Expressive Power of Broadcast Consensus Protocols

Population protocols are a formal model of computation by identical, anonymous mobile agents interacting in pairs. Their computational power is rather limited: Angluin et al. have shown that they can only compute the predicates over $\mathbb{N}^k$ expressible in Presburger arithmetic. For this reason, several extensions of the model have been proposed, including the addition of devices called cover-time services, absence detectors, and clocks. All these extensions increase the expressive power to the class of predicates over $\mathbb{N}^k$ lying in the complexity class NL when the input is given in unary. However, these devices are difficult to implement, since they require that an agent atomically receives messages from all other agents in a population of unknown size; moreover, the agent must know that they have all been received. Inspired by the work of the verification community on Emerson and Namjoshi’s broadcast protocols, we show that NL-power is also achieved by extending population protocols with reliable broadcasts, a simpler, standard communication primitive.

💡 Research Summary

**
The paper investigates the computational power of population protocols—systems of identical, anonymous agents that interact pairwise—and shows how to extend them with a simple, well‑studied communication primitive: reliable broadcast. Classical population protocols are limited to predicates expressible in Presburger arithmetic (i.e., first‑order formulas over addition). To go beyond this limitation, earlier works introduced powerful but hard‑to‑implement devices such as cover‑time services, absence detectors, and clocks, which raise the expressive power to exactly the class NL (nondeterministic logarithmic space) when inputs are given in unary. However, those devices require an agent to receive messages from every other agent and to know when this reception is complete—an unrealistic assumption in many distributed settings.

Motivated by Emerson and Namjoshi’s broadcast protocols, the authors propose broadcast consensus protocols. In addition to the usual pairwise rendezvous transitions, a broadcast transition is a triple (q → r; f) where an agent in state q broadcasts a signal, moves to state r, and every other agent simultaneously applies a deterministic transfer function f to its own state. The broadcast is reliable: all agents receive it, and the transition is atomic from the perspective of the broadcasting agent.

The central technical contribution is a proof that broadcast consensus protocols compute exactly the predicates in NL. The proof proceeds in two main steps:

1. Silent semi‑computation – a weaker notion than standard stabilization. A protocol silently semi‑computes a predicate if, on every fair execution, it eventually reaches a configuration that never changes again (silent) and whose output equals the predicate value. The authors show how to simulate any NL‑machine using broadcasts: the machine’s configuration (state, tape head, and logarithmic‑size work tape) is encoded as a multiset of agents. Broadcasts are used to make nondeterministic choices, to propagate global checks, and crucially to restart the whole population when an inconsistency is detected. A special “reset” broadcast r forces every agent back to the initial configuration; fairness guarantees that a correct computation path will be tried infinitely often, leading to a silent consensus on the correct answer.

2. Equivalence of silent semi‑computation and full computation for NL – the authors prove that any predicate that can be silently semi‑computed by a broadcast protocol can also be fully computed (i.e., stabilizes to the correct output) by a standard population protocol with broadcasts. This step uses the fact that NL is closed under complement and under log‑space reductions, allowing the construction of a protocol that, once it reaches a silent configuration, simply halts further changes.

Together these results establish the main theorem: A predicate over ℕ^k is computable by a broadcast consensus protocol (with unary input) iff it belongs to NL. Consequently, broadcast consensus protocols are strictly more powerful than ordinary population protocols (they can decide, for example, whether the number of agents is a power of two) but no more powerful than the previously studied extensions that rely on global synchronization.

The paper also explores minimality aspects:

• Single‑signal broadcast: It suffices that only one agent can broadcast a single, fixed signal; all other agents react deterministically. This already yields NL power.
• Reset does not increase power: Adding an explicit reset operation (which returns all agents to their initial states) does not enlarge the class of computable predicates beyond NL.
• Leaderless protocols: Although many constructions use a distinguished set of leader agents, the authors show that any protocol with leaders can be simulated by a leaderless one. A first broadcast can elect a unique leader, after which the rest of the computation proceeds as in the leader‑based version.

A concrete example is given: a broadcast protocol that decides whether the number of agents is a power of two. The protocol repeatedly halves the number of agents in a designated state using pairwise interactions; broadcasts detect odd remainders and trigger a reset. Fairness ensures that the correct halving sequence is eventually followed, leading to a stable consensus exactly when the input size is a power of two—an example of a predicate outside Presburger arithmetic.

The paper is organized as follows:

1. Introduction – motivation, related work, and overview of contributions.
2. Preliminaries – formal definitions of multisets, population protocols, fairness, and the notion of computed predicates.
3. Broadcast consensus protocols – formal model, definition of broadcast transitions, and illustrative examples (including the power‑of‑two protocol and leaderless simulation).
4. Expressiveness equals NL – detailed construction of NL‑machine simulation, silent semi‑computation, and proof of equivalence.
5. Minimal requirements – analysis of single‑signal, single‑broadcaster, reset, and leaderless variants.
6. Conclusion and future work – discussion of practical implementation prospects (wireless sensor networks, synthetic biology) and open questions such as extensions to richer input encodings or probabilistic variants.

In summary, the authors demonstrate that adding a reliable broadcast primitive to population protocols yields a model that is both theoretically optimal (exactly NL) and practically appealing (requires only a simple broadcast, no global synchronization). This bridges the gap between the high expressive power of previously proposed extensions and the feasibility of real‑world distributed systems, opening avenues for implementing NL‑level distributed algorithms in sensor networks, swarm robotics, and synthetic biological circuits.