Skip to main content

Stable leader election in population protocols requires linear time


A population protocol stably elects a leader if, for all n, starting from an initial configuration with n agents each in an identical state, with probability 1 it reaches a configuration \(\mathbf {y}\) that is correct (exactly one agent is in a special leader state \(\ell \)) and stable (every configuration reachable from \(\mathbf {y}\) also has a single agent in state \(\ell \)). We show that any population protocol that stably elects a leader requires \(\varOmega (n)\) expected “parallel time”—\(\varOmega (n^2)\) expected total pairwise interactions—to reach such a stable configuration. Our result also informs the understanding of the time complexity of chemical self-organization by showing an essential difficulty in generating exact quantities of molecular species quickly.

This is a preview of subscription content, access via your institution.


  1. Some recent work on PPs [13, 10, 11, 20] allows the number of states to grow with the number of agents. This paper uses the original model [4] with state set that is constant with respect to the population size. (See Sect. 1.7.)

  2. Some work allows “non-deterministic” transitions, in which the transition function maps to subsets of \(\varLambda \times \varLambda \). Our results are independent of whether the transition function is deterministic or nondeterministic in this manner.

  3. In the most generic model, there is no restriction on which agents are permitted to interact. If one prefers to think of the agents as existing on nodes of a graph, then it is the complete graph \(K_n\) for a population of n agents.

  4. The set of valid initial configurations for a “self-stabilizing” PP is \(\mathbb {N}^\varLambda \), where leader election is provably impossible [9]. We don’t require the PP to work if started in any possible configuration, but rather allow potentially “helpful” initial configurations as long as they don’t already have small count states (see “\(\alpha \)-dense” below).

  5. What “correct” means depends on the task. For computing a predicate, for example, \(\varLambda \) is partitioned into “yes” and “no” voters, and a “correct” configuration is one in which every state present has the correct vote.

  6. See “Open questions” for the distinction between time to converge and time to stabilize. In this paper, the time lower bound we prove is on stabilization.

  7. If it were possible to detect when a leader election protocol such as \(\ell ,\ell \rightarrow \ell ,f\) has stabilized, in the sense that each agent carries a bit \(\{g,s\}\) in which all agents start with g and only transition to s after a single \(\ell \) exists, then one could consider the product state \((\ell ,s)\) to be the “true” leader state, and \((\ell ,g)\) is considered only a “candidate” leader state that may have count \(>1\) prior to the stabilization to a single \(\ell \).

  8. Our result holds for any CRN that obeys Theorem 4.3, the precise constraints of which are specified in [17] (those constraints automatically apply to all PPs). Importantly, to generalize to CRNs, we never assume that the count of agents is fixed, but rather use n to indicate the initial count.

  9. We give a slightly different formalism than that of [8] for population protocols. The main difference is that since we are not deciding a predicate, there is no notion of inputs being mapped to states or states being mapped to outputs. Another difference is that we assume the transition function is symmetric (so there is no notion of a “sender” and “receiver” agent as in [8]; the unordered pair of states completely determines the next pair of states). However, the results of this paper hold even if we allow the transition function to be non-symmetric or even to be non-deterministic (allowing transitions such as \(a,b \rightarrow c,d\) and \(a,b \rightarrow x,y\) to coexist).

  10. When the initial configuration to which a transition sequence is applied is clear from context, we may overload terminology and refer to \((\mathbf {c}_0, \mathbf {c}_1, \ldots )\) as a transition sequence or path.

  11. Since PP’s have a finite reachable configuration space, this is equivalent to requiring that for all \(\mathbf {x}\) reachable from \(\mathbf {c}\), there is a \(\mathbf {c}' \in C\) reachable from \(\mathbf {x}\).

  12. Recall that the condition \(\mathsf {Pr}[\mathbf {i}\mathop {\Longrightarrow }\nolimits Y]=1\) is equivalent to \([(\forall \mathbf {c}\in \mathbb {N}^\varLambda )\ \mathbf {i}\mathop {\Longrightarrow }\nolimits \mathbf {c}\) implies \((\exists \mathbf {y}\in Y)\ \mathbf {c}\mathop {\Longrightarrow }\nolimits \mathbf {y}]\).

