PaperlayerView exact revision on arXiv

Source-linked AI summary

An exponential strong converse for private communication over degradable quantum channels

Mark M. Wilde

arXiv:2609.32515v1quant-phcs.ITmath.IT

TL;DR

Private communication must be reliable while keeping information secret from the environment, and a capacity formula alone does not rule out above-capacity transmission with error bounded away from its maximum. This paper combines privacy-test overlap bounds with a single-filter reduction to prove exponential strong converses under a joint fidelity criterion. The result covers finite-dimensional degradable channels above the coherent-information threshold, with additional zero-rate and erasure-channel conclusions.

  • Problem

    A capacity formula alone does not rule out transmission above capacity with error bounded away from its maximal value, and private communication must satisfy both reliability and secrecy from the environment.

    Source: §1.1 · p. 3

  • Method

    The proof combines a shield-dimension-independent overlap bound for privacy tests with a single Gaussian-filter reduction that controls the common receiver system while preserving exchange symmetry.

    Source: §1.1 · p. 4, §1.2 · p. 4

  • Results

    The paper proves an exponential strong converse above the coherent-information threshold for finite-dimensional degradable channels, and at zero for antidegradable channels, with the quantum erasure channel covered throughout its parameter range.

    Source: §1.2 · p. 5

  • Takeaways & Limitations

    Under the joint criterion, the result rules out simultaneous reliability and secrecy above the private capacity threshold.

    Source: §1.2 · p. 5

    The finite-dimensional theorem does not settle the quantum capacity or unassisted private-capacity strong-converse questions for the pure-loss bosonic channel; a cutoff would require controlling dimension-dependent factors and truncation error.

    Source: §8.2 · p. 35

Abstract

from arXiv · show

We prove an exponential strong converse for the unassisted private capacity of every finite-dimensional degradable quantum channel. We use a single error criterion: the infidelity between the actual message-estimate-environment state and an ideal state consisting of uniformly distributed, perfectly correlated messages that are independent of the environment. At every rate strictly above the channel's maximum coherent information, the optimized fidelity to this ideal state decays exponentially in the number of channel uses. The same bound holds when the ideal environment state is fixed to the actual environment marginal. The proof applies to arbitrary mixed-state encoders and arbitrary collective decoders. Its main operator ingredient is an overlap bound for privacy tests: two receiver placements with a common subsystem of dimension $d_C$ have overlap at most $d_C/M$, where $M$ is the key size, independently of both shield dimensions. We combine this bound with the signed averaging construction presented in arXiv:2608.01308 and a single-filter reduction to symmetric states. A Gaussian filter replaces the common receiver subsystem by a system of controlled dimension from which the original correlations can be approximately recovered using quantum side information. The filter preserves exact exchange symmetry, and a uniform smooth-entropy estimate bounds its output dimension for arbitrary correlated channel inputs. We also obtain an exponential strong converse at zero for antidegradable channels under the same joint criterion and, consequently, for the private capacity of the quantum erasure channel throughout its parameter range.

1 Introduction

The introduction establishes an exponential strong converse for unassisted private communication over finite-dimensional degradable channels. It motivates a joint reliability-and-secrecy criterion and develops privacy-test overlap and single-filter tools for proving the result.

1.1 Background and motivation

  • The motivation is that private tests involve shield systems whose twisting unitaries may not commute, while equality of quantum and private capacities does not equate their finite-error coding problems.These issues motivate a shield-dimension-independent overlap bound and control of the receiver subsystem common to different placements.Source: §1.1 · p. 4

1.2 Summary of results

  • The decoded privacy-test overlap is at most dC/M when receiver placements share a subsystem of dimension dC, independently of both shield dimensions.For disjoint receiver systems, the overlap is exactly 1/M.Source: §1.2 · p. 4
  • A single Gaussian filter controls the additional receiver system’s dimension while preserving exchange symmetry and allowing recovery of the original state using quantum side information.A uniform smooth-entropy estimate bounds the output dimension for arbitrary correlated inputs, including exponentially small recovery errors.Source: §1.2 · p. 4
  • The proof gives an exponential strong converse above the coherent-information threshold for degradable channels, plus an exponential strong converse at zero for antidegradable channels and for the quantum erasure channel throughout its parameter range.The joint fidelity criterion also covers the actual-environment-marginal formulation with the same exponent.Source: §1.2 · pp. 4–5

1.3 Paper organization

  • The introduction is organized around preliminaries and the main theorem, overlap and symmetric-state bounds, the filter and its application, finite-block converses, and appendices on dilation, entropy, polynomial approximation, and the actual-marginal criterion.Section 8 concludes the paper.Source: §1.3 · p. 5

2 Preliminaries and the main theorem

This section defines the joint error criterion and coherent-information threshold, then states an exponential strong converse for degradable channels and sketches the proof reduction.

