Optimality of Gaussian Entanglement of Formation

AI summary

Overview

Research area: Continuous-variable quantum information theory — entanglement quantification for Gaussian states of light and microwave fields. Technical level: Advanced (a mathematical physics proof paper; the results are stated in plain terms but the derivations rely on Schmidt decompositions, covariance matrices, and logarithmic-Sobolev inequalities). Scope: A single-author theoretical paper (Gerardo Adesso, University of Nottingham, arXiv:2608.01909v3) proving that the entanglement of formation of every two-mode Gaussian state equals its Gaussian restriction, that it is additive, and that it therefore equals the entanglement cost.

What This Paper Is About

For a quantum state, the entanglement of formation is the least average pure-state entanglement needed to prepare it, and its asymptotic version is the entanglement cost. For Gaussian states one can compute a "Gaussian" version of this quantity by restricting the optimization to Gaussian ensembles, which gives only an upper bound in general. This paper proves that for every two-mode Gaussian state no non-Gaussian ensemble can do better, so the bound is exact, and then shows the quantity is additive and equals the entanglement cost. The problem of whether the inequality could be strict was listed in 2005 as Open Quantum Problem 29.

Key Contributions

1. Exactness of the Gaussian entanglement of formation for all two-mode Gaussian states. The paper proves E_F(ρ_V) = E_F^G(V) for every two-mode Gaussian state, resolving the full two-mode instance of Open Quantum Problem 29.
2. A sharp affine extremality theorem for arbitrary pure two-mode states. Theorem 1 gives E(ψ) + λ_{q,t}[⟨D̂_t⟩ψ − d_t(q)] ≥ ℰ(q) for all 0 < q < t ≤ 1, where D̂_t is a generalized Einstein–Podolsky–Rosen (EPR) observable with relative gain t and λ{q,t} = −2q ln q/[(t−q)(1−tq)] > 0. The result holds for non-Gaussian states as well.
3. Additivity and entanglement cost. Theorem 2 extends the argument to m pairs of modes, giving E_C(ρ_V) = E_F(ρ_V) = E_F^G(V) for every two-mode Gaussian state, and yields closed-form entanglement costs for thermal-loss, thermal-amplifier and additive-noise bosonic channels.
4. Certified lower bounds for arbitrary non-Gaussian states. The same inequalities imply E_F(ρ) ≥ E_C(ρ) ≥ E_F^G(V) as written in Eq. (23), so any two-mode state is at least as entangled as the Gaussian state with the same covariance matrix, and the bound is measurable from second moments alone.

Main Findings

• Gaussian ensembles are optimal: For every two-mode Gaussian state, mixing non-Gaussian pure states cannot reduce the entanglement of formation below the Gaussian value; the equality E_F(ρ_V) = E_F^G(V) holds in general, extending an earlier result known only for symmetric two-mode states.
• Sharp trade-off between entanglement and EPR correlations: At fixed entanglement ℰ(q) with q ≤ t, the two-mode squeezed vacuum gives the smallest generalized EPR expectation d_t(q) = (1−tq)²/(1−q²). This proves the extremality conjecture of Ivan and Simon, and removes the extra condition on Schmidt coefficients required by the earlier rearrangement argument.
• Quantitative entanglement witness: Averaging the affine inequality over an arbitrary ensemble gives E_F(ρ) ≥ sup over 0 < q < t ≤ 1 of {ℰ(q) − λ_{q,t}[Tr(ρD̂_t) − d_t(q)]}. In the balanced case t = 1, with δ = Tr(ρD̂_1), this becomes E_F(ρ) ≥ g((1−δ)²/(4δ)) for 0 < δ < 1, where g(y) = (y+1)ln(y+1) − y ln y, and zero for δ ≥ 1.
• Additivity on tensor products: For a product of two-mode Gaussian states, the product of the optimal Gaussian ensembles attains the many-copy bound, so E_F(ρ^{⊗m}) = m E_F(ρ) and E_C = E_F = E_F^G for all two-mode Gaussian states.
• Closed-form channel costs: For thermal-loss, thermal-amplifier and additive-noise channels, E_C(𝒩) = g((1−δ_𝒩)²/(4δ_𝒩)) for 0 < δ_𝒩 < 1 and 0 for δ_𝒩 ≥ 1, where δ_𝒩 = (√(N_B+1)+√(N_B))² |1−√τ|/(1+√τ) for transmissivity or gain τ with N_B thermal photons, and δ_𝒩 = ξ for additive noise of variance ξ. Costs are in nats per use, with division by ln 2 giving the Bell-pair rate.
• Bisymmetric multimode extension: Any bisymmetric (m+n)-mode Gaussian state is locally Gaussian-unitarily equivalent to a correlated two-mode core tensored with m+n−2 uncorrelated local thermal modes, and E_F(ρ_bisym) = E_F^G(ρ_bisym) = E_F(ρ_core).
• Structural insight into the optimal ensemble: The optimal decomposition of a two-mode Gaussian state consists of identical two-mode squeezed vacua displaced by classical correlated noise whose covariance N_{t,u,v} is singular along the generalized EPR quadratures; the mixedness is entirely classical.
• Numerical illustration for non-Gaussian states: For a photon-pair state √(1−p)|00⟩ + √p|11⟩ sent through pure-loss channels with transmissivities η_A = 0.9 and η_B = 0.2, at p = 0.01 the optimized bound captures 98% of the exact value (from Wootters' formula) versus 88% for the balanced measurement t = 1. For a photon-subtracted state obtained by removing one photon from each mode of a two-mode squeezed vacuum, the bound captures at least 80% of the exact value over the range shown, and the certified increase persists under equal losses.
• Proof techniques: The result on Gaussian discord, and a connection between single-mode entropy and the Wigner–Yanase skew information, are also reported as by-products.

