PaperlayerView exact revision on arXiv

Source-linked AI summary

Strong converse for the quantum capacity of the pure-loss bosonic channel

Mark M. Wilde

arXiv:2609.16608v1quant-phcs.IT

TL;DR

The paper asks whether unconstrained quantum communication over a pure-loss bosonic channel obeys a strong converse despite infinite-dimensional, potentially high-energy and correlated code states. It combines balanced-channel fluctuation bounds with testing inequalities and a degrading-channel reduction. The result is an O(1/n) fidelity bound above capacity for every code, establishing the strong converse while leaving exponential rates and finite-energy extensions open.

  • Problem

    A capacity formula does not determine above-capacity fidelity, especially for bosonic channels with infinite-dimensional inputs and arbitrarily high-energy code states.

  • Method

    The proof combines quantum Chebyshev and hockey-stick testing with a uniform relative-entropy-variance bound for balanced loss, then reduces general transmissivity through a degrading attenuator.

  • Results

    At every fixed rate gap above capacity, every unconstrained code has entanglement-generation fidelity bounded by O(1/n) and therefore tending to zero.

  • Takeaways & Limitations

    The unconstrained quantum capacity of every pure-loss bosonic channel satisfies the strong converse for arbitrary encoded states and joint decoders.

  • Takeaways & Limitations

    The paper does not establish an exponential strong-converse exponent for the unconstrained bosonic channel.

Abstract

from arXiv · show

This paper reports the proof of a strong converse for the unconstrained quantum capacity of the pure-loss bosonic channel. At every fixed rate above capacity, the entanglement-generation fidelity of every code is bounded by a constant times the reciprocal of the number of channel uses. The bound holds without an energy constraint and for arbitrary encoded states, including states correlated across all input modes, and arbitrary joint decoders. The proof combines quantum Chebyshev and hockey-stick testing inequalities with a uniform relative-entropy-variance bound for the balanced pure-loss channel, corresponding to transmissivity $η=1/2$. The variance bound follows by expressing the balanced beam splitter in bright and dark modes: the dark modes are exactly in vacuum, and any state orthogonal to that vacuum contains at least one dark photon. For general transmissivity, dilating the degrading attenuator reduces the problem to this balanced-channel setting and bounds the decoder test by precisely the factor that produces the known quantum-capacity threshold. The resulting argument establishes the strong converse at the unconstrained quantum capacity for every pure-loss bosonic channel.

1 Introduction

The paper establishes a strong converse for unconstrained quantum communication over pure-loss bosonic channels, addressing the gap between capacity formulas and above-capacity code performance. Its proof combines uniform fluctuation control for balanced loss with a degrading-channel reduction.

  • Motivation: Above the quantum capacity, every code sequence has entanglement-generation fidelity tending to zero, unlike a merely weak or pretty-strong converse.This distinction matters for bosonic channels because infinite-dimensional inputs allow arbitrarily high-energy code states.
  • Background: For transmissivity η > 1/2, the unconstrained quantum capacity is log2[η/(1 −η)] qubits per mode; for η ≤ 1/2, it is zero.The paper studies the unassisted, unconstrained setting rather than energy-constrained or assisted capacities.
  • Contribution and proof idea: The proof gives an O(1/n) fidelity bound at every fixed positive rate gap for arbitrary encoded states and decoders.The bound is independent of photon support, reference dimension, and correlations across input modes.
  • Contribution and proof idea: For general transmissivity, dilating the degrading attenuator converts the channel to a balanced-beam-splitter problem and controls the decoder comparison trace without adding the extra bosonic output dimension.The paper develops these estimates directly rather than invoking finite-dimensional strong-converse reductions.
  • Contribution and proof idea: A balanced beam splitter makes receiver and environment symmetric, while its dark vacuum mode yields a uniform relative-entropy-variance bound of at most 4n.The estimate applies to correlated code states and does not depend on input energy.

2 Channel, code model, and main theorem

The paper defines the pure-loss channel and unrestricted entanglement-generation code model, then proves a quantitative strong-converse theorem. The proof uses balanced-channel symmetry, variance control, and a degrading-channel factorization to extend the result to arbitrary transmissivity.

