Quantum Shannon theory made robust: a tale of three protocols for almost i.i.d. sources

Abstract: The asymptotic rates of information-theoretic protocols - including error exponents, compression rates, and channel capacities - are traditionally defined under the idealised assumption that the underlying resource (state or channel) is independent and identically distributed (i.i.d.). Somewhat surprisingly, even slight departures from the exact i.i.d. structure can lead to a drastic breakdown of these protocols. The asymptotic rates of information theoretic protocols - error exponents, compression rates, capacities - were originally evaluated taking for granted that the underlying source (state or channel) is i.i.d. Differently from what we might expect at first glance, it is not hard to exhibit instances of protocols that may drastically fail when the i.i.d. assumption holds only approximately rather than exactly. If the precise nature of the perturbation from the i.i.d. regime is known (e.g. a pointwise defect), we could design a bespoke protocol that compensates for the defect (for example, by discarding the corrupted subsystem). However, in any realistic setting, neither can the i.i.d. behaviour of the system be precisely guaranteed, nor can the deviations from the ideal regime be determined exactly. In this paper we answer the following question: are there protocols that can still achieve the optimal asymptotic rates when the i.i.d. resource is replaced by any arbitrary almost i.i.d. resource along it? What is the nature of the unknown perturbation under which protocols like these are possible? We focus, in particular, on hypothesis testing, data compression, and channel coding. As a by-product of our analysis, we introduce the notion of club distance, as a variant of the well-known diamond distance, and of an almost i.i.d. process, which may be of independent interest.

• The paper demonstrates that optimal rates in hypothesis testing, data compression, and channel coding remain intact for almost i.i.d. quantum sources using universal protocols.
• It rigorously establishes a hierarchy of almost i.i.d. properties, distinguishing weakly almost i.i.d., Wasserstein almost i.i.d., and MSR almost i.i.d. states and channels.
• The study introduces the club distance metric to capture locality-aware deviations, ensuring robust performance despite sparse local defects in quantum systems.

Robust Quantum Shannon Theory: Stability of Information-Theoretic Protocols under Almost i.i.d. Resources

Introduction and Motivation

This work addresses the fundamental question of robustness in quantum information theory: when canonical protocols in hypothesis testing, data compression, and channel coding—originally constructed for perfectly independent and identically distributed (i.i.d.) resources—remain stable under physically realistic deviations from the i.i.d. assumption. The classical i.i.d. approximation is rarely satisfied exactly in experimental or physical settings, where weak correlations, sporadic defects, or time-dependent non-stationarities generically arise. The paper provides a systematic and operational approach for identifying classes of "almost i.i.d." quantum states and channels under which the optimal asymptotic rates (error exponents, compression rates, and capacities) of information-theoretic protocols remain unchanged. It also introduces the club distance as a physically meaningful metric for channel proximity tailored to multipartite, locally varying quantum resources.

Notions of Almost i.i.d. Structure

The paper establishes a rigorous hierarchy of almost i.i.d. notions for quantum states (Figure 1), clarifying the relation and operational implications among several previously developed concepts:

Figure 1: Hierarchy of almost i.i.d. quantum state classes showing inclusions—from weakly almost i.i.d. through Wasserstein almost i.i.d. to MSR almost i.i.d.—as well as relationships to trace distance almost i.i.d. sources.

Three principal definitions are analyzed:

• Weakly almost i.i.d.: Only requires $k$-body marginals ($k$ fixed, $n\to\infty$) to converge (in average trace norm) to $\rho^{\otimes k}$ for reference state $\rho$. This captures local indistinguishability but not global properties.
• Wasserstein almost i.i.d.: $W_1$-distance (quantum optimal transport with Hamming local structure) between $\rho_n$ and $\rho^{\otimes n}$ vanishes per subsystem as $n\to\infty$. This ensures stability under local perturbations and ensures global continuity of the von Neumann entropy.
• Mazzola--Sutter--Renner (MSR) almost i.i.d.: Enforces structure akin to the quantum de Finetti theorem, guaranteeing not just local marginals but also that $\rho_n$ can be written (up to few subsytems) as a symmetrized mixture of product states.

In addition, weaker trace distance convergence is discussed and shown to be insufficient for operational robustness, as local defects may yield orthogonal states, failing to guarantee entropy stability or operational performance in global tasks. The analysis directly connects the operational stability of protocols to these structural properties, formalized further in subsequent sections.

Quantum Hypothesis Testing Robustness

Quantum Stein's lemma underpins the rate of asymmetric quantum hypothesis testing in the i.i.d. setting: the optimal asymptotic type-II error exponent equals the Umegaki relative entropy $k$0. This work demonstrates:

Figure 2: Schematic of hypothesis testing when one or both hypotheses are replaced by (potentially unknown) almost i.i.d. states, highlighting robustness and universality of protocol design.

