Source-linked AI summary
The exact strong converse exponent for private communication over quantum channels
Hao-Chung Cheng, Christoph Hirche, Marco Tomamichel
TL;DR
The paper asks how secret-key fidelity decays above private capacity, and develops a Rényi private information whose regularization characterizes the exact exponent. Its results establish exponential decay above capacity and show that separate reliability and secrecy criteria can have a different exponent.
Problem
Capacity characterizes rates with asymptotically perfect transmission, but does not determine optimal fidelity above capacity or its decay exponent.
Source: §1 · pp. 1–2Method
The paper introduces a Rényi private information and uses its tilted-state integral representation and continuity under regularization, alongside converse and matching achievability bounds.
Source: §1 · p. 2, §2.1 · p. 4Results
The regularized Rényi private information characterizes the exact strong converse exponent for squared joint-key fidelity over finite-dimensional quantum wiretap channels.
Source: §2.2 · p. 6, §8 · p. 34Takeaways & Limitations
The results establish exponential fidelity decay at every rate above private capacity, while the classical specialization has a variational form with a relative-entropy penalty.
Source: §8 · p. 34The exact decay rate under separate reliability and secrecy requirements need not coincide with the squared joint-key-fidelity exponent.
Source: §2.3 · p. 7
Abstract
from arXiv · showhide
We determine the exact strong converse exponent for secret-key transmission and generation over general finite-dimensional quantum wiretap channels, measuring reliability and secrecy jointly by squared fidelity to an ideal secret key. We introduce a novel Renyi private information whose regularization characterizes this exponent. An exact integral representation in terms of ordinary private information gives a uniform continuity bound, showing that the regularized quantity converges to private capacity as the Renyi order approaches one. This establishes for any wiretap channel an exponential fidelity decay at every rate above private capacity.
1 Introduction
Private communication sends classical information reliably to Bob while keeping it secret from Eve; the paper studies how fidelity decays when the transmission rate exceeds private capacity. It introduces a Rényi private information and uses continuity and matching bounds to determine the exact strong converse exponent.
1 Introduction
- Private communication requires reliable classical transmission to Bob while keeping the message secret from Eve, assessed here by squared fidelity to an ideal key.In the ideal key, Alice’s message and Bob’s estimate agree and are independent of Eve.Source: §1 · p. 1
- A strong converse requires optimal fidelity to vanish at every rate above private capacity; an exponential strong converse specifies its decay rate.Source: §1 · pp. 1–2
- The exact exponent requires a converse valid for every code and a sequence of codes attaining the same exponent.Source: §1 · p. 2
- Existing general converse bounds can guarantee fidelity decay only above rates that exceed private capacity, while prior exponential results cover specified channel classes or criteria.Source: §1 · p. 2
- The paper introduces a Rényi private information whose regularization characterizes the exponent, with an integral representation yielding continuity at order one.Source: §1 · p. 2
- A one-shot converse and a matching achievability strategy establish the exponent; achievability uses a channel change followed by local pinching and double blocking.Source: §1 · pp. 2–3
2 Main results and discussion
The main results define a Rényi private information and use its continuity at order one to characterize the exact strong converse exponent. The results also cover secret-key generation and distinguish joint-fidelity exponents from those under separate error criteria.
2.1 Rényi private information
- The Rényi private information is additive for product cqq states, but channel optimization over correlated input ensembles means regularization is still needed.It also obeys data processing in opposite directions: processing Bob’s system cannot increase it, while processing Eve’s system cannot decrease it.Source: §2.1 · p. 3
- Its integral representation averages ordinary private information along tilted states, enabling continuity at order one after regularization.For classical channels, the variational form maximizes private information under a changed joint law, penalizing departures from the original law by relative entropy.Source: §2.1 · pp. 4–5
2.2 Strong converse exponent
- Continuity at order one lets the authors choose a Rényi order with information below any rate above private capacity, yielding exponential fidelity decay.Source: §2.2 · p. 6
- The exact strong converse exponent is characterized by the regularized Rényi private information, with the operational liminf established as a limit.The characterization combines a regularized converse optimized over Rényi order with a matching achievability bound.Source: §2.2 · p. 6
- Achievability changes the channel, uses universal reference states and local pinching, transfers the construction back, then blocks arbitrary input ensembles to obtain regularization.Source: §2.2 · p. 6
2.3 Secret-key generation and other error criteria
- Secret-key generation has the same exponent as transmission: a generation code converts to a transmission code of equal blocklength and key size with at most a factor-four squared-fidelity loss.Source: §2.3 · p. 7
- The exact decay rate under separate reliability and secrecy requirements need not match the squared joint-key-fidelity exponent.Source: §2.3 · p. 7
3 Rényi private information and its properties
The section develops Rényi private information through equivalent norm and dual quasi-norm representations, then establishes how it behaves under local processing at Bob’s and Eve’s terminals.
