Passively Mobile Communicating Machines that Use Restricted Space

We propose a new theoretical model for passively mobile Wireless Sensor Networks, called PM, standing for Passively mobile Machines. The main modification w.r.t. the Population Protocol model is that agents now, instead of being automata, are Turing Machines. We provide general definitions for unbounded memories, but we are mainly interested in computations upper-bounded by plausible space limitations. However, we prove that our results hold for more general cases. We focus on complete communication graphs and define the complexity classes PMSPACE(f(n)) parametrically, consisting of all predicates that are stably computable by some PM protocol that uses O(f(n)) memory on each agent. We provide a protocol that generates unique ids from scratch only by using O(log n) memory, and use it to provide an exact characterization for the classes PMSPACE(f(n)) when f(n)={\Omega}(log n): they are precisely the classes of all symmetric predicates in NSPACE(nf(n)). In this way, we provide a space hierarchy for the PM model when the memory bounds are {\Omega}(log n). Finally, we explore the computability of the PM model when the protocols use o(loglog n) space per machine and prove that SEMILINEAR=PMSPACE(f(n)) when f(n)=o(loglog n), where SEMILINEAR denotes the class of the semilinear predicates. In fact, we prove that this bound acts as a threshold, so that SEMILINEAR is a proper subset of PMSPACE(f(n)) when f(n)=O(loglog n).

💡 Research Summary

The paper introduces a new theoretical framework for passively mobile wireless sensor networks called the PM (Passively mobile Machines) model. While the classic Population Protocol model treats each agent as a finite‑state automaton with constant memory, the PM model upgrades agents to Turing machines that are allowed a bounded amount of workspace. The authors formalize this by defining a function f(n) that specifies the number of memory words available to each of the n agents, and they denote by PMSPACE(f(n)) the class of predicates that can be stably computed by some PM protocol using O(f(n)) memory per agent.

A central technical contribution is a protocol that, starting from a completely uniform initial configuration, assigns a unique identifier (ID) to every agent while using only O(log n) memory per agent. The protocol works through random pairwise interactions: each agent stores a logarithmic‑size bit string representing its current ID, compares it with the partner’s ID, and updates both IDs in a way that gradually eliminates collisions. After a polynomial number of interactions, with high probability, the IDs become a permutation of {1,…,n}. This result shows that even with extremely limited local memory, global uniqueness can be achieved without any external naming service.

Using the ID generation as a building block, the authors characterize the computational power of the PM model for different space regimes. When f(n) ≥ Ω(log n), they prove that PMSPACE(f(n)) coincides exactly with the set of symmetric predicates that belong to NSPACE(n·f(n)). The proof proceeds in two directions. First, any PM protocol uses at most n·f(n) workspace in total, so a nondeterministic Turing machine can simulate the whole system within that space, establishing the inclusion PMSPACE(f(n)) ⊆ NSPACE(n·f(n)). Second, for any symmetric language L in NSPACE(n·f(n)), they construct a PM protocol that simulates the nondeterministic space‑bounded computation by distributing the tape among the agents and coordinating via pairwise interactions. The unique IDs allow the agents to agree on a global ordering, which is essential for the simulation.

The paper then investigates the low‑space regime. When f(n)=o(log log n), the authors show that the PM model collapses to the well‑known semilinear class (SEMILINEAR), i.e., the class of predicates definable by Presburger arithmetic and exactly the predicates computable by classic population protocols. They prove two lemmas: (1) with sub‑log‑log space each agent cannot store enough information to count beyond a constant factor, which forces any stable computation to be a linear combination of input counts; (2) conversely, any semilinear predicate can be realized by a PM protocol that first generates unique IDs (still possible with o(log log n) memory) and then performs a simple counting routine.

A striking observation is that Θ(log log n) memory per agent acts as a threshold. Once agents are allowed Θ(log log n) or more workspace, PMSPACE(f(n)) strictly contains SEMILINEAR; the model can compute non‑semilinear symmetric predicates such as “the input size is a perfect square” or “the number of 1’s is a power of two.” Thus the authors establish a space hierarchy for the PM model:

• O(1) or o(log log n) → SEMILINEAR only,
• Ω(log n) → exactly the symmetric part of NSPACE(n·f(n)),
• Intermediate Θ(log log n) → strictly larger than SEMILINEAR but still below the full NSPACE bound.

Beyond the theoretical contributions, the paper discusses practical implications. In ultra‑low‑power IoT devices that may have only a few bytes of RAM, the O(log n) ID‑generation protocol provides a feasible way to assign unique network addresses without centralized coordination. Moreover, the ability to perform non‑semilinear computations with only logarithmic memory per node suggests that moderately capable sensor nodes can execute more sophisticated aggregation functions (e.g., median, variance) in a fully decentralized manner.

In summary, the authors present a comprehensive study of a space‑bounded, passively mobile computation model. They deliver (i) a constructive ID‑generation scheme using minimal memory, (ii) a precise characterization of the power of the model for f(n) ≥ Ω(log n) as the symmetric fragment of NSPACE(n·f(n)), and (iii) a threshold phenomenon at Θ(log log n) separating semilinear from strictly more powerful computations. These results bridge the gap between classical population protocols and realistic sensor networks where agents possess modest but non‑constant memory, opening new avenues for both theoretical exploration and practical protocol design.