Local Computation: Lower and Upper Bounds

The question of what can be computed, and how efficiently, are at the core of computer science. Not surprisingly, in distributed systems and networking research, an equally fundamental question is what can be computed in a \emph{distributed} fashion. More precisely, if nodes of a network must base their decision on information in their local neighborhood only, how well can they compute or approximate a global (optimization) problem? In this paper we give the first poly-logarithmic lower bound on such local computation for (optimization) problems including minimum vertex cover, minimum (connected) dominating set, maximum matching, maximal independent set, and maximal matching. In addition we present a new distributed algorithm for solving general covering and packing linear programs. For some problems this algorithm is tight with the lower bounds, for others it is a distributed approximation scheme. Together, our lower and upper bounds establish the local computability and approximability of a large class of problems, characterizing how much local information is required to solve these tasks.

💡 Research Summary

The paper tackles a central question in distributed computing: how much global information can be inferred from purely local views, and what are the inherent limits on the efficiency of such local computation. The authors focus on a broad class of classic graph optimization problems—minimum vertex cover, minimum (connected) dominating set, maximum matching, maximal independent set, and maximal matching—and establish, for the first time, poly‑logarithmic lower bounds on the number of communication rounds required to achieve any non‑trivial approximation when nodes are restricted to information within a bounded radius.

The lower‑bound technique combines a “fake‑graph” construction with information‑theoretic arguments. By designing families of graphs that are indistinguishable to any node within r hops, the authors show that any r‑round algorithm cannot differentiate between instances that have vastly different optimal solutions. A probabilistic analysis of the algorithm’s output then yields concrete lower bounds that depend on the maximum degree Δ of the network: for minimum vertex cover and minimum dominating set, an Ω(log Δ / log log Δ) round complexity is necessary to beat the trivial 2‑approximation; for maximal independent set and maximal matching, an Ω(log Δ) bound is proved. These results close a long‑standing gap in the literature, where only constant‑ or logarithmic‑round upper bounds were known, but no matching lower bounds existed.

On the upper‑bound side, the authors present a unified distributed algorithmic framework for solving general covering and packing linear programs (LPs). The core of the framework is a primal‑dual scaling scheme that iteratively updates local primal variables and dual (Lagrange multiplier) variables while maintaining feasibility within a polylogarithmic number of rounds. Each round requires only O(log n)‑bit messages, fitting the CONGEST model, and the algorithm converges to a (1 + ε)‑approximate solution in (\tilde O(\log^2 n)) rounds. When instantiated for specific combinatorial problems, the framework yields algorithms that match the newly proved lower bounds: a Θ(log Δ) round 2‑approximation for minimum vertex cover, a Θ(log Δ) round (1 + ε)‑approximation scheme for minimum dominating set, and a Θ(log Δ) round 2‑approximation for maximum matching. For some problems the algorithm even provides a distributed approximation scheme (DAS), allowing the trade‑off between ε and round complexity to be tuned arbitrarily.

The paper also discusses practical considerations. The primal‑dual updates are fully local; each node needs only the current values of its incident edges and neighboring constraints. The message size remains bounded, and the algorithm works under synchronous round assumptions, making it suitable for real‑world networks such as data‑center fabrics, IoT deployments, and ad‑hoc wireless systems.

In conclusion, the work delivers a comprehensive characterization of what can be computed locally: it proves that poly‑logarithmic round complexity is fundamentally required for non‑trivial approximations of many classic problems, and it supplies a versatile algorithmic toolkit that attains these limits. The authors outline future directions, including extensions to asynchronous models, handling of non‑linear objectives, and empirical validation on large‑scale testbeds. By bridging the gap between lower‑bound impossibility results and constructive upper‑bound algorithms, the paper sets a new benchmark for the theory and practice of local distributed computation.