  13. This is the main theorem of [17], of which Theorem 4.3 is a corollary.

  14. If \(r_1 \ne r_2\) and \(\mathbf {c}(r_1)=\mathbf {c}(r_2)=b\), then the probability to pick the first agent in one of the states \(r_1\) or \(r_2\) is \(\frac{2b}{n}\), and the probability to pick the second agent in the other state is \(\frac{b}{n-1}\), so the total probability of both is \(\frac{2b^2}{n(n-1)}.\) The case for \(r_1=r_2\) gives \(\frac{b}{n}\) for the first times \(\frac{b-1}{n-1}\) for the second, resulting in lower total probability \(\frac{b^2-b}{n(n-1)}\).

  15. With a nondeterministic transition function, the total number of transitions would replace the quantity \(|\varLambda |^2\) in the conclusion, but it would remain a constant independent of the size of the initial configuration.

  16. Theorem 4.3 was proven in a more general model for Chemical Reaction Networks (CRNs) that obey a certain technical condition [17]. As observed in that paper, the class of CRNs obeying that condition includes all PPs, so the theorem holds unconditionally for PPs. The theorem proved in [17] is more general than Theorem 4.3, but we have stated a corollary of it here. A similar statement is implicit in the proof sketch of Lemma 5 of a technical report on a variant model called “urn automata” that has PPs as a special case [5].

  17. Note the need for \(\mathbf {p}\) to ensure that the total count never goes negative. In writing “\(a=7, b=6, c=-3\)”, we are examining the effect only on \(\mathbf {y}_m\) of modifying \(p_m\), but in applying the lemma, the starting configuration is \(\mathbf {x}_m+\mathbf {p}\), not merely \(\mathbf {x}_m\), so the actual count of each state s will be \(\mathbf {p}(s)\) larger than just stated.


  1. Alistarh, D., Aspnes, J., Eisenstat, D., Gelashvili, R., Rivest, R.L.: Time-space trade-offs in molecular computation. Technical Report, arXiv:1602.08032 (2016)

  2. Alistarh, D., Gelashvili, R.: Polylogarithmic-time leader election in population protocols. In: ICALP 2015: Proceedings of the 42nd International Colloquium on Automata, Languages, and Programming, Kyoto, Japan (2015)

  3. Alistarh, D., Gelashvili, R., Vojnović, M.: Fast and exact majority in population protocols. In: PODC 2015: Proceedings of the 2015 ACM Symposium on Principles of Distributed Computing, pp. 47–56. ACM, New York (2015)

  4. Angluin, D., Aspnes, J., Diamadi, Z., Fischer, M., Peralta, R.: Computation in networks of passively mobile finite-state sensors. Distrib. Comput. 18, 235–253 (2006). Preliminary version appeared in PODC (2004)

    Article  Google Scholar 

  5. Angluin, D., Aspnes, J., Diamadi, Z., Fischer, M.J., Peralta, R.: Urn automata. Technical Report YALEU/DCS/TR-1280, Yale University, November (2003)

  6. Angluin, D., Aspnes, J., Eisenstat, D.: Fast computation by population protocols with a leader. Distrib. Comput. 21(3), 183–199 (2008). Preliminary version appeared in DISC (2006)

    Article  Google Scholar 

  7. Angluin, D., Aspnes, J., Eisenstat, D.: A simple population protocol for fast robust approximate majority. Distrib. Comput. 21(2), 87–102 (2008)

    Article  Google Scholar 

  8. Angluin, D., Aspnes, J., Eisenstat, D., Ruppert, E.: The computational power of population protocols. Distrib. Comput. 20(4), 279–304 (2007)

    Article  Google Scholar 

  9. Angluin, D., Aspnes, J., Fischer, M.J., Jiang, H.: Self-stabilizing population protocols. In: Anderson, J.H., Prencipe, G., Wattenhofer, R. (eds.) Principles of Distributed Systems, pp. 103–117. Springer, Berlin (2006)