2 Preliminaries and the main theorem

  • The framework treats finite-dimensional channels through a fixed Stinespring representation, with the environment assigned to the eavesdropper.A channel is degradable when receiver-output processing reproduces its complementary channel.Source: §2 · p. 6

2.1 A single joint error criterion

  • Since the actual environment marginal is feasible in the optimized criterion, f_marg ≤ f and optimized-fidelity bounds transfer without changing the exponent.The optimized and actual-marginal fidelities need not be equal at finite blocklength.Source: §2.1 · p. 7

2.2 Why coherent information is the private-capacity threshold

  • For degradable channels, additivity of maximum coherent information gives P(N) = q_N, while the capacity formula alone addresses only vanishing error.The section distinguishes this threshold identity from the strong converse above the threshold.Source: §2.2 · p. 8

2.3 Main result

  • The private capacity has an exponential strong converse at P(N) = q_N under uniform optimized, maximal optimized, and uniform actual-marginal joint criteria.The bound is uniform over codes and independent of message states, their ranks, and local purifying-system dimensions.Source: §2.3 · p. 9
  • A single filter reduces degradable codes to symmetric states with log L ≤ nq_N plus a controlled recovery-error correction, enabling a test bound involving L/M.The symmetric-state test bound is independent of shield dimensions, and the proof makes approximation errors exponentially small.Source: §2.3 · p. 9

3 Privacy tests and their overlaps

The section converts a private code’s joint fidelity into a privacy-test acceptance probability, then bounds overlaps between alternative decoded tests. The overlap depends on the shared receiver dimension, not on shield dimensions.

3.1 Private states and coherent implementations

  • A coherent implementation retains the encoder and decoder purifications as shields and gives a privacy-test acceptance probability at least the code’s joint fidelity.The decoder’s local purification leaves the environment marginal unchanged.Source: §3.1 · pp. 9–11

3.2 Decoded test projections

  • Pulling the privacy test back through the decoder gives a projection on the enlarged input space, with the same acceptance probability as decoding before testing.The receiver-only auxiliary system G remains in the operator space so the pulled-back test is a projection.Source: §3.2 · p. 12

3.3 A shield-independent overlap bound

  • Two decoded privacy tests sharing a receiver subsystem of dimension dC have overlap at most dC/M, independently of sender and receiver shield dimensions.Here M is the key size; the shared sender shield contributes no dimension factor.Source: §3.3 · pp. 13–14
  • With disjoint receiver systems, the decoded privacy-test projections have overlap exactly 1/M, even when their shields are arbitrary and large.The argument matches the common key label, while shield operators combine into unitaries.Source: §3.3 · pp. 13, 15

4 Symmetric outputs with a common receiver system

Section 4 bounds the common acceptance probability of interchangeable privacy-test placements by combining exchange symmetry, limited overlap, and signed averaging, without requiring a uniform key marginal.

4 Symmetric outputs with a common receiver system

  • The bound is formulated for an arbitrary normalized state invariant under separate swaps of each output pair, with no condition on its key marginal or correlations involving C.The receiver placements are alternative mathematical tests; only one decoder is used in the protocol.Source: §4 · p. 16, §4.1 · pp. 17–18

4.1 Decoder placements and their common Hilbert space

  • Exchange symmetry makes every decoder placement have acceptance probability p, while distinct placements overlap by at most their shared receiver dimension divided by M.The common factor has dimension Ld^s when placements agree at s coordinates; the sender shield and dedicated auxiliary systems do not contribute.Source: §4.1 · pp. 17–18, §4.2 · p. 18

