An exponential separation between randomized and deterministic complexity in the local model

Yi Jun Chang, Tsvi Kopelowitz, Seth Pettie

Research output: Contribution to journalArticlepeer-review

41 Scopus citations


Over the past 30 years numerous algorithms have been designed for symmetry breaking problems in the \sansL \sansO \sansC \sansA \sansL model, such as maximal matching, MIS, vertex coloring, and edge coloring. For most problems the best randomized algorithm is at least exponentially faster than the best deterministic algorithm. In this paper we prove that these exponential gaps are necessary and establish numerous connections between the deterministic and randomized complexities in the \sansL \sansO \sansC \sansA \sansL model. Each of our results has a very compelling take-away message: Fast \Delta -coloring of trees requires random bits. Building on a recent randomized lower bound of Brandt et al. [A lower bound for the distributed Lov\' asz local lemma, in Proceedings of the 48th ACM Symposium on Theory of Computing (STOC), ACM, New York, 2016, pp. 479-488], we prove that the randomized complexity of \Delta -coloring a tree with maximum degree \Delta is O(log \Delta log n + log \ast n) for any \Delta \geq 55, whereas its deterministic complexity is \Omega (log \Delta n) for any \Delta \geq 3. This also establishes a large separation between the deterministic complexity of \Delta -coloring and (\Delta + 1)-coloring trees. There is a gap in the deterministic complexity hierarchy. We show that any deterministic algorithm for a natural class of problems that runs in O(1) + o(log \Delta n) rounds can be transformed to run in O(log \ast n - log \ast \Delta + 1) rounds. If the transformed algorithm violates a lower bound (even allowing randomization), then one can conclude that the problem requires \Omega (log \Delta n) time deterministically. This gives an alternate proof that deterministically \Delta -coloring a tree with small \Delta takes \Omega (log \Delta n) rounds. Graph shattering is necessary. We prove that the randomized complexity of any natural problem on instances of size n is at least its deterministic complexity on instances of size \surd log n. This shows that any randomized O(1)+o(log \Delta log n)-round algorithm can be derandomized to run in deterministically O(1)+o(log \Delta n) rounds and hence can be transformed to run in O(log \ast n - log \ast \Delta + 1) rounds. This also shows that a deterministic \Omega (log \Delta n) lower bound for any problem (\Delta -coloring a tree, for example) implies a randomized \Omega (log \Delta log n) lower bound. It illustrates that the graph shattering technique employed in recent randomized symmetry breaking algorithms is absolutely essential to the \sansL \sansO \sansC \sansA \sansL model. For example, it is provably impossible to improve the 2 O (\surd log log n ) terms in the complexities of the best MIS and (\Delta + 1)-coloring algorithms without also improving the 2 O (\surd log n ) -round Panconesi-Srinivasan algorithms.

Original languageEnglish
Pages (from-to)122-143
Number of pages22
JournalSIAM Journal on Computing
Issue number1
StatePublished - 2019

Bibliographical note

Publisher Copyright:
© 2019 Society for Industrial and Applied Mathematics. All Rights Reserved.


  • Coloring
  • Distributed algorithm
  • Local model
  • Symmetry breaking


Dive into the research topics of 'An exponential separation between randomized and deterministic complexity in the local model'. Together they form a unique fingerprint.

Cite this