Sublinear Bounds for Randomized Leader Election

This paper concerns {\em randomized} leader election in synchronous distributed networks. A distributed leader election algorithm is presented for complete $n$-node networks that runs in O(1) rounds and (with high probability) uses only $O(\sqrt{n}\log^{3/2} n)$ messages to elect a unique leader (with high probability). When considering the “explicit” variant of leader election where eventually every node knows the identity of the leader, our algorithm yields the asymptotically optimal bounds of O(1) rounds and O(n) messages. This algorithm is then extended to one solving leader election on any connected non-bipartite $n$-node graph $G$ in $O(\tau(G))$ time and $O(\tau(G)\sqrt{n}\log^{3/2} n)$ messages, where $\tau(G)$ is the mixing time of a random walk on $G$. The above result implies highly efficient (sublinear running time and messages) leader election algorithms for networks with small mixing times, such as expanders and hypercubes. In contrast, previous leader election algorithms had at least linear message complexity even in complete graphs. Moreover, super-linear message lower bounds are known for time-efficient {\em deterministic} leader election algorithms. Finally, we present an almost matching lower bound for randomized leader election, showing that $\Omega(\sqrt n)$ messages are needed for any leader election algorithm that succeeds with probability at least $1/e + \eps$, for any small constant $\eps > 0$. We view our results as a step towards understanding the randomized complexity ofleader election in distributed networks.

💡 Research Summary

The paper investigates the fundamental communication complexity of randomized leader election in synchronous distributed networks. It introduces a novel algorithm for the complete graph on n nodes that elects a unique leader in a constant number of rounds while using only O(√n · log³⁄² n) messages with high probability. The algorithm works by letting each node become “active” with probability p = Θ(1/√n · log½ n) in each round. Active nodes exchange a single message with a randomly chosen partner; if no collision occurs, the node becomes a leader candidate. Collisions cause the candidate set to shrink, and the parameters are tuned so that after a constant number of rounds the candidate set is reduced to one with probability 1 – 1/poly(n). Consequently, the expected total number of messages is sublinear, namely O(√n · log³⁄² n).

For the “explicit” variant—where every node must eventually learn the identity of the elected leader—the same constant‑round structure is retained, but an additional broadcast phase is required. This adds Θ(n) messages, matching the known lower bound for explicit leader election on a complete graph and proving optimality in that setting.

The authors then extend the technique to any connected, non‑bipartite graph G. The key parameter is the mixing time τ(G) of a simple random walk on G, i.e., the number of steps needed for the walk to approach the stationary distribution within a constant factor. The algorithm proceeds in two phases: (1) a sampling phase where each node performs a τ(G)-step random walk and probabilistically becomes a candidate, and (2) an dissemination phase where the surviving candidates are spread throughout the network using another τ(G)-step walk. Because the walk mixes quickly when τ(G) is small (as in expanders, hypercubes, and many small‑world graphs), the candidate set remains uniformly distributed and the collision analysis from the complete‑graph case carries over. The total running time is O(τ(G)) rounds, and the total message count is O(τ(G)·√n·log³⁄² n). For graphs with τ(G)=O(log n), this yields sub‑logarithmic message complexity, a dramatic improvement over prior deterministic or randomized schemes that required Ω(n) messages even on complete graphs.

A complementary lower‑bound result shows that any randomized leader election algorithm that succeeds with probability at least 1/e + ε (for any constant ε > 0) must send at least Ω(√n) messages. The proof combines an information‑theoretic argument—showing that the system must acquire Ω(√n) independent pieces of information to break symmetry—with a probabilistic collision analysis that bounds the chance of success when fewer messages are exchanged. This lower bound matches the upper bound up to polylogarithmic factors, establishing that the presented algorithm is essentially optimal in the message‑complexity dimension.

The paper also discusses related work, emphasizing that earlier deterministic algorithms achieving constant‑time leader election inevitably incurred linear message costs, and that previous randomized approaches could not beat the linear barrier even on complete graphs. By exploiting randomization together with rapid mixing properties of the underlying network, the authors break this barrier and provide the first sublinear‑message, constant‑time leader election algorithm.

Finally, the authors outline future directions: extending the results to asynchronous models, handling bipartite or directed graphs, and performing empirical evaluations on real‑world network topologies. Overall, the work advances our theoretical understanding of the trade‑off between time and communication in distributed symmetry breaking, and it offers practical algorithms for networks where rapid mixing can be leveraged.