  • 2.1 Notation and the pure-loss channel: The pure-loss channel mixes each input mode with vacuum at a beam splitter of transmissivity η and discards one output; qη is the known unconstrained capacity threshold.The converse derives this threshold from beam-splitter structure, while achievability is not reproved.
  • 2.2 Codes and fidelity: A code permits arbitrary reference states, encoders, correlations across all n input modes, and joint decoding, without assistance or an energy constraint.The model includes encoder-generated states and does not require purity, finite energy, or Gaussian structure.
  • 2.2 Codes and fidelity: For any rate exceeding qη by a fixed gap, the theorem bounds fidelity by a term of order 1/n plus an exponentially decaying term, forcing fidelity to zero.The resulting corollary establishes the strong converse at the unconstrained quantum-capacity threshold.
  • 2.3 Why balanced loss suffices: At η = 1/2, receiver–environment symmetry makes coherent information vanish, and quantum Chebyshev testing uses the variance bound to control decoder acceptance.The balanced estimate is first developed for finite photon support and then extended by fidelity convergence.

3 A quantum Chebyshev bound in hockey-stick form

This section develops a noncommutative testing bound that converts relative-entropy variance into an acceptance-probability estimate, including for unnormalized comparison operators.

  • 3.1 Comparison operators and testing: The hockey-stick framework allows an unnormalized comparison operator S, with Tr ST interpreted as a comparison weight rather than a probability.This form is used because the coding proof compares against operators such as I_R ⊗ σ_B.
  • 3.1 Comparison operators and testing: Support compression restricts the analysis to the positive support of S without changing either tested trace or the hockey-stick quantity.The state and comparison operator vanish outside this support, making the remaining logarithms well-defined.
  • 3.1 Comparison operators and testing: The logarithms are defined on positive supports, with singular-state terms handled by 0 ln 0 = 0 and kernel extensions annihilated by the state.The weighted expressions therefore do not depend on how ln σ is extended on its kernel.
  • 3.2 Logarithmic trace inequality: The logarithmic trace inequality replaces the invalid noncommutative identification between {ρ − γS > 0} and a logarithmic likelihood-ratio threshold.It bounds a weighted logarithmic average on the actual Neyman–Pearson subspace.
  • 3.2 Logarithmic trace inequality: For D = D(ρ∥S) and variance V, the centered quantum Chebyshev argument bounds every effect T by V/(V+t^2) plus e^(D+t)Tr ST.The threshold projector is P_t = {ρ − e^(D+t)S > 0}.
  • 3.2 Logarithmic trace inequality: The proof applies the logarithmic trace inequality to the centered log-likelihood operator, whose mean is zero and whose squared expectation is V.A centered-projector estimate yields the denominator V+t^2.
  • 3.2 Logarithmic trace inequality: Replacing the refined V/(V+t^2) estimate by the prior V/t^2 form changes a finite-blocklength denominator but not the strong-converse conclusion.The general variance-to-converse strategy follows established work by Sharma–Warsi and Cheng–Hsieh.

4 A uniform information-variance bound for balanced loss

The balanced pure-loss channel has zero relative entropy and variance at most 4n uniformly over finite-support correlated inputs. Bright–dark mode symmetry and a one-photon estimate provide the channel-specific control.