Main results:

• If the null hypothesis is weakly almost i.i.d. along $k$1 and the alternative is MSR almost i.i.d. along $k$2, then the Stein exponent remains at least $k$3—no strict decrease is possible for universal testing strategies.
• Universal protocols exist: a single test sequence (not tailored to the explicit defect instance) will achieve Stein rate $k$4 for all almost i.i.d. sources in the appropriate class.
• Strict inequality can appear: there are pathological (non-symmetric, local-defect) sources for which the attainable Stein exponent exceeds $k$5, but not below. Conversely, replacing the alternative hypothesis by merely trace distance or Wasserstein almost i.i.d. can fail dramatically (e.g., yield an exponent zero), confirming the necessity of the MSR condition for stability.
• The results extend the operational robustness from classical to quantum hypothesis testing using symmetrization and measurement techniques, align with the generalized (quantum) Sanov theorem, and confirm the need for care in robustness statements.

Robust Classical and Quantum Data Compression

In the data compression setting, the paper generalizes Shannon’s and Schumacher’s compression theorems by showing both classical and quantum optimal compression rates are robust to weakly almost i.i.d. sources:

Figure 3: Robustness of classical data compression: a code for i.i.d.\ sources retains vanishing error probability for any weakly almost i.i.d.\ source.

Figure 4: Robustness of quantum data compression: a code attaining high entanglement fidelity for i.i.d.\ sources works for any weakly almost i.i.d.\ source, even in the presence of an external reference.

Results:

• For any weakly almost i.i.d. source along $k$6 (classical) or $k$7 (quantum), universal codes achieve compression rates $k$8 (Shannon entropy) and $k$9 (von Neumann entropy), respectively, with vanishing error probability (classical) or asymptotic entanglement fidelity approaching 1 (quantum).
• This robustness is achieved via codes constructed from robust hypothesis tests—hypothesis testing serves as a unifying tool for compression.
• The MSR class is not required; weakly almost i.i.d. suffices because the critical global property is local compression per marginal.

Channel Coding: Capacity Stability under Almost i.i.d. Processes

The work generalizes the coding paradigm from i.i.d. product channels to "almost i.i.d." channels, formalized via the club (♣) distance utilizing quantum $n\to\infty$0.

Figure 5: Multiple uses of a channel modeled with general almost i.i.d. process, in contrast to the traditional i.i.d. tensor product setting.

Figure 6: Illustration of communication over an almost i.i.d. quantum channel, leveraging robust, universal (shared-randomized) codes to achieve the i.i.d. channel capacity.

Main theorem:

• If a sequence of channels $n\to\infty$1 is almost i.i.d. along a reference channel $n\to\infty$2 in club distance (vanishing per subsystem as $n\to\infty$3), then the unassisted classical capacity equals the i.i.d. channel capacity: $n\to\infty$4.
• Universal coding is possible: there exist codes, depending only on $n\to\infty$5, not on the specific perturbation, attaining this capacity for all almost i.i.d. processes.
• The result extends to settings with shared randomness, enabling protocol universality even if the noise model is only known up to the general almost i.i.d. specification.

Figure 7: Coupling interpretation for almost i.i.d. channels: an i.i.d. channel followed by a controlled noise process that spoils at most $n\to\infty$6 output symbols per input.

Crucially, while capacity is robust, the paper proves that the reliability function (error exponent rate) is not universally robust for generic almost i.i.d. channels. Specific classes of perturbations can yield non-exponential error decay, even if capacity remains maximal.

Quantum Wasserstein $n\to\infty$7 and Club Distance

The quantum $n\to\infty$8 metric is central to the robustness properties proven. The metric is shown to interpolate between trace distance and global perturbations by being sensitive to the locality structure, permitting strong continuity bounds on entropy per particle and on channel output distinguishability. The club distance on quantum channels leverages quantum $n\to\infty$9, capturing the relevant physical notion of "closeness" for quantum processes subject to sparse local defects.

Implications and Future Directions

The results establish that only sufficiently stringent or physically meaningful notions of almost i.i.d. structure (Wasserstein or MSR for channels, weakly almost i.i.d. for sources) guarantee operational robustness of key information-theoretic protocols. Trace distance convergence, while stricter, fails to capture operationally relevant global properties in the presence of sparse but strong local defects. Conversely, local-marginal convergence is too weak for entropic and capacity-related tasks.

The strict separation between reliability function and capacity robustness underlines the fundamental difference between rate and exponent in non-i.i.d. information theory. The paper’s universal protocol constructions—with explicit use of symmetrization and measurement reduction—enable practical coding in settings where the details of the non-i.i.d. perturbation are unknown.

Prospective extensions include one-shot generalizations, further analysis of club/Wasserstein distances for quantum channels, and applications to quantum many-body learning [Rouzé, Quantum 8, 1319 (2024); De Palma, J. Math. Phys. 65, 5.0178897 (2024)]. The rapid advances in quantum optimal transport theory signal the possibility of deeper connections between locality-aware quantum distances and operational capacities across broader classes of protocols.

Conclusion

This work provides a comprehensive operational foundation for robust quantum Shannon theory in the presence of realistic, weakly correlated perturbations away from idealized i.i.d. assumptions. By identifying the necessary and sufficient structures for the preservation of optimal rates in hypothesis testing, data compression, and channel coding, and by introducing metrics tailored to physically relevant notions of closeness, it establishes a rigorous framework for universal, robust information processing protocols in quantum information theory.

Reference:

"Quantum Shannon theory made robust: a tale of three protocols for almost i.i.d. sources" (2605.18726)