3.1 Norm and dual quasi-norm representations
- Equivalent norm and dual quasi-norm representations exchange the outer optimizations after the conditional auxiliary states are optimized.The resulting objective is convex in Bob’s reference and concave in Eve’s reference.Source: §3.1 · pp. 7–8
3.2 Local data processing
- Processing Bob’s system cannot increase the Rényi private information, while processing Eve’s system cannot decrease it.These opposite directions are the local data-processing inequalities for the two terminals.Source: §3.2 · p. 9
3.3 The order-one limit for a fixed ensemble
For a fixed ensemble, the order-one limit recovers ordinary private information, with uniform control supporting continuity. At the order-infinity endpoint, the quantity is related to the ideal-key fidelity.
3.3 The order-one limit for a fixed ensemble
- The optimized leading term differs from I(X : B)ω − I(X : E)ω by at most −2 log c, yielding the ordinary private-information limit as α approaches one and ε vanishes.The expansion has a uniform O(α − 1) remainder, including for singular signal marginals.Source: §3.3 · p. 10
- For fixed α > 1, Pα is continuous on cqq states and unchanged when B and E are restricted to their average-marginal supports.The support reduction and continuity result is stated in Corollary 3.3.Source: §3.3 · p. 10
3.4 The order-infinity limit
- The endpoint limit follows by restricting the reference states to compact sets, taking the limit there, and then removing the restriction.The passage states Qu,ε → Q1,ε and then Qu → Q1.Source: §3.4 · p. 10
- At the order-infinity endpoint, Q1 = dFkey(ω), relating the endpoint quantity to the ideal-key fidelity.Only blocks with matching classical labels contribute to the fidelity identity.Source: §3.4 · p. 12
3.5 Fixed-point identities
The fixed-point identities connect optimized states to the conditional and average marginals of a tilted state. The optimizing states also have uniqueness, faithfulness, and regularity properties under the stated conditions.
3.5 Fixed-point identities
- The optimizers in Definition 2.1 are the conditional and average marginals of a tilted cqq state constructed from the original state.The tilted state is formed by applying local operators and normalizing, under the stated faithfulness assumptions.Source: §3.5 · p. 12
The minimizing conditional states on Eve’s system satisfy
- The fixed-point identities identify the optimized conditional and average marginals that enter the tilted-state representation.The paper relates these identities to fixed-point identities for optimized Rényi information in public classical and quantum communication [16, Lemmas 6 and 9].Source: §3.5 · p. 12, The optimizing reference states satisfy · p. 14
The optimizing reference states satisfy
- For each fixed Bob reference, the maximizing Eve reference and auxiliary states are unique and faithful, and the maximized objective is continuously differentiable in u and τB.The optimizing states and related quantities may depend on u and the ensemble.Source: The optimizing reference states satisfy · p. 13
3.6 Additivity for fixed ensembles
The Rényi private information is additive for fixed ensembles under tensor products. This does not imply additivity of the channel quantity, which still optimizes over correlated block inputs.
3.6 Additivity for fixed ensembles
- The Rényi private information is additive under tensor products for any two finite cqq ensembles.The fixed-ensemble identity is established in Lemma 3.5.Source: §3.6 · p. 14
- This fixed-ensemble identity does not imply channel additivity because the channel quantity optimizes over correlated block inputs.The channel optimization allows ensembles correlated across uses.Source: §3.6 · p. 15
3.7 The tilted state and integral representation
The section constructs a tilted ensemble from the optimizing states and differentiates its optimized normalization. Integrating the resulting relation yields the stated integral representation.