  • 4 A uniform information-variance bound for balanced loss: The balanced-channel lemma proves D(ω_RB^n∥I_R ⊗ σ_B^n) = 0 and V(ω_RB^n∥I_R ⊗ σ_B^n) ≤ 4n.The bound is independent of the photon cutoff, reference dimension, and correlations among modes.
  • 4 A uniform information-variance bound for balanced loss: Receiver–environment exchange symmetry at transmissivity 1/2 makes the receiver and environment marginals equal, including inputs correlated with the reference and across modes.Purity then identifies the relevant entropy difference with coherent information and yields the zero relative entropy.
  • 4 A uniform information-variance bound for balanced loss: Finite photon-number support guarantees that the logarithmic operators, commutators, and dark-mode number operator are defined on finite-dimensional photon sectors throughout the proof.The argument then extends the resulting variance estimate through the displayed commutator bound and its sum over modes.
  • 4 A uniform information-variance bound for balanced loss: The dark-mode number operator has eigenvalue zero on the joint vacuum and at least one on its orthogonal complement, providing a spectral gap.The balanced output lies in the dark vacuum, so the vacuum projector can isolate the orthogonal sector.
  • 4 A uniform information-variance bound for balanced loss: In bright–dark coordinates, the bright modes carry the signal while the dark modes remain in the vacuum entering the unused beam-splitter ports.The coordinate change is an inverse beam-splitter transformation for analysis, not an additional protocol operation.
  • 4 A uniform information-variance bound for balanced loss: Receiver–environment exchange acts as dark-mode photon-number parity, sending the logarithmic fluctuation Y|ψ⟩ into the odd dark-parity sector orthogonal to the vacuum.This removes the dark-vacuum component needed for the number-operator estimate.
  • 4 A uniform information-variance bound for balanced loss: Creating one dark photon converts the fluctuation estimate into commutator norms, which are bounded independently of input energy through a cross-partial-trace argument.The positive-support range property ensures lowering does not map supported vectors into the kernel, allowing the logarithmic comparison.
  • 4 A uniform information-variance bound for balanced loss: The variance estimate yields a testing inequality for every effect T, bounding Tr ωT by a variance term plus e^a Tr[(I_R ⊗ σ)T].This is the form later applied to decoder tests.

5 The decoder comparison trace at general transmissivity

For general transmissivity, a degrading attenuator factors the channel through a balanced splitter. Applying the balanced testing bound gives the comparison weight that yields the quantum-capacity threshold.

  • 5 The decoder comparison trace at general transmissivity: For η ≥ 1/2, the receiver can be further attenuated so the resulting three-output dilation contains a balanced pure-loss channel.The balanced splitter acts on the intermediate system, while the remaining output is the auxiliary system used in the comparison.
  • 5 The decoder comparison trace at general transmissivity: The two Gaussian-isometry factorizations agree on every Fock vector and therefore remain equal for correlated multimode inputs; at η = 1/2 the auxiliary mode is fixed vacuum.This realizes the symmetric-channel construction directly on bosonic Fock space.
  • 5 The decoder comparison trace at general transmissivity: The balanced-channel testing inequality applies with the enlarged reference RF^n, while the logical target remains the original reference R and message dimension M.Finite photon support permits restricting the auxiliary reference to a finite-dimensional support.
  • 5 The decoder comparison trace at general transmissivity: The degrading attenuation contributes a comparison factor τ^-n, with −ln τ = q_η ln 2 for η ≥ 1/2.The identity for the attenuation map is established through its Kraus operators and bounded positive extension.
  • 5 The decoder comparison trace at general transmissivity: The construction uses finite-support encoded states and compressed effects, while leaving the physical channel unchanged.The finite-support assumptions make the relevant traces finite; optional compression to supp σ does not change fidelity or the comparison trace.
  • 5 The decoder comparison trace at general transmissivity: The decoder test has comparison weight at most [η/(1 − η)]^n/M, producing the capacity term in the final bound.The test is evaluated on the dilated state without changing the code fidelity.

6 Completion of the proof for arbitrary codes

The proof extends the fidelity bound from balanced channels and finite-support pure inputs to arbitrary transmissivity, arbitrary pure inputs, and all normal encoded states. At any positive rate gap, the resulting fidelity decays as O(1/n).

  • Extension to arbitrary transmissivity: For 0 ≤η ≤1/2, cascading the channel as Lη = L2η ◦L1/2 reduces the proof to the balanced-channel case and preserves the bound for all transmissivities.The decoder is composed with L2η^⊗n, and the balanced-channel estimate is applied with τ = 1 and qη = 0.
  • Removing the photon-support cutoff: The cutoff removal yields the same fidelity inequality for arbitrary pure inputs, including states with infinite mean photon number.The estimate is independent of the finite photon-support cutoff, so convergence is taken for a bounded measurement probability before asymptotic blocklength limits.
  • Mixed inputs and arbitrary encoders: Convex spectral decomposition extends the pure-state bound to every normal encoded state, including outputs of arbitrary CPTP encoders acting on one half of ΦM.All spectral components use the same logical register and satisfy the same bound, whose convex combination therefore does as well.
  • Evaluation at the rate gap: If the rate gap satisfies δn ≥ n∆ > 0, both terms in the finite-blocklength bound decrease with δn, yielding the theorem’s fixed-gap converse.Replacing δn by n∆ gives the stated bound in (12).