4.3 The symmetric-output bound

  • The averaged test operator obeys ∥A∥≤ζ+√ℓη, with η=min{1, Ld^t/M}; for small-fraction t, the exponentially small L/M ratio offsets d^t and row-sum growth.The bound applies to every normalized symmetric state in the stated setup, including states with a nonuniform key marginal.Source: §4.2 · p. 18, §4.3 · p. 23

4.2 Signed averaging

  • The signed matrix connects only nearly complementary placements yet acts nearly as a projection onto the uniform label vector, with zero diagonal when t<n.Signed entries reconcile sparse support with the near-projection property, so self-overlaps are excluded.Source: §4.2 · p. 21

5 Compression by a single filter

Section 5 uses one linear filter to replace part of a pure state by a controlled-dimension system, then recovers the original correlations using quantum side information.

5.1 Why a filter is sufficient

  • A filtered state whose reference marginal remains close to the original admits recovery on the retained systems by Uhlmann’s theorem.The filter need not be trace-preserving; the argument needs one normalized recoverable output, not a deterministic coding operation.Source: §5.1 · p. 23

5.3 A Gaussian estimate for the reference marginal

  • The Gaussian estimate controls the reference-marginal change without dependence on the reference dimension or its smallest nonzero eigenvalue.The averaged filter transformation satisfies E[ΓZ(X)]=Tr_A X.Source: §5.3 · pp. 25–26

5.4 Smoothing, normalization, and recovery

  • Smoothing, normalization, and Uhlmann recovery yield the stated whole-state recovery guarantee, and applying the final filter to the original pure state preserves exact exchange symmetry.The trace-norm contraction used to transfer the smoothing estimate is an average bound; an individual Gaussian filter need not contract trace norm.Source: §5.4 · pp. 26–27

5.2 The one-shot statement

  • The one-shot construction bounds log|C| by the positive part of h+4 log(1/δ)+4, plus 1, where h is the relevant conditional smooth max-entropy.The filter output is normalized before recovery; no lower bound on the probability of implementing it is asserted.Source: §5.2 · p. 24, §5.4 · p. 27

6 From a degradable code to symmetric outputs

The section turns a degradable code into a symmetric state with a controlled-size common receiver system, then uses recovery to relate its privacy-test probability to the original code. The construction preserves exact exchange symmetry and works uniformly for correlated channel inputs.

6.1 The symmetric factorization

  • A symmetric factorization of a degradable channel exposes two interchangeable outputs while retaining the same isometric output for every input, including inputs entangled with a reference.An independent maximally mixed receiver qubit may be appended if needed.Source: §6.1 · p. 28, Figure 6 · p. 28
  • Every original private code has a corresponding code with exactly the same optimized joint fidelity and actual-marginal joint fidelity.This lets the symmetric reduction preserve the two joint error criteria.Source: §6.1 · p. 28

6.2 Filtering while preserving the exact symmetry

  • One fixed filter on F^n produces a controlled-dimension system C while preserving every exchange symmetry exactly.Recovery uses C and E′^n to reconstruct F^nE′^n with the other systems serving as quantum side information.Source: §6.2 · pp. 29–30

6.3 A uniform dimension estimate for correlated inputs

  • The filter dimension bound holds uniformly for every coherent encoder, including arbitrarily correlated inputs, although the filter may depend on the code and δ.The stated bound includes an explicit 8 log(1/δ) term.Source: §6.3 · pp. 30–31

6.4 Returning to the original privacy test

  • Composing recovery with the original decoder compares the original privacy test directly with the filtered symmetric state, without treating filtering as protocol postselection.The recovery estimate does not multiply or divide by the probability of realizing the filter.Source: §6.2 · p. 30, §6.4 · pp. 31–32

7 The finite-block converse and exponential decay

The finite-block converse yields exponential decay of joint fidelity above the coherent-information threshold for degradable channels. The same approach gives a zero-threshold converse for antidegradable channels, covering the quantum erasure channel across its parameter range.