3.7 The tilted state and integral representation
- The tilted ensemble has letter probabilities r_x(u) proportional to p_xℓ_x(u), with normalized conditional states Θ_BE,x.Its normalization is the optimized sum over p_xℓ_x(u).Source: §3.7 · p. 15
- The conditional stationarity identities and averaging establish the stated relations for the tilted ensemble.The derivation uses the identities for the optimizing states and averages over the tilted letter probabilities.Source: §3.7 · pp. 15–16
- Differentiating the optimized sum gives the derivative relation for almost every u, without requiring a differentiable choice of optimizers.The proof holds the optimizing families fixed during differentiation.Source: §3.7 · p. 16
- Integrating the derivative relation and taking ε down to zero proves the stated integral representation.The argument uses the bound on tilted-state private information and the limiting relation as u approaches zero.Source: §3.7 · p. 16
4 Continuity of the regularized Rényi private information
The section proves continuity at order one by uniformly comparing the regularized Rényi private information with ordinary private information. It bounds the cost of restoring tilted states to the channel image, yielding convergence as the Rényi order approaches one.
4 Continuity of the regularized Rényi private information
- The regularized Rényi private information differs from private capacity by a dimension-dependent bound that vanishes as α approaches one.The theorem applies for 1 < α ≤ 2, with Γ_dB,dE(α) → 0 as α ↘ 1.Source: §4 · p. 16
4.1 Local estimates for reversing the tilt
- Local cancellation bounds Eve’s correction independently of the untouched dimension, ensemble size, and spectra.For d_C = 1, the bound is zero.Source: §4.1 · p. 17
4.2 Projecting the tilt back to the channel image
- The support test rejects the tilted state with probability O_dB,dE(u^2), uniformly in blocklength and ensemble.This controls how far the tilted state lies outside the single-use Stinespring output subspace.Source: §4.1 · p. 16, §4.2 · p. 20
4.3 Information cost of restoring the channel output
- Restoring the channel output incurs a per-use information cost that vanishes with u, allowing the regularized comparison to pass to the channel image.Sequential restoration preserves the untouched marginal, and the final restored state lies in the image of V^⊗n.Source: §4.3 · pp. 20–21
5 Converse bounds
The paper derives one-shot fidelity and trace-distance converses in terms of the Rényi private information. Combined with continuity at order one, these bounds establish exponential decay above private capacity, while the exponent is zero below capacity.
5 Converse bounds
- The one-shot fidelity converse bounds a secret-key transmission code’s fidelity using its Rényi private information.The proposition applies to every one-use code and every α > 1.Source: §5 · pp. 21–22
- The one-shot trace-distance converse relates the reliability–secrecy gap to the joint trace-distance error.For every one-use code, [p_succ − δ_T]_+ is bounded by one minus the minimum trace distance to an ideal key state and an Eve reference state.Source: §5 · p. 23
5.1 Exponential strong converse
- For every rate R > P(W), the strong converse exponent is positive for finite-dimensional channels.The one-shot converses and continuity at order one yield exponential fidelity decay above private capacity.Source: §5.1 · p. 24
- For R < P(W), the strong converse exponent is zero; the paper also establishes zero at capacity.The argument appends an independent label to a lower-rate code, dividing fidelity by the label size.Source: §5.1 · p. 24
6 Achievability of the strong converse exponent
This section establishes achievability of the strong converse exponent using variational coding, local pinching, transfer to the original channel, and blocking or block regularization. It also shows that unassisted secret-key generation has the same exact exponent as transmission.
