Optimality of Gaussian Entanglement of Formation

Technical Summary: Optimality of Gaussian Entanglement of Formation

Problem Statement
The paper addresses a longstanding open problem in continuous-variable (CV) quantum information theory, specifically Open Quantum Problem 29 posed in 2005. The central question is whether the entanglement of formation ($E_F$) of a two-mode Gaussian state coincides with its Gaussian restriction ($E_F^G$).

For a bipartite state $\rho$, the entanglement of formation is defined as the infimum of the average entanglement over all possible pure-state ensemble decompositions:
$$E_F(\rho) = \inf_{\{p_j, |\psi_j\rangle\}} \sum_j p_j E(|\psi_j\rangle)$$
where $E(|\psi_j\rangle)$ is the entropy of the reduced state. For a Gaussian state $\rho_V$, the Gaussian entanglement of formation $E_F^G(V)$ restricts this optimization to Gaussian pure-state ensembles only. Since the set of Gaussian ensembles is a subset of all possible ensembles, the inequality $E_F(\rho_V) \leq E_F^G(V)$ always holds. The open problem asks whether this inequality is strict for general two-mode Gaussian states, or if the optimal decomposition is always Gaussian. Previous work had resolved this for symmetric states but left the general asymmetric case unresolved, with various approaches failing to exclude non-Gaussian ensembles rigorously.

Methodology
The author develops a proof based on a sharp affine relation between entanglement and a generalized Einstein–Podolsky–Rosen (EPR) observable, valid for arbitrary pure two-mode states (Gaussian and non-Gaussian).

1. Generalized EPR Observable: The author defines a parameterized observable $\hat{D}_t$ involving a relative gain parameter $t$ ($0 < t \leq 1$):
   $$\hat{D}_t = (ta - b^\dagger)^\dagger (ta - b^\dagger) = \frac{1}{2} \left[ (tx_A - x_B)^2 + (tp_A + p_B)^2 + 1 - t^2 \right]$$
   where $t=1$ corresponds to the standard balanced EPR observable, while $t < 1$ accommodates asymmetric states.

2. Affine Extremality Theorem (Theorem 1): The core of the proof is an inequality relating the entanglement $E(\psi)$ of any normalized pure two-mode state $|\psi\rangle$ with finite mean energy to its expectation value of $\hat{D}_t$. For any $0 < q < t \leq 1$:
   $$E(\psi) + \lambda_{q,t} \left[ \langle \hat{D}_t \rangle_\psi - d_t(q) \right] \geq E(q)$$
   Here, $E(q)$ is the entanglement of a two-mode squeezed vacuum (TMSV) with squeezing parameter $q$, $d_t(q)$ is the expectation value of $\hat{D}_t$ for that TMSV, and $\lambda_{q,t}$ is a positive coefficient. This inequality implies that at a fixed entanglement, the TMSV minimizes the generalized EPR expectation.

3. Proof Strategy: The proof of Theorem 1 proceeds in three steps:

   • Singular-Value Inequality: A new inequality reduces the EPR functional to an expression depending only on the ordered Schmidt coefficients of the state, utilizing a "positive part" operation to avoid artificial penalties for upward steps in the rescaled sequence.
   • Geometric Block-Averaging: A procedure is introduced to average adjacent blocks of Schmidt coefficients (weighted by geometric factors $t^{2n}$) to produce a new sequence that satisfies a geometric ratio condition ($c_{n+1} \leq t c_n$) without increasing the entropy or the EPR functional. This step removes the restrictive geometric-ratio condition required by previous rearrangement arguments.
   • Logarithmic-Sobolev Inequality: A meta logarithmic-Sobolev inequality for phase-covariant Gaussian channels is applied to show that the remaining entropy-EPR trade-off is minimized by a geometric sequence (corresponding to a TMSV).

4. Extension to Mixed States: Because the inequality in Theorem 1 is affine in the expectation value of the observable, it can be averaged over any ensemble. This allows the author to construct a lower bound for the entanglement of formation of any mixed state (Gaussian or not) based on the measured expectation of $\hat{D}_t$.

Key Results

• Equality for Two-Mode Gaussian States: By applying the affine bound to the canonical decomposition of a two-mode Gaussian state (which can be brought via local symplectic transformations to a form involving a TMSV plus orthogonal noise), the author proves that the lower bound matches the known Gaussian upper bound. Consequently, they establish:
  $$E_F(\rho_V) = E_F^G(V)$$
  for every two-mode Gaussian state $\rho_V$. This confirms that the optimal decomposition for any two-mode Gaussian state is indeed Gaussian.

• Extension to Bisymmetric Multimode States: The result extends to $(m+n)$-mode Gaussian states that are bisymmetric across the $A:B$ partition. Such states are locally unitarily equivalent to a correlated two-mode core tensored with uncorrelated local thermal modes. Since local thermal modes do not affect $E_F$ or $E_F^G$, the equality holds for the entire bisymmetric state.

• Lower Bound for Non-Gaussian States: The derived inequality provides a measurable, quantitative lower bound on the entanglement of formation for arbitrary non-Gaussian two-mode states. This bound depends only on the variances of two generalized EPR quadratures and does not require full state tomography or photon-number-resolving detection.

Significance and Claims
The paper claims to resolve the full two-mode instance of Open Quantum Problem 29, establishing that the Gaussian restriction is not merely an approximation but the exact value of the entanglement of formation for all two-mode Gaussian states.

• Theoretical Impact: The result promotes established analytical formulas and efficient numerical algorithms for $E_F^G$ (previously restricted to Gaussian ensembles) to exact statements for the unrestricted convex roof $E_F$.
• Experimental Utility: The author demonstrates that the same experimental data used for standard EPR tests (phase-resolved homodyne or microwave measurements) can be reprocessed to provide a rigorous lower bound on $E_F$ for non-Gaussian states. In a specific example involving a lossy non-Gaussian "vacuum-pair" state, the optimized bound captured 98.0% of the exact entanglement of formation.
• Methodological Contribution: The proof reduces an infinite-dimensional optimization over arbitrary Schmidt bases to a one-dimensional optimization over a squeezing parameter. The techniques introduced, specifically the singular-value defect inequality and the geometric block-averaging method, are noted as potentially useful for other bosonic convex roof problems.

The paper concludes that while the generic nonsymmetric multimode problem remains open, the two-mode and bisymmetric cases are now fully solved, providing a definitive link between Gaussian optimization and the fundamental resource measure of entanglement of formation.