Multiparty Equality Function Computation in Networks with Point-to-Point Links

In this report, we study the multiparty communication complexity problem of the multiparty equality function (MEQ): EQ(x_1,…,x_n) = 1 if x_1=…=x_n, and 0 otherwise. The input vector (x_1,…,x_n) is distributed among n>=2 nodes, with x_i known to node i, where x_i is chosen from the set {1,…,M}, for some integer M>0. Instead of the “number on the forehand” model, we consider a point-to-point communication model (similar to the message passing model), which we believe is more realistic in networking settings. We assume a synchronous fully connected network of n nodes, the node IDs (identifiers) are common knowledge. We assume that all point-to-point communication channels/links are private such that when a node transmits, only the designated recipient can receive the message. The identity of the sender is known to the recipient. We demonstrate that traditional techniques generalized from two-party communication complexity problem are not sufficient to obtain tight bounds under the point-to-point communication model. We then introduce techniques which significantly reduce the space of protocols to study. These techniques are used to study some instances of the MEQ problem.

💡 Research Summary

The paper investigates the communication complexity of the multiparty equality (MEQ) function in a realistic point‑to‑point (P2P) message‑passing setting. In the MEQ problem n ≥ 2 nodes each hold a private input x_i drawn from {1,…,M}. The goal is for all nodes to output 1 iff all inputs are identical, and 0 otherwise, while minimizing the total number of bits transmitted over private, directed links in a synchronous fully‑connected network.

Traditional two‑party communication‑complexity techniques—information‑theoretic lower bounds, rectangle partitions, and reductions to set‑disjointness—are shown to be insufficient when directly transplanted to the P2P model. The key difficulty is that, unlike the “number‑on‑the‑forehead” model where every player sees all other inputs, here each message is visible only to its intended recipient, and the sender’s identity is known. Consequently, the usual symmetry arguments break down, and naïve extensions yield overly loose bounds.

To overcome these obstacles the authors introduce two complementary methodological advances.

1. Protocol Compression – The space of possible messages is reduced by mapping each input value to an equivalence‑class identifier that can be communicated using ⌈log₂ C⌉ bits, where C is the number of classes (typically much smaller than M). A shared codebook is established in a one‑time setup phase, after which each node only needs to transmit its class ID. This compression dramatically shrinks the protocol search space from exponential in M to exponential in C, enabling tighter analysis.

2. Connection‑Graph Decomposition – The complete communication graph Kₙ is partitioned into a collection of sub‑graphs (stars, chains, bipartite cliques, etc.). Within each sub‑graph a local equality test is performed, often by a designated hub node that gathers the inputs of its neighbours, compares them, and disseminates the result. The global answer is then obtained by aggregating the sub‑graph outcomes. This hierarchical approach distributes the communication load, avoids the need for every pair of nodes to exchange messages, and aligns naturally with the private‑link constraint.

The paper applies these techniques to concrete instances. For (n = 3, M = 4) the naïve two‑party bound would suggest 3·⌈log₂ 4⌉ = 6 bits, but using protocol compression the authors construct a 5‑bit protocol that is optimal for this setting. For (n = 4, M = 2) a star topology centered at one node yields a total cost of 2·⌈log₂ 2⌉ = 2 bits, proving that centralization can be optimal when M is small.

Beyond specific cases, the authors discuss how the two techniques scale. Protocol compression relies on a logarithmic encoding of inputs; the choice of the class granularity parameter κ determines a trade‑off between compression ratio and the difficulty of class identification. Graph decomposition can be framed as a min‑cut or max‑matching problem; although finding the exact optimal decomposition is NP‑hard, standard approximation algorithms provide sufficiently good partitions for practical networks. By combining compressed messages with an appropriately chosen decomposition, the total communication cost approaches the lower bounds known for the two‑party model, even under the stricter P2P constraints.

The work concludes that, in realistic networking environments where links are private and directed, the MEQ function can be computed far more efficiently than previously thought. The introduced methods have immediate relevance to distributed consensus, blockchain validation, and secure multiparty computation where privacy of messages is mandatory. Future directions include extending the analysis to sparse or dynamic topologies, handling asynchronous communication, and incorporating adversarial behavior to derive robust security‑aware lower bounds.