7 The finite-block converse and exponential decay

  • Above the maximum coherent-information threshold, the optimized joint fidelity decays exponentially, with the same conclusion for maximal joint infidelity and the actual-environment-marginal criterion.The finite-block bound combines recovery error, approximation of the uniform averaging matrix, and common-receiver overlap.Source: §7 · pp. 32–33

7.1 Antidegradable channels

  • For every finite-dimensional antidegradable channel, joint fidelity decays exponentially at every positive rate, giving an exponential strong converse at zero.The stated criteria are joint infidelities.Source: §7.1 · p. 33

7.2 The quantum erasure channel

  • For every erasure probability, the unassisted private capacity has an exponential strong converse under the joint criteria.The corollary covers all p in [0, 1] for finite input dimension.Source: §7.2 · p. 34

8 Conclusion

The conclusion summarizes an exponential strong converse for unassisted private communication over finite-dimensional degradable channels under joint infidelity criteria. It identifies open questions about exponent tightness, bosonic channels, broader channel classes, and other assistance models.

8.1 Summary

  • For finite-dimensional degradable channels, joint fidelity decays exponentially above maximum coherent information, uniformly over correlated inputs and collective decoders.The conclusion covers both an optimized ideal environment state and one fixed to the actual marginal.Source: §8.1 · p. 34
  • The proof combines privacy-test overlap bounds with a recoverable single-filter reduction that preserves exact exchange symmetry and controls dimension for correlated inputs.The signed Boolean construction is attributed to Ref. [KBK+26], based on Ref. [dW08].Source: §8.1 · p. 35

8.2 Future directions

  • The exponent in the present converse is not asserted to be optimal.Sharper averaging and filter estimates, and a variational characterization, remain questions.Source: §8.2 · p. 35
  • The finite-dimensional theorem does not settle bosonic private-capacity converses, since dimension-dependent bounds and truncation errors must be controlled for allowed inputs.The discussion raises pure-loss channels and photon-number constraints as settings for further work.Source: §8.2 · p. 35
  • Extensions beyond degradability or to other assistance models require reductions and bounds suited to the channels and resources allowed.The proof does not establish a converse for arbitrary feedback-assisted communication.Source: §8.2 · pp. 35–36

Statement on AI-assisted preparation

The statement describes extensive AI assistance in developing and preparing the paper, alongside the author’s direction and responsibility. The appendix constructs a symmetric dilation of a degradable channel through an involutive flag and an added receiver qubit.

Statement on AI-assisted preparation

  • AI tools assisted with mathematical exploration, proof checking, drafting, revision, bibliographic verification, and preparation of code and figures.The author proposed the research problem, directed revisions, and supplied detailed comments.Source: Statement on AI-assisted preparation · p. 36
  • The paper includes self-review but not independent external peer review or formal verification, and the author accepts responsibility for its contents.Numerical calculations checked finite-dimensional identities and estimates rather than replacing proofs.Source: Statement on AI-assisted preparation · p. 36

A.1 An involutive dilation

  • A coherent flag makes exchanging the two complementary outputs equivalent to an involution on the retained auxiliary system.The flag construction leaves both complementary marginals unchanged.Source: §A.1 · p. 39

A.2 Removing the involution

  • An added receiver qubit correlates the involution sectors with symmetric or antisymmetric Bell states, canceling their exchange signs and yielding an exactly symmetric output pair.The receiver qubit is maximally mixed and is purified by an environment qubit.Source: §A.2 · p. 41

A.3 Preservation of coherent information and the joint criterion

The auxiliary-qubit construction preserves the coherent-information threshold and the joint fidelity criterion exactly. The appendix also establishes uniform smooth-entropy bounds for correlated channel inputs and identifies the approximation input used in the converse.

A.3 Preservation of coherent information and the joint criterion

  • Appending the auxiliary qubits adds one bit to both receiver and environment entropies, leaves maximized coherent information unchanged, and preserves the optimized joint fidelity exactly.The decoder can ignore the added receiver subsystem, so the factorization applies to the original coding problem without an additional approximation.Source: §A.3 · p. 42

B.1 Conventions and elementary inputs

  • The appendix fixes smoothing conventions and retains explicit smoothing changes because it does not assert concavity of smoothed max-entropy.It also notes that normalized and subnormalized states matter in the entropy comparisons.Source: §B.1 · pp. 42–43

B.2 Postselection for arbitrary channel inputs

  • For arbitrary channel inputs, permutation averaging, postselection domination, and smoothing comparisons yield the correlated-input entropy bound with only logarithmic dimension terms.The choices and bounds are independent of the input state; no product-state or full-rank restriction is imposed.Source: §B.2 · pp. 43–45

