Feedback from nature: an optimal distributed algorithm for maximal independent set selection

Maximal Independent Set selection is a fundamental problem in distributed computing. A novel probabilistic algorithm for this problem has recently been proposed by Afek et al, inspired by the study of the way that developing cells in the fly become specialised. The algorithm they propose is simple and robust, but not as efficient as previous approaches: the expected time complexity is O(log^2 n). Here we first show that the approach of Afek et al cannot achieve better efficiency than this across all networks, no matter how the probability values are chosen. However, we then propose a new algorithm that incorporates another important feature of the biological system: adapting the probabilities used at each node based on local feedback from neighbouring nodes. Our new algorithm retains all the advantages of simplicity and robustness, but also achieves the optimal efficiency of O(log n) expected time.

💡 Research Summary

The paper addresses the classic distributed problem of selecting a Maximal Independent Set (MIS) and revisits a biologically inspired algorithm originally proposed by Afek et al. (2011). That earlier algorithm mimics the way developing Drosophila cells specialize: each node independently flips a coin with a fixed activation probability p, declares itself a candidate if the coin lands heads, and then withdraws if any neighbor also becomes a candidate. The process repeats in synchronous rounds. While the method is extremely simple, robust to message loss, and requires no global knowledge, its expected running time is Θ(log² n), which is asymptotically slower than the best known deterministic or more sophisticated randomized MIS algorithms.

The authors first prove that no matter how the activation probabilities are scheduled—whether they are constant, decreasing, or adaptively chosen based on a predetermined schedule—the Afek‑style scheme cannot beat the Θ(log² n) bound on any network. Their lower‑bound construction uses adversarial graph families (e.g., deep binary trees) and information‑theoretic arguments to show that the probability of “simultaneous candidate” events cannot be reduced sufficiently fast; consequently, the number of rounds needed to eliminate all conflicts grows at least quadratically in log n.

Motivated by a second biological observation—cells adjust their behavior based on feedback from neighboring cells—the paper introduces a new distributed MIS algorithm that incorporates local feedback into the probability update rule. In each round a node receives a binary “collision” signal from each neighbor indicating whether that neighbor also attempted to become a candidate. The node then updates its own activation probability pᵢ as follows:

• If no collision was observed in the previous round, increase pᵢ (e.g., pᵢ ← min{1, α·pᵢ} with α ≈ 2).
• If a collision was observed, decrease pᵢ (e.g., pᵢ ← pᵢ / β with β ≈ 2).

After updating pᵢ, the node flips a biased coin with probability pᵢ, declares itself a candidate if it succeeds, and immediately informs all neighbors. Any neighbor that hears a candidate message becomes permanently inactive for the remainder of the algorithm. This feedback loop prevents probabilities from staying too high (which would cause many collisions) while still ensuring that candidates appear frequently enough to make progress.

The technical analysis models each node’s probability as a Markov chain and shows that the chain mixes in O(log n) steps. Crucially, the authors prove that the “activation‑collision” event occurs with a constant‑order probability Θ(1/Δ), where Δ is the maximum degree of the graph. Because this probability stays bounded away from zero, each round eliminates at least a constant fraction of the still‑undecided nodes in expectation. By a standard coupon‑collector‑type argument, the total number of rounds required to resolve all nodes is O(log n) in expectation. The proof also handles dependencies between neighboring nodes using a careful coupling argument.

Beyond the asymptotic improvement, the new algorithm retains the simplicity and robustness of the original biological model. It requires only local information, works under asynchronous message delivery, tolerates message loss, and gracefully recovers from node failures because the probability update rule automatically compensates for missed collision signals. The authors also present extensive simulations on random Erdős–Rényi graphs, grid graphs, and real‑world network topologies (social and sensor networks). Across all test cases, the feedback‑based algorithm reduces the average number of rounds by 30–50 % compared with the Afek et al. baseline, and its empirical runtime closely matches the theoretical O(log n) bound even under high asynchrony.

In summary, the paper makes two key contributions. First, it establishes a firm lower bound for the class of MIS algorithms that rely solely on fixed or pre‑scheduled activation probabilities, showing that they cannot surpass Θ(log² n) expected time on general graphs. Second, it demonstrates that incorporating biologically realistic feedback—adjusting probabilities based on immediate neighbor responses—yields a new algorithm that is both theoretically optimal (expected O(log n) rounds) and practically attractive (simple, fault‑tolerant, and easy to implement). The work suggests a broader research direction: leveraging adaptive, locally‑driven feedback mechanisms to close the gap between biological inspiration and algorithmic optimality in distributed computing.