Methodology in Plain English

A pure two-mode state can be written as a matrix of coefficients in the photon-number basis, and the entanglement depends only on the singular values of that matrix (the Schmidt coefficients). The generalized EPR observable, however, depends on the Schmidt bases, which for a non-Gaussian state can be arbitrary — this is the obstacle. The proof removes it in three steps. First, a new singular-value inequality (Eq. 11) bounds the EPR functional using only the ordered Schmidt coefficients: G_t(C) ≥ Σ_n (n+1)(tc_n − c_{n+1})₊². Second, whenever adjacent coefficients violate the ratio condition c_{n+1} > t c_n, they are averaged over blocks with geometric weights t^{2n}, which enforces c̃_{n+1} ≤ t c̃_n without increasing either the EPR functional or the entropy. Third, a logarithmic-Sobolev inequality for phase-covariant Gaussian channels shows the remaining entropy–EPR trade-off is minimized by a geometric sequence — and a geometric sequence is exactly the Schmidt spectrum of a two-mode squeezed vacuum, so the Gaussian optimizer emerges from the argument rather than being assumed.

For the multimode additivity theorem, the strategy is variational: at a state minimizing the left-hand side, second variations along 2m Bogoliubov directions must be nonnegative, and by stationarity their sum reduces to 2×2 quadratic forms over pairs of Schmidt coefficients that are shown to be negative unless the state is the joint vacuum of those operators — again a tensor product of two-mode squeezed vacua. Combined with the covariance-geometry lemma that brings any entangled two-mode covariance matrix to V_q + N_{t,u,v} with q ≤ t ≤ 1, this pins the optimum from both sides.

Why This Matters

Impact on research: The paper closes a question open since 2005, upgrades previously conjectural formulas and numerical algorithms for E_F^G into exact statements about the unrestricted entanglement of formation, and settles the entanglement cost for a central family of states. It also shows a route around the Wolf–Giedke–Cirac extremality theorem, which requires strong superadditivity and therefore does not apply to the entanglement of formation.

Real-world applications (as discussed in the paper):

• Benchmarking entanglement in teleportation, communication networks, distributed quantum sensing, integrated photonics, and hybrid microwave–optical interfaces.
• Certifying entanglement-based communication rates from covariance matrices measured by balanced homodyne detection in optics, and by linear amplification plus heterodyne digitization in superconducting microwave circuits.
• Quantifying the entanglement cost of thermal-loss, thermal-amplifier and additive-noise channels used in optical communication.
• Assessing photon-subtraction-based entanglement distillation, where the certified increase in entanglement is bounded from second moments.

Industry relevance: Because the bound requires only measured second moments — no photon-number-resolving detection, no non-Gaussian tomography, no verification of Gaussianity — it can be folded into existing characterization workflows for optical and microwave quantum hardware, and a single data set serves for any gain t.

Future Directions

• Extending Gaussian optimality and additivity beyond two-mode cores to general multimode Gaussian states, which the paper explicitly leaves open.
• Using the measurable bounds in more experimental settings, including calibration and finite-statistics issues discussed in the Supplemental Material.
• Applying local filtering before quadrature measurement when the covariance matrix hides the entanglement of a non-Gaussian state, so that the filtered bound times the success probability certifies both E_F and E_C for the original state.
• Pursuing the connections noted in the outlook: optimality of Gaussian measurements for quantum discord (including its regularization) in two-mode Gaussian states with one vacuum normal mode, and the link between single-mode entropy and the Wigner–Yanase skew information.

Target Audience

Quantum information theorists and mathematical physicists working on continuous-variable entanglement, especially those concerned with entanglement monotones, Gaussian channels, and convex-roof optimization. Experimentalists in optical and microwave quantum technology who characterize entanglement from covariance matrices will also benefit, as will graduate students seeking a worked example of how Gaussian optimizers can be derived rather than assumed. The paper notes substantive use of AI tools in developing the results, which may interest readers tracking AI-assisted mathematical physics.

Authors’ abstract

We prove that the entanglement of formation of every two-mode Gaussian state coincides with its Gaussian restriction, solving a longstanding open problem in continuous variable quantum information theory. The key result is a sharp affine relation between entanglement and a generalized Einstein--Podolsky--Rosen observable, valid for arbitrary pure two-mode states, including non-Gaussian ones. We further prove that the entanglement of formation is additive on arbitrary tensor products of two-mode Gaussian states, and hence coincides with the entanglement cost, settling the latter for all two-mode Gaussian states. As a consequence, we obtain in closed form the entanglement cost of thermal-loss, thermal-amplifier and additive-noise bosonic channels. Our results, which also extend to bisymmetric multimode states, establish that no two-mode state is less entangled than the Gaussian state with the same covariance matrix, yielding measurable lower bounds on both the entanglement of formation and the entanglement cost of arbitrary non-Gaussian states.