Gaussian-measurement equality with the Holevo bound
- Field
- Topics
Problem
When does the optimal single-copy Gaussian-measurement cost equal the Holevo bound? Give necessary and sufficient conditions in terms of the local first moments, covariance, their first derivatives, and the weight matrix defined below. Let \(\rho_\theta\) be a smooth faithful \(n\)-mode Gaussian model, with integers \(n,p\geq1\) and \(\theta\in\Theta\subset\mathbb R^p\) for an open set \(\Theta\). Write \(R=(q_1,p_1,\ldots,q_n,p_n)^{\mathsf T}\) for the canonical quadrature vector. Its mean vector and covariance are defined in Eq. (1). The commutation convention is Eq. (2).
Assume the symmetric logarithmic derivative quantum Fisher information matrix is nonsingular. Let \(W\) be a real positive-definite \(p\times p\) weight matrix.
Define the optimal single-copy Gaussian-measurement cost by Eq. (3).
where \(F_M\) is the classical Fisher information matrix of measurement \(M\), and singular \(F_M\) has infinite cost. Gaussian measurements mean Gaussian-ancilla preparations followed by Gaussian unitaries and homodyne detection with arbitrary classical processing.
Define the Holevo bound \(C_H(\theta,W)\) as the infimum of Eq. (4).
The infimum is over Hermitian observables \(X_1,\ldots,X_p\) with finite second moments, subject to Eq. (5).
The measurement is optimized locally at the specified \(\theta\); its setting is held fixed when taking derivatives for \(F_M\). The symmetric logarithmic derivatives \(L_j\) satisfy \(2\partial_j\rho_\theta=\rho_\theta L_j+L_j\rho_\theta\). Their Fisher matrix has entries \(\operatorname{Re}\operatorname{Tr}(\rho_\theta L_jL_k)\). The requested criterion must characterize \(C_G(\theta,W)=C_H(\theta,W)\), rather than merely restate these two optimizations. Equality of the infima is the target. A limiting sequence of Gaussian measurements counts, even if no individual measurement attains the value.
Source
This attainability classification is derived from the Gaussian-measurement restriction identified as an open direction in the Discussion of Chang, Genoni, and Albarelli [CGA26]. It is narrower than asking how to evaluate the Holevo bound itself.
Progress
The Holevo minimization for a faithful Gaussian model can be restricted exactly to centered linear and symmetrized quadratic quadrature observables. Chang, Genoni, and Albarelli prove the required invariant-subspace property in Derivations and give a finite semidefinite program in Eqs. (36) and (72) [CGA26]. This computes \(C_H\) from the local moments and derivatives. It does not show that a Gaussian measurement attains it.
For Gaussian shift models with parameter-independent covariance, Gaussian measurements attain the Holevo bound. Bradshaw, Lam, and Assad construct the measurement in Sec. III.B and extend it to arbitrary finite mode and displacement-parameter counts in Appendix C [BLA18]. A positive weight can be absorbed into an invertible linear reparameterization. The result does not cover general covariance parameters.
Comment
The unresolved question is a general criterion for equality with the restricted single-copy measurement cost. A finite optimization of the Holevo bound already exists. Quadratic optimizing observables can require non-Gaussian detection, and collective asymptotic attainability is a different property.