Near-optimal pure state estimation with adaptive Fisher-symmetric measurements

1. Problem Statement

Quantum state estimation (tomography) is fundamental for quantum information processing, including communication, computation, and metrology. The goal is to reconstruct an unknown quantum state $\rho$ from measurement data.

• The Challenge: For a $d$-dimensional pure state, the fundamental limit on estimation accuracy using individual (separable) measurements is given by the Gill-Massar Lower Bound (GMB). The average infidelity $\bar{I}$ is bounded by $\bar{I} \ge \frac{d-1}{N}$, where $N$ is the sample size.
• Limitations of Existing Methods:
  • Fisher Symmetric Measurements (FSM): These are optimal POVMs that saturate the GMB. However, they are locally informationally complete, meaning they only work accurately if the unknown state is close to a pre-defined "fiducial" state. If the state is far from the fiducial state, the estimation error is dominated by approximation errors, rendering the method ineffective without prior knowledge.
  • Global/Collective Measurements: Methods that achieve the GMB without prior knowledge often require collective measurements on multiple copies of the state, which are experimentally difficult to implement.
  • Adaptive Methods: Fully adaptive methods exist but often require a large number of iterations or measurement bases, increasing the total number of outcomes significantly.

The authors aim to develop an adaptive estimation protocol that achieves the GMB (near-optimal accuracy) for arbitrary unknown pure states using only local (separable) measurements, while keeping the number of measurement outcomes linear in the dimension $d$.

2. Methodology: Three-Stage Adaptive Protocol

The authors propose a three-stage adaptive method (AFSM) that combines a single-shot measurement with two sets of FSMs.

• Stage 0: Fiducial State Selection (Single-Shot)

  • Perform a single-shot projective measurement on the unknown state $|\Psi\rangle$ in an arbitrary basis.
  • Select the outcome vector $|\phi\rangle$ as the fiducial state.
  • Purpose: This ensures the fiducial state has a non-zero overlap with the unknown state, avoiding the singularity where the overlap is zero.

• Stage 1: Preliminary Estimation (Two Non-Optimal FSMs)

  • Construct two FSMs, $E_+$ and $E_-$, based on the fiducial state selected in Stage 0. These are defined as $|\phi^\alpha_\pm\rangle = \beta^\alpha_0 |0\rangle \pm \sum \omega^\alpha_k |k\rangle$.
  • Measure the unknown state using these two POVMs with a sample size $N_1$.
  • Analytical Reconstruction: Use the difference in measurement probabilities ($P^\alpha_+ - P^\alpha_-$) to analytically solve for the coefficients of the state relative to the fiducial basis.
  • Output: A preliminary estimate $|\tilde{\Psi}\rangle$. While this estimate is not optimal (due to the state potentially being far from the fiducial), it is sufficiently accurate to serve as a new fiducial state for the next stage.

• Stage 2: Near-Optimal Adaptation (Adapted FSM)

  • Construct a new FSM, $\tilde{E}$, by applying a unitary transformation $U$ to a standard FSM. This transformation maps the standard fiducial state $|0\rangle$ to the preliminary estimate $|\tilde{\Psi}\rangle$ (i.e., $|\tilde{\Psi}\rangle = U|0\rangle$).
  • Measure the unknown state using this adapted FSM $\tilde{E}$ with sample size $N_2$.
  • Final Estimation: Combine the data from Stage 1 and Stage 2. Use Maximum Likelihood Estimation (MLE) with $|\tilde{\Psi}\rangle$ as the starting point to derive the final estimate $|\hat{\Psi}\rangle$.
  • Total Outcomes: The protocol requires $7d - 3$ total measurement outcomes (or $7d$ if using bases instead of POVMs), scaling linearly with $d$.

3. Key Contributions

1. Extension of FSMs to Arbitrary States: The paper demonstrates that FSMs, traditionally limited to local neighborhoods, can be applied to any unknown pure state by using an adaptive strategy to update the fiducial state.
2. Theoretical Error Bounds:
   • The authors derive finite-sample, high-probability upper bounds on the reconstruction infidelity.
   • They prove that for sufficiently large $N$, the infidelity scales as $O(d/N)$, which matches the scaling of the Gill-Massar bound.
   • They quantify the impact of the "approximation error" ($\kappa$) caused by the linearization of probabilities, showing that this error becomes negligible when the sample size is large enough to ensure the preliminary estimate is close to the true state.
3. Efficiency: The method avoids collective measurements and achieves near-optimal accuracy with a total number of outcomes scaling as $7d - 3$. This is more efficient than standard 5-basis tomography (5BBT) which scales as $O(d^{1.87})$ and less resource-intensive than methods requiring $O(d^2)$ outcomes.
4. Robustness to Noise: The paper extends the method to slightly mixed states (affected by depolarizing noise), showing that the protocol can estimate the purity parameter $\lambda$ and the pure state component simultaneously, achieving performance close to $2 \times$ the GMB under weak noise.

4. Results

• Numerical Simulations: Monte Carlo simulations were performed for dimensions $d \in \{4, 8, 16, 32, 64\}$ and various sample sizes.
  • Stage 1 Performance: The preliminary estimate showed infidelity scaling roughly as $O(d^2/N)$, consistent with theoretical predictions for non-optimal FSMs.
  • Stage 2 Performance: The final adapted estimate achieved an average infidelity extremely close to the GMB ($\bar{I} \approx \frac{d-1}{N}$).
  • Scaling: The numerical fits confirmed the scaling $\bar{I} \approx C \frac{d-1}{N}$, where $C \approx 1.3$ for the optimal sample splitting ($N_1 = N_2 = N/2$).
• Comparison:
  • vs. 5-Basis Tomography (5BBT): The proposed method achieves better accuracy with fewer measurement bases (effectively 7 bases vs. 5, but with better scaling and adaptivity).
  • vs. Projected Least Squares (PLST): AFSM offers a quadratic advantage in sample size $N$ compared to PLST for pure states.
• Convergence: The simulations showed that for large $N$, the approximation error $\kappa$ vanishes, and the protocol saturates the GMB. For small $N$, the approximation error dominates, but the adaptive nature still provides a significant advantage over non-adaptive methods.

5. Significance

• Experimental Feasibility: The protocol is designed for local measurements only, making it highly suitable for current experimental platforms like Integrated Photonic Processors (IPPs) and quantum computers, where collective measurements are infeasible.
• Scalability: The linear scaling of measurement outcomes ($O(d)$) makes the method viable for high-dimensional quantum systems (qudits), which are increasingly important in quantum computing.
• Benchmarking: The method provides a rigorous, near-optimal tool for characterizing and benchmarking quantum hardware, ensuring that state preparation and gate operations are verified with minimal resource overhead.
• Theoretical Insight: The work bridges the gap between locally optimal measurements (FSM) and global optimality, demonstrating that a simple two-step adaptation (fiducial selection + FSM adaptation) is sufficient to overcome the locality constraint of FSMs.

In conclusion, this paper presents a practical, theoretically grounded, and experimentally feasible protocol for quantum state tomography that achieves near-optimal accuracy for arbitrary pure states without requiring collective measurements, significantly advancing the state of the art in quantum characterization.