7 Conclusion and future research

The paper establishes an unconstrained strong converse for unassisted quantum communication over every pure-loss bosonic channel, while identifying limits of the present polynomial-rate proof and several open extensions.

  • Conclusion: At each fixed positive rate gap, every encoded state and joint decoder has entanglement-generation fidelity tending to zero as O(1/n), without energy, Gaussianity, or independence assumptions.The result applies even to encoded states with infinite mean photon number.
  • Proof mechanism: The proof combines balanced-channel symmetry, dark-mode vacuum fluctuations, a 4n variance bound, and quantum Chebyshev testing in hockey-stick form.For general transmissivity, the decoder test is evaluated through the balanced-channel reduction.
  • Open questions: The argument proves polynomial rather than exponential fidelity decay and does not establish an exponential strong-converse exponent.An exponential result would require uniformly controlled noncommutative higher or exponential moments, beyond the variance estimate.
  • Open questions: Near capacity, the bound already forces vanishing fidelity when δn/√n →∞, but the order-√n regime and a sharp second-order expansion remain unresolved.The paper neither identifies a dispersion coefficient nor proves matching achievable bounds.
  • Open questions: An energy-constrained strong converse remains a separate question because mean-photon-number and photon-number-occupation constraints are not interchangeable.An energy-dependent refinement must retain the relevant constraint in the comparison-trace or fluctuation bound.
  • Open questions: Extensions to thermal-noise attenuators and other Gaussian channels require replacements for the exact balanced-output symmetry, dark-vacuum estimate, and comparison-trace bound.The present proof does not establish those extensions.

Statement on AI-assisted preparation

The manuscript reports extensive use of ChatGPT Pro 6 Astra during development, while assigning responsibility for verification and final content to the author.

  • Statement on AI-assisted preparation: ChatGPT Pro 6 Astra assisted with proof strategies, literature searches, proof review, exposition, LATEX preparation, and manuscript preparation.The author states that the initial proof methods were not provided in advance.
  • Statement on AI-assisted preparation: The author revised the manuscript and remains responsible for verifying the mathematical arguments, calculations, citations, conclusions, and final content.

A Proof of the logarithmic trace inequality

The appendix proves the logarithmic trace inequality by interpolation, differentiating the logarithm through a resolvent integral, and decomposing each integrand into nonnegative trace terms. Singular cases follow by regularization.

  • Scope of the inequality: The appendix’s inequality is a scalar trace inequality, not an operator-positivity claim about weighted products or logarithms compressed by the difference projector.The proof is related to earlier weighted trace inequalities but makes the scalar scope explicit.
  • Strictly positive interpolation path: For strictly positive A and B, the proof interpolates with Zs = (1 −s)B + sA and differentiates the logarithm under an integral representation.Uniform positivity supplies an integrable derivative bound that justifies the differentiations and integrations.
  • Nonnegative integrand decomposition: The derivative is rewritten using positive and negative parts of the difference, producing a nonnegative decomposition for every integrand.Cyclicity and positivity of the relevant operators make the resulting traces real and nonnegative.
  • Singular first argument: For a singular first argument, replacing A and B by Aε and Bε preserves the positive-difference projector and permits passage to the limit ε ↓0.Continuity of x ln x at zero and convergence of ln Bε establish the limiting inequality.

B Receiver–environment exchange as dark-mode parity

The appendix proves that receiver–environment exchange equals the parity operator for the total number of dark-mode photons. Consequently, bright–dark Fock states with even dark-photon number are invariant, while those with odd number acquire a minus sign.

  • The proof uses the spectral-calculus definition (−1)^Ndk := e^iπNdk for the total dark-mode number operator.
  • The exchange leaves bright-mode creation operators unchanged but multiplies each dark-mode creation operator by −1.
  • Equality of the swap and parity operators follows because they agree on every bright–dark Fock basis vector, whose linear span is complete.
  • The joint vacuum is invariant under receiver–environment exchange, consistent with its zero dark-photon parity.
  • Receiver–environment exchange is exactly total dark-mode photon-number parity, acting as +1 on even and −1 on odd dark-photon sectors.The identity is established by comparing the operators on a complete bright–dark Fock basis and extending equality by continuity.
Loading 2609.16608v1…