $\rho$).
• Rigorous Non-Asymptotic Bounds: Unlike asymptotic (large $N$) or order-of-magnitude estimates, the approach yields immediately computable guarantees applicable to practical scenarios, including modest or moderate $N$.
• Comparison to Existing Work: In contrast to region estimators—whose volume and geometry complicate their interpretation—the point estimator ENM yields error probabilities directly meaningful for operator distance measures.
• Adaptability: The scheme is shown to extend to quantum process tomography (via Choi-Jamiolkowski duality) and accommodates common experimental imperfections such as finite detection efficiency.

Numerical and Analytical Results

The framework is instantiated for one- and two-qubit state tomography, considering realistic modeling of detection losses (parameterized via efficiency $\eta$). Explicit expressions for $c_\alpha$ are derived for standard Pauli measurement protocols, yielding concrete sample size requirements for desired precision/confidence trade-offs. For example, for a single qubit, achieving $\delta=0.07$ infidelity at $99\%$ confidence with $\eta=0.9$ detection efficiency requires $N=7500$ samples.

Implications and Future Directions

This methodology closes a critical gap in the statistical interpretation of QST data, elevating the rigor with which experimental claims (e.g., fidelity to a given target state) can be justified. The result is directly applicable to the certification of state preparation (including protocols requiring high-fidelity, such as quantum information processing and quantum error correction tasks), benchmarking, and any context where quantitative confidence thresholds are operationally necessary.

Potential directions for future exploration include:

• Tighter Confidence Levels: The bounds are guaranteed but may not be tightest possible, particularly near the boundaries of state space or when certain loss functions are used. Sharper concentration inequalities or state-dependent analysis could yield improvements.
• Optimality and Computational Complexity: Numerical studies of tightness and computational cost for high-dimensional systems, and comparison to alternative (e.g., Bayesian or region-based) approaches, remain open.
• Extension to Process Tomography and Ancilla-Assisted Schemes: While the Choi-representation provides a route, explicit generalizations and computational tools specific to QPT will further enhance experimental utility.
• Systematic and Numerical Uncertainties: The framework accommodates imperfect knowledge of measurement operators and algorithmic error, with explicit inclusion of supplementary error terms, and this can be extended to settings with complex noise and experimental drift.

Conclusion

The introduction of the ENM estimator and the corresponding precision-guaranteed confidence-level formula represents a robust advancement for quantum tomography, enabling objective certification of experimental outcomes even with finite data and imperfect systems. Its practicality, generality, and computability position it as a new standard for assessing tomographic precision in both foundational and applied quantum experiments ["Precision-guaranteed quantum tomography" (1306.4191)].