Lower Bound for the Communication Complexity of the Russian Cards Problem

In this paper it is shown that no public announcement scheme that can be modeled in Dynamic Epistemic Logic (DEL) can solve the Russian Cards Problem (RCP) in one announcement. Since DEL is a general model for any public announcement scheme we conclude that there exist no single announcement solution to the RCP. The proof demonstrates the utility of DEL in proving lower bounds for communication protocols. It is also shown that a general version of RCP has no two announcement solution when the adversary has sufficiently large number of cards.

💡 Research Summary

The paper investigates the communication complexity of the Russian Cards Problem (RCP) using Dynamic Epistemic Logic (DEL) and establishes rigorous lower bounds on the number of public announcements required to solve the problem. The classic RCP involves three players—Anne, Bill, and Cath—who are dealt 3, 3, and 1 cards respectively from a known set of seven cards. The goal is for Anne and Bill to learn the complete distribution of cards through public, truthful announcements while ensuring that Cath gains no knowledge about any card other than her own.

Modeling in DEL
The authors formalize the problem with a Kripke model where each world corresponds to a possible deal (A, B, C) with |A|=|B|=3, |C|=1, and A∪B∪C = {0,…,6}. The epistemic accessibility relations R(a), R(b), and R(c) are defined so that each player can distinguish worlds only by the cards she holds. Consequently, Anne’s and Bill’s relations partition the 35 possible deals into 35 components of four worlds each, while Cath’s relation yields 7 components of twenty worlds each.

Single‑announcement impossibility
A public announcement is represented by an action model μ. Because the announcement is deterministic, the same action must be applied to every world within a component of Anne’s relation. After executing μ, Anne still cannot separate the four worlds of her component, contradicting the requirement that she end up knowing the exact deal. Lemma 1 formalizes the necessary condition that the final model must contain a singleton component for both Anne and Bill, which cannot be achieved with a single announcement. Theorem 1 therefore proves that no one‑announcement protocol can solve the RCP, regardless of the content or structure of the announcement.

Generalized problem RCP(k; l)
The paper extends the analysis to a family RCP(k; l) where Anne and Bill each hold k cards, Cath holds l cards, and the deck contains 2k + l cards. The same Kripke construction yields (\binom{2k+l}{k}) components for Anne and Bill. Assuming a two‑announcement protocol, the first announcement selects a set of components Tα. Lemma 2 shows that to keep Cath ignorant of any specific card, Tα must either contain all cards or none, which is impossible in practice. Lemma 3 establishes a combinatorial inequality (\lceil\frac{2k+l}{k}\rceil \times \binom{k+l}{k} > \binom{2k+l}{k}) that holds whenever (l \ge 2k^{2}\ln k). This inequality implies that even after the first announcement, Cath can infer at least one card’s location, violating the secrecy requirement. Lemma 4 then demonstrates that for (k\ge2) and (l>2k^{2}\ln k), any first announcement satisfying Lemma 2 leaves at least two indistinguishable worlds for Bill, making a second announcement insufficient to resolve the uncertainty. Consequently, Theorem 2 concludes that two public announcements cannot solve RCP(k; l) in the specified parameter regime.

Significance
The results showcase DEL not merely as a descriptive tool for epistemic updates but as a powerful framework for proving lower bounds on communication protocols. By exploiting the partition structure of epistemic relations and the deterministic nature of action models, the authors derive impossibility results that apply to any public‑announcement scheme, not just specific constructions. This methodology can be transferred to other distributed‑security settings where agents must exchange information publicly without leaking secrets.

Conclusion
The paper rigorously proves that the classic RCP requires at least two public announcements and that, for a broad class of generalized instances, even two announcements are insufficient when the adversary holds sufficiently many cards. These findings confirm the optimality of known two‑announcement solutions and highlight DEL’s utility in establishing fundamental limits on information‑theoretic security in multi‑agent systems.