Abstract
Random constraint satisfaction problems exhibit several phase transitions when their density of constraints is varied. One of these threshold phenomena, known as the clustering or dynamic transition, corresponds to a transition for an information theoretic problem called tree reconstruction. In this article we study this threshold for two CSPs, namely the bicoloring of k-uniform hypergraphs with a density \(\alpha \) of constraints, and the q-coloring of random graphs with average degree c. We show that in the large k, q limit the clustering transition occurs for \(\alpha = \frac{2^{k-1}}{k} (\ln k + \ln \ln k + \gamma _{\mathrm{d}} + o(1))\), \(c= q (\ln q + \ln \ln q + \gamma _{\mathrm{d}}+ o(1))\), where \(\gamma _{\mathrm{d}}\) is the same constant for both models. We characterize \(\gamma _{\mathrm{d}}\) via a functional equation, solve the latter numerically to estimate \(\gamma _{\mathrm{d}} \approx 0.871\), and obtain an analytic lowerbound \(\gamma _{\mathrm{d}} \ge 1 + \ln (2 (\sqrt{2}-1)) \approx 0.812\). Our analysis unveils a subtle interplay of the clustering transition with the rigidity (naive reconstruction) threshold that occurs on the same asymptotic scale at \(\gamma _{\mathrm{r}}=1\).
We’re sorry, something doesn't seem to be working properly.
Please try refreshing the page. If that doesn't work, please contact support so we can address the problem.
References
Abou-Chacra, R., Thouless, D.J., Anderson, P.W.: A selfconsistent theory of localization. J. Phys. C 6(10), 1734 (1973). https://doi.org/10.1088/0022-3719/6/10/009. http://stacks.iop.org/0022-3719/6/i=10/a=009
Achlioptas, D., Coja-Oghlan, A.: Algorithmic barriers from phase transitions. In: Proceedings of FOCS 2008, p. 793 (2008). https://doi.org/10.1109/FOCS.2008.11
Achlioptas, D., Ricci-Tersenghi, F.: On the solution-space geometry of random constraint satisfaction problems. In: Proc. of 38th STOC, pp. 130–139. ACM, New York (2006). https://doi.org/10.1145/1132516.1132537
Asmussen, S., Rosinski, J.: Approximations of small jumps of lévy processes with a view towards simulation. J. Appl. Probab. 38(2), 482–493 (2001). https://doi.org/10.1239/jap/996986757
Barbier, J., Krzakala, F., Zdeborová, L., Zhang, P.: The hard-core model on random graphs revisited. J. Phys. 473, 012021 (2013). https://doi.org/10.1088/1742-6596/473/1/012021
Bhatnagar, N., Sly, A., Tetali, P.: Reconstruction threshold for the hardcore model. In: M. Serna, R. Shaltiel, K. Jansen, J. Rolim (eds.) Approximation, Randomization, and Combinatorial Optimization. Algorithms and Techniques, pp. 434–447 (2010)
Biroli, G., Monasson, R., Weigt, M.: A variational description of the ground state structure in random satisfiability problems. Eur. Phys. J. B 14, 551 (2000). https://doi.org/10.1007/s100510051065
Braunstein, A., Mézard, M., Zecchina, R.: Survey propagation: an algorithm for satisfiability. Random Struct. Algorithms 27(2), 201–226 (2005). https://doi.org/10.1002/rsa.20057
Brightwell, G.R., Winkler, P.: A second threshold for the hard-core model on a bethe lattice. Random Struct. Algorithms 24(3), 303–314 (2004). https://doi.org/10.1002/rsa.20006
Budzynski, L., Ricci-Tersenghi, F., Semerjian, G.: Biased landscapes for random constraint satisfaction problems. J. Stat. Mech. 2019(2), 023302 (2019). https://doi.org/10.1088/1742-5468/ab02de
Chi, Z.: Nonnormal small jump approximation of infinitely divisible distributions. Adv. Appl. Probab. 46(4), 963–984 (2014). https://doi.org/10.1239/aap/1418396239
Dall’Asta, L., Ramezanpour, A., Zecchina, R.: Entropy landscape and non-Gibbs solutions in constraint satisfaction problems. Phys. Rev. E 77, 031118 (2008). https://doi.org/10.1103/PhysRevE.77.031118
Ding, J., Sly, A., Sun, N.: Proof of the satisfiability conjecture for large k. In: Proceedings of the Forty-seventh Annual ACM Symposium on Theory of Computing, STOC’15, pp. 59–68 (2015). https://doi.org/10.1145/2746539.2746619
Feller, W.: An introduction to probability theory and its applications, vol. 2, 2nd edn. Wiley, New York (1971)
Gabrié, M., Dani, V., Semerjian, G., Zdeborová, L.: Phase transitions in the q-coloring of random hypergraphs. J. Phys. A 50(50), 505002 (2017). https://doi.org/10.1088/1751-8121/aa9529
Garey, M., Johnson, D.: Computers and Intractability: A Guide to the Theory of NP-Completeness. Freeman, San Francisco (1979)
Krzakala, F., Montanari, A., Ricci-Tersenghi, F., Semerjian, G., Zdeborova, L.: Gibbs states and the set of solutions of random constraint satisfaction problems. Proc. Natl. Acad. Sci. 104(25), 10318–10323 (2007). https://doi.org/10.1073/pnas.0703685104
Kschischang, F.R., Frey, B., Loeliger, H.A.: Factor graphs and the sum-product algorithm. IEEE Trans. Inform. Theory 47(2), 498–519 (2001). https://doi.org/10.1109/18.910572
Mertens, S., Mézard, