Bipartite Bound Information Exists

Technical Summary: "Bipartite Bound Information Exists"

Problem Statement
The paper addresses a fundamental question in classical cryptography posed by Gisin and Wolf twenty-five years ago: Does "bound information" exist? Analogous to "bound entanglement" in quantum theory—states that require entanglement to create but from which no pure entanglement (singlets) can be distilled—bound information refers to a classical probability distribution shared by two honest parties (Alice and Bob) and an eavesdropper (Eve) that requires secret bits to create (positive formation cost, $I_{\text{form}} > 0$) yet yields no extractable secret key (zero secret-key rate, $S = 0$). While bound entanglement is a known quantum phenomenon, the existence of its classical counterpart remained an open problem, with previous attempts only showing sources where the formation cost exceeded the key rate by a large factor, but never exactly zero.

Methodology
The authors employ a novel proof mechanism based on a gap between two distinct ways of comparing an eavesdropper's information:

1. Simulation: Can Eve process her data $Z$ to simulate another variable $J$?
2. Less-Noise Dominance: Is Eve's channel "less noisy" than another channel $J$? This is defined by the condition $I(U; Z) \geq I(U; J)$ for all auxiliary variables $U$ jointly distributed with the inputs $(X, Y)$.

The authors construct a specific source where Eve's observation $Z$ dominates a hypothetical variable $J$ in the "less-noisy" sense (implying $S(X; Y \| Z) = 0$ because $J$ decouples $X$ and $Y$), yet $Z$ cannot simulate $J$. This "rigidity" ensures that while Eve can prevent key distillation, the source retains positive intrinsic information ($I^\downarrow > 0$), which serves as a lower bound for the formation cost.

Key Contributions and Results

1. Explicit Construction of Bound Information (Theorem 1):
   The paper provides the first explicit example of a bipartite source with bound information. The source consists of two bits ($X, Y$) and a trit ($Z$), defined by the joint distribution in Table I.

   • Mechanism: The distribution is generated via a latent bit $L$. Alice and Bob's variables are correlated with $L$ via a specific channel. Eve's variable $Z$ is $L$ passed through an erasure channel (erasing $L$ with probability $1/2$). A hypothetical variable $J$ is $L$ passed through a flip channel (flipping with probability $1/5$).
   • Proof of $S=0$: The flip channel $J$ decouples $X$ and $Y$ (making them independent given $J$). The authors prove that Eve's channel $Z$ is "less noisy" than $J$ ($P_{Z|XY} \succeq_{\text{ln}} P_{J|XY}$). By a result of Gohari, Günlü, and Kramer, this dominance implies the secret-key rate $S(X; Y \| Z)$ is zero.
   • Proof of $I_{\text{form}} > 0$: The authors demonstrate that any processing of $Z$ that decouples $X$ and $Y$ must exactly reproduce the distribution of $J$. However, no channel exists that can transform $Z$ into $J$ due to incompatible probability constraints. Consequently, the intrinsic information $I(X; Y \downarrow Z)$ remains strictly positive, proving $0 < I_{\text{form}}$.

2. Resolution of the Historical Candidate (Theorem 2):
   The paper investigates the specific family of distributions originally proposed by Gisin and Wolf, derived from standard-basis measurements of bound-entangled qutrit states ($\sigma_\alpha$ for $3 < \alpha \leq 4$).

   • Result: Contrary to previous evidence suggesting these distributions might be bound, the authors prove that for every $\alpha$ in the bound-entangled range ($3 < \alpha \leq 4$), a secret key is extractable ($S > 0$).
   • Method: They utilize finite-blocklength witnesses based on the Gohari-Günlü-Kramer criterion, showing that even with just two copies of the source, a positive key rate can be certified.
   • Implication: The analogy between bound entanglement and bound information does not hold at the level of individual states and their standard measurement outcomes.

3. Resource-Level Analogy:
   The authors show that while the standard measurements of the qutrit family do not yield bound information, alternative measurements on the purifications of these states (including separable states) do yield the same bound-information distribution found in Theorem 1. This establishes that the analogy holds at the level of resources (the set of achievable distributions) rather than as a direct dictionary mapping specific quantum states to specific classical distributions.

Significance and Claims
The paper settles the existence question of bound information, confirming that the extreme irreversibility of quantum bound entanglement has a direct classical counterpart in probability theory.

• Conceptual Shift: The work clarifies that the "bound" nature is a property of the resource (the set of correlations), not necessarily of the specific quantum state or its standard measurement outcomes.
• Methodological Advance: The proof relies on the distinction between "simulation" and "less-noisy dominance," a tool that may have broader applications in comparing eavesdroppers and bounding secret-key rates.
• Modesty: The authors explicitly state that their result does not imply that all bound-entangled states yield bound information under standard measurements (in fact, they prove the opposite for the historical candidate). They also note that other historical candidates (e.g., Renner and Wolf's second candidate) remain open, as their mechanism cannot be applied there.

The paper concludes that the irreversibility of "positive creation cost with zero distillable yield" is not an exclusively quantum phenomenon but exists within classical probability, driven by the specific structural rigidity of the constructed source.