    Google Scholar 

  10. Aspnes, J., Beauquier, J., Burman, J., Sohier, D.: Time and space optimal counting in population protocols. (2016)

  11. Beauquier, J., Burman, J., Clavière, S., Sohier, D.: Space-optimal counting in population protocols. In: DISC 2015: Proceedings of the 29th International Symposium on Distributed Computing, pp. 631–646 (2015)

    MATH  Google Scholar 

  12. Bower, J.M., Bolouri, H.: Computational Modeling of Genetic and Biochemical Networks. MIT Press, Cambridge (2004)

    Google Scholar 

  13. Chen, H.-L., Cummings, R., Doty, D., Soloveichik, D.: Speed faults in computation by chemical reaction networks. In: DISC 2014: Proceedings of the 28th International Symposium on Distributed Computing, Austin, TX, USA, pp. 16–30 (2014)

  14. Chen, Y.-J., Dalchau, N., Srinivas, N., Phillips, A., Cardelli, L., Soloveichik, D., Seelig, G.: Programmable chemical controllers made from DNA. Nat. Nanotechnol. 8(10), 755–762 (2013)

    Article  Google Scholar 

  15. Cunha-Ferreira, I., Bento, I., Bettencourt-Dias, M.: From zero to many: control of centriole number in development and disease. Traffic 10(5), 482–498 (2009)

    Article  Google Scholar 

  16. Dickson, L.E.: Finiteness of the odd perfect and primitive abundant numbers with \(n\) distinct prime factors. Am. J. Math. 35(4), 413–422 (1913)

    Article  MathSciNet  Google Scholar 

  17. Doty, D.: Timing in chemical reaction networks. In: SODA 2014: Proceedings of the 25th Annual ACM-SIAM Symposium on Discrete Algorithms, pp. 772–784 (2014)

  18. Doty, D., Soloveichik, D.: Stable leader election in population protocols requires linear time. In: DISC 2015: Proceedings of the 29th International Symposium on Distributed Computing. Lecture Notes in Computer Science, pp. 602–616. Springer, Berlin (2015)

    MATH  Google Scholar 

  19. Gillespie, D.T.: Exact stochastic simulation of coupled chemical reactions. J. Phys. Chem. 81(25), 2340–2361 (1977)

    Article  Google Scholar 

  20. Izumi, T., Kinpara, K., Izumi, T., Wada, K.: Space-efficient self-stabilizing counting population protocols on mobile sensor networks. Theor. Comput. Sci. 552, 99–108 (2014)

    Article  MathSciNet  Google Scholar 

  21. Karp, R.M., Miller, R.E.: Parallel program schemata. J. Comput. Syst. Sci. 3(2), 147–195 (1969)

    Article  MathSciNet  Google Scholar 

  22. Petri, C.A.: Communication with automata. Technical report, DTIC Document (1966)

  23. Soloveichik, D., Seelig, G., Winfree, E.: DNA as a universal substrate for chemical kinetics. Proc. Natl. Acad. Sci. 107(12), 5393 (2010). Preliminary version appeared in DNA 2008

    Article  Google Scholar 

  24. Volterra, V.: Variazioni e fluttuazioni del numero dindividui in specie animali conviventi. Mem. Acad. Lincei Roma 2, 31–113 (1926)

    MATH  Google Scholar 

Download references


The authors thank Anne Condon and Monir Hajiaghayi for several insightful discussions. We also thank the attendees of the 2014 Workshop on Programming Chemical Reaction Networks at the Banff International Research Station, where the first incursions were made into the solution of the problem of PP stable leader election. We are also grateful to anonymous reviewers whose comments have significantly improved the presentation. David Doty was supported by NSF Grants CCF-1619343, CCF-1219274, and CCF-1162589 and the Molecular Programming Project under NSF Grant 1317694. David Soloveichik was supported by an NIGMS Systems Biology Center Grant P50 GM081879 and NSF Grant CCF-1618895.

Author information

Authors and Affiliations


Corresponding author

Correspondence to David Doty.

Additional information

A preliminary version of this article appeared as [18]; the current version has been revised for clarity, and includes several omitted proofs.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Doty, D., Soloveichik, D. Stable leader election in population protocols requires linear time. Distrib. Comput. 31, 257–271 (2018).

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI:


  • Population protocols
  • Leader election
  • Time lower bound
  • Chemical reaction network