6 Achievability of the strong converse exponent
- The operational exponent is a limit, convex, nondecreasing, and 1-Lipschitz; blocking arbitrary input blocks yields the exponent in terms of Pα(W).Source: §6 · pp. 24–25
6.1 Achievability of the variational exponent
- Ordinary private coding on a modified channel, followed by a relative-entropy performance penalty and optimization, achieves the variational exponent.Source: §6.1 · pp. 25–26
6.2 Comparison by local pinching
- Local pinching makes the trace and variational formulations asymptotically comparable, with their difference vanishing per channel use and a uniform weighted estimate permitting optimization over α.Source: §6.2 · p. 27, §6.2.2 · p. 29
6.2.1 Universal reference states
- Faithful permutation-invariant universal reference states dominate every permutation-invariant state up to polynomial factors, giving uniform errors across ensembles and α.Source: §6.2.1 · pp. 27–28
6.2.2 Comparison after controlled local pinching
- Controlled local pinches bound the remaining comparison costs by logarithms of polynomial projection counts, so these costs vanish after division by block length.Source: §6.2.2 · pp. 28–29
6.3 Transferring the coding construction
- Invariant acceptance tests remove word-dependent center pinches, while input-independent output pinches incur a controlled fidelity cost, transferring the coding construction to the original channel.Source: §6.3 · p. 29
- Taking limits in the transferred coding construction establishes the achievability bound, using minimax optimization and a vanishing polynomial correction.The argument first passes to large code blocks and then takes the pinching and rate limits.Source: §6.3 · p. 31
6.4 Regularization by blocking
- Theorem 6.9 establishes an achievability bound for every finite-dimensional channel by optimizing over input ensembles and block lengths.The proof applies the bound to channel blocks, optimizes over ensembles and block sizes, and uses convex duality.Source: §6.4 · pp. 31–32
6.5 Secret-key generation has the same exponent
- An unassisted generation code with squared fidelity f yields a transmission code with fidelity at least f/4, so their optimized exponents are equal.The conversion duplicates weighted labels and retains the M labels with the largest individual root-fidelity contributions.Source: §6.5 · pp. 32–33
7 Classical channels
For classical joint channels, the Rényi private information admits a variational form. A diagonal reduction turns its optimization into one over distributions, and the resulting channel formula applies at every blocklength, including stochastic encodings.
7 Classical channels
- A classical joint channel measures the input in a fixed basis and prepares diagonal outputs with law W(·, ·|a).For such channels, the Rényi private information has the variational form stated in Section 2.Source: §7 · p. 33
- The Rényi private-information optimization reduces to distributions through dephasing and data processing on Bob’s and Eve’s outputs.The argument makes the minimizing references and auxiliary optimizers diagonal.Source: §7 · p. 33
- The variational proof orders the reference optimizations, optimizes auxiliary distributions, and uses a minimax exchange to obtain the stated representation.The argument invokes Sion’s theorem and faithful approximation for singular distributions.Source: §7 · pp. 33–34
- The classical channel formula applies at every blocklength, so the general exponent theorem covers classical channels, including stochastic encodings.The proof establishes the reverse inequality in the channel formula using the variational representation.Source: §7 · p. 34
8 Conclusions
The paper determines the exact strong converse exponent for squared joint-key fidelity over finite-dimensional quantum wiretap channels using regularized Rényi private information. Its results establish exponential fidelity decay above private capacity, while identifying additivity and other error criteria as open questions.
8 Conclusions
- The exact strong converse exponent for squared joint-key fidelity is characterized by regularized Rényi private information for finite-dimensional quantum wiretap channels.A matching change-of-channel and pinching argument, together with a tilted-state integral, proves exponential fidelity decay above private capacity.Source: §8 · p. 34
- For classical channels, Rényi private information reduces to a variational expression involving ordinary private information and a relative-entropy penalty.Source: §8 · p. 34
- Joint-key fidelity and marginal error criteria can have different exponents and therefore must be treated separately.Source: §8 · p. 34
- Identifying quantum wiretap channel classes with additive optimized Rényi private information remains open; additivity would yield a single-letter exponent.The exact exponent of [psucc −δT ]+ and its relation to the squared-fidelity exponent are also open questions.Source: §8 · p. 34
A On other error criteria
The appendix compares squared joint-key fidelity with other error criteria and gives examples where their strong converse exponents differ. It also distinguishes the joint-channel setting from independently specified quantum marginals, which need not be compatible.
A On other error criteria
- The exact formula concerns squared joint-key fidelity for a specified joint channel; independently specified quantum marginals need not be compatible.Every pair of classical channels admits a joint implementation.Source: A On other error criteria · p. 35
- The criteria detect the same vanishing-error regime, but the bounds do not identify their strong converse exponents, which can differ.Source: A On other error criteria · p. 35
- The largest reliability–secrecy gap [psucc −δT ]+ is 1/M when Bob’s marginal channel is a postprocessing of Eve’s, giving exponent R at rate R.Source: A On other error criteria · p. 35
- 5/8 is the fidelity of the erasure-channel example’s positive decoding block against the stated common Eve reference.The example uses complementary outputs of a binary erasure channel with erasure probability 1/2.Source: A On other error criteria · p. 35