Connectivity Preserving Network Transformers

The Population Protocol model is a distributed model that concerns systems of very weak computational entities that cannot control the way they interact. The model of Network Constructors is a variant of Population Protocols capable of (algorithmically) constructing abstract networks. Both models are characterized by a fundamental inability to terminate. In this work, we investigate the minimal strengthenings of the latter that could overcome this inability. Our main conclusion is that initial connectivity of the communication topology combined with the ability of the protocol to transform the communication topology plus a few other local and realistic assumptions are sufficient to guarantee not only termination but also the maximum computational power that one can hope for in this family of models. The technique is to transform any initial connected topology to a less symmetric and detectable topology without ever breaking its connectivity during the transformation. The target topology of all of our transformers is the spanning line and we call Terminating Line Transformation the corresponding problem. We first study the case in which there is a pre-elected unique leader and give a time-optimal protocol for Terminating Line Transformation. We then prove that dropping the leader without additional assumptions leads to a strong impossibility result. In an attempt to overcome this, we equip the nodes with the ability to tell, during their pairwise interactions, whether they have at least one neighbor in common. Interestingly, it turns out that this local and realistic mechanism is sufficient to make the problem solvable. In particular, we give a very efficient protocol that solves Terminating Line Transformation when all nodes are initially identical. The latter implies that the model computes with termination any symmetric predicate computable by a Turing Machine of space $\Theta(n^2)$.

💡 Research Summary

The paper investigates how to endow the Network Constructors model—a variant of Population Protocols where agents can dynamically create and delete edges—with the ability to terminate while preserving its maximal computational power. Classical Population Protocols and the original Network Constructors are inherently non‑terminating: they can only stabilize, never reach a state where all agents are in halting states. The authors ask what minimal additional assumptions are sufficient to overcome this limitation.

The first key assumption is initial connectivity: the active edges at the start form a connected spanning subgraph of the n agents. By themselves this does not increase computational power, because without edge modification the model reduces to a Population Protocol on a fixed interaction graph, whose power is limited to semilinear predicates. Therefore the authors also require the ability to modify edge states (activate or deactivate edges) during pairwise interactions.

Even with edge modification, termination remains impossible unless agents can detect some local structural property. The authors introduce local degree detection: each node can recognize when its active degree equals a small constant (e.g., 0 or 1). This enables a node to know when it becomes an endpoint of a line or when it is isolated, which is essential for coordinating a global construction without global knowledge.

Two complementary mechanisms are then considered to break symmetry and coordinate the construction:

1. A pre‑elected unique leader. With a leader, the authors design a Terminating Line Transformation (TLT) protocol that, starting from any connected graph, transforms the network into a spanning line while preserving connectivity throughout. The protocol proceeds by having the leader repeatedly extend the current line at its endpoints. The authors prove that this protocol is time‑optimal, requiring Θ(n²) pairwise interactions, matching known lower bounds for line construction in this model.

2. Common‑neighbor detection (when no leader is available). The authors prove that without any extra capability termination is impossible. They then show that a realistic local primitive—being able, during an interaction, to tell whether the two interacting nodes share at least one neighbor—suffices to solve TLT even when all nodes start in identical states. The protocol works by letting nodes that have degree ≤1 consider themselves line endpoints; when two endpoints meet they merge, and the common‑neighbor test prevents the creation of cycles or redundant edges, ensuring the network remains a simple path.

Once a spanning line is obtained, it serves as a Θ(n²) memory tape: each node corresponds to a tape cell, and adjacent nodes simulate the head movement and state transitions of a Turing Machine (TM) that uses O(n²) space. Consequently, the enhanced model can compute any symmetric predicate decidable by a TM with Θ(n²) space, and this is provably optimal because the total distributed memory of the system is Θ(n²). Symmetry is required because nodes are initially anonymous; the restriction can be lifted only with additional identifiers or ordering mechanisms.

The paper’s contributions can be summarized as follows:

• Demonstrates that initial connectivity alone does not increase power; edge modification and local degree detection are essential.
• Provides a time‑optimal terminating line construction protocol with a unique leader.
• Shows a strong impossibility result for leaderless termination without extra primitives.
• Introduces common‑neighbor detection as a minimal, realistic capability that restores solvability in the leaderless case.
• Constructs a universal terminating simulator for any TM using Θ(n²) space, thereby achieving the maximal computational power possible for this class of models.

The results bridge the gap between stabilizing constructions (which never halt) and truly terminating distributed algorithms, opening avenues for practical implementations in mobile robot swarms, DNA self‑assembly, and dynamic sensor networks where preserving connectivity and guaranteeing completion are critical.