Closing a single-copy gap to the Holevo bound with two-copy measurements

Technical Summary: Closing a Single-Copy Gap to the Holevo Bound with Two-Copy Measurements

Problem Statement
The Holevo Cramér-Rao bound ($C_H$) defines the ultimate asymptotic precision limit for quantum multiparameter estimation when collective measurements on infinitely many identical copies are permitted. However, in practical scenarios involving finite resources, a fundamental question remains: can the Holevo bound be attained exactly with a finite number of copies ($n$)?

For two-parameter estimation models, the "gap-persistence theorem" establishes that if the optimal single-copy estimation cost ($C^{(1)}$) is strictly greater than the Holevo bound ($C^{(1)} > C_H$), this gap persists for any finite number of copies ($C^{(n)} > C_H$ for all finite $n$). It was an open question whether this persistence holds universally for models with three or more parameters. Specifically, it was unknown whether a strict single-copy gap could be closed by a collective measurement on a finite number of copies (e.g., $n=2$) in higher-dimensional parameter spaces.

Methodology
The author addresses this question through a combination of theoretical construction, rigorous proof, and experimental implementation:

1. Theoretical Construction of a Counterexample:
   The author constructs a specific three-parameter, rank-two quantum estimation model acting on a four-dimensional Hilbert space ($\mathcal{H} = \mathbb{C}^2_r \otimes \mathbb{C}^2_t$). The state family is defined as $\rho_\theta = U_\theta \rho_0 U_\theta^\dagger$, where $\rho_0 = |0\rangle\langle 0|_r \otimes I_t/2$ and the unitary $U_\theta$ encodes three parameters ($\theta_1, \theta_2, \theta_3$) via Pauli operators.

   • They analytically derive the Holevo bound for this model, finding $C_H = 5/8$.
   • They calculate the exact optimal single-copy cost ($C^{(1)}$) by optimizing conditional measurements on the qubit subsystem, finding $C^{(1)} = (3 + 2\sqrt{2})/8 \approx 0.729$. This confirms a strict single-copy gap ($C^{(1)} > C_H$).
   • They construct an explicit collective measurement on two copies ($n=2$) using a blockwise estimator observable strategy. This measurement is shown to be locally unbiased and achieves a normalized cost of exactly $5/8$, thereby attaining the Holevo bound.

2. Theoretical Proof for Full-Rank Models:
   To characterize the necessity of the counterexample's properties, the author proves a theorem stating that for any smooth, locally identifiable, full-rank model (where the density matrix is strictly positive), a strict single-copy gap implies a persistent gap for all finite $n$. The proof utilizes Naimark dilation to show that exact Holevo saturation in full-rank models requires the commutativity of estimator operators, which, combined with the strict convexity of the Holevo functional, forces the single-copy optimizer to also commute and attain the bound. This establishes that rank deficiency (specifically, first-order motion of the state support) is a necessary condition for finite-copy gap closure.

3. Experimental Implementation:
   The theoretical two-copy protocol was implemented on the IBM superconducting quantum processor ibm_boston using two transmon qubits.

   • Hybrid Protocol: The experiment utilized a hybrid approach where the classical flags (outcomes of measuring the $t$-qubit in the $\sigma_x$ basis) were sampled classically rather than physically measured, allowing the conditional encodings to be applied to single transmons. This reproduced the exact statistics of the ideal two-copy model.
   • Noise Analysis: The author introduced controlled depolarizing noise ($\epsilon$) to the system. They demonstrated that while exact attainment of the Holevo bound is impossible for any $\epsilon > 0$ (due to the state becoming full-rank), the two-copy collective measurement retains a precision advantage over the single-copy optimum and separate measurements over a finite noise range.

Key Results

• Disproof of Universal Persistence: The constructed three-parameter model demonstrates that a strict single-copy gap ($C^{(1)} > C_H$) does not necessarily persist for all finite $n$. Specifically, the gap closes exactly at $n=2$. This disproves the universality of the gap-persistence theorem beyond two parameters.
• Nagaoka-Hayashi Bound Closure: The study also shows that the gap between the Nagaoka-Hayashi (NH) bound and the Holevo bound, which is strict for single copies, closes at two copies for this model.
• Rank Deficiency as a Condition: The paper proves that for full-rank models, finite-copy gap closure is impossible. Thus, rank deficiency is identified as a necessary condition for the phenomenon observed.
• Experimental Verification: The experiment measured a two-copy estimation cost of $C^{(2)}_{exp} = 0.709(9)$, which is significantly lower than the certified single-copy lower bound of $C^{(1)}_{opt} = 0.741(7)$ (a 2.8 standard deviation advantage). The collective measurement also outperformed the separate-measurement bound ($0.778(8)$) by 5.9 standard deviations.
• Noise Robustness: The precision advantage of the two-copy measurement persists even when depolarizing noise is introduced, although exact saturation of the Holevo bound is lost immediately upon the introduction of noise ($\epsilon > 0$).

Significance and Claims
The paper claims to resolve a long-standing open question regarding the attainability of the Holevo bound with finite resources. By providing a counterexample, it establishes that the "gap-persistence" phenomenon is not universal in multiparameter quantum estimation. The work clarifies the structural requirements (rank deficiency) for closing such gaps and demonstrates that the ultimate precision limits of quantum multiparameter estimation can be accessed with finite ensembles under specific conditions.

The author emphasizes that their results connect the attainability of ultimate precision to the specific structure of the estimation model and the availability of collective measurement resources. They note that while exact saturation requires noiseless, rank-deficient models, a measurable collective advantage persists in noisy, realistic settings, bridging the gap between asymptotic theory and finite-resource experimental reality. The paper does not claim to solve the general problem for all models but rather identifies the specific class of models (rank-deficient with moving support) where finite-copy optimality is achievable.