B.3 A dimension-uniform independent-copy estimate

  • The independent-copy smooth-entropy estimate is dimension-uniform and holds for every blocklength, including when the smoothing parameter depends exponentially on blocklength.This uniformity avoids state-dependent thresholds and constants involving the smallest nonzero eigenvalue.Source: §B.3 · p. 45

C The polynomial approximation input

  • The appendix specializes an external approximate-degree theorem to NOR_n and clarifies that neither a new approximation theorem nor a new signed-weight construction is claimed.The required approximation has small-error dependence, while its degree restriction supplies the support condition used in the converse.Source: C The polynomial approximation input · p. 46

C.1 Specializing the approximate-degree theorem

The approximate-degree input for NOR_n supplies both small-error approximation and a degree restriction used in the converse. The Hadamard representation separates spectral approximation from coordinate support, while the actual-marginal target preserves the converse bound.

C.1 Specializing the approximate-degree theorem

  • For NOR_n, the specialized approximate-degree bound provides a polynomial with exponentially small error at linear t, while its degree restriction enforces support s(x, y) ≤ t.A constant-error approximation alone would not give an exponentially small operator-norm difference.Source: §C.1 · p. 46

C.2 The Hadamard transform and the two bases

  • The Hadamard representation makes Wt close to U0 through small nonuniform eigenvalues, while the low degree of q separately restricts coordinate support.The spectral approximation and entrywise support statements are distinct and neither substitutes for the other.Source: §C.2 · p. 46

D Using the actual environment marginal

  • Fixing the ideal environment state to the actual marginal preserves the finite-block converse bound and its exponential rate without loss.The distinction from an optimized environment state is discussed in Ref. [WTB17, Appendix B].Source: D Using the actual environment marginal · p. 47

D.1 Comparison with the optimized criterion

The actual-marginal fidelity is no larger than the optimized fidelity, so every finite-block upper bound transfers to it. Consequently, the exponential strong converse and vanishing-error capacity conclusion hold for the actual-marginal criterion as well.

D.1 Comparison with the optimized criterion

  • The actual-marginal fidelity is bounded above by the optimized fidelity, so every finite-block upper bound and the vanishing-error capacity transfer.The two criteria need not have equal finite-error values; near zero, ε_marg ≤ 4ε.Source: §D.1 · p. 47
  • The exponential strong converse remains valid with the actual environment marginal as the ideal state, with the same constants and exponent.For rates log M ≥ n(q_N + ∆), the joint infidelity is at least 1 − 2^-γn for n ≥ n0.Source: §D.1 · p. 47

D.2 The two finite-error fidelities can be different

A degradable-channel code example shows that the optimized and actual-marginal criteria can differ at finite error. The example deliberately introduces decoding errors to distinguish the criteria, rather than to optimize communication.

D.2 The two finite-error fidelities can be different

  • Only correctly decoded blocks overlap the ideal key state, which is why the example distinguishes the two environment-state choices.The channel transmits the encoded basis state and appends a flag correlated with the environment.Source: §D.2 · p. 48
  • The two finite-error fidelities can differ: in the example, optimizing the environment state attains p, whereas using the actual environment marginal gives a different value.The example is realized by an unassisted code for a degradable channel whose decoder deliberately errs on one flag.Source: §D.2 · p. 48

D.3 Consequence for separate decoding and secrecy errors

The two-error formulation yields a lower bound on the sum of decoding and secrecy errors, but does not imply that either error separately converges to one. The proof specifies the environment state and fidelity convention for this comparison.

D.3 Consequence for separate decoding and secrecy errors

  • The comparison is stated in the paper’s notation to specify the environment state and fidelity convention.The proof says that combining the established bound with Corollary D.2 gives the stated consequence.Source: §D.3 · pp. 48–49
  • The proof replaces the estimate with a correct copy of K while preserving the KEn marginal, then uses blockwise trace norms and an isometry to compare the states.The argument proceeds via the trace-norm inequality and the other Fuchs–van de Graaf inequality.Source: §D.3 · p. 49
  • The lower bound concerns the sum of the two errors; it does not require either error separately to converge to one or equate separate-error definitions with either joint criterion at fixed nonzero error.The comparison is identified with Proposition 28 of Ref. [WTB17].Source: §D.3 · pp. 48–49
Loading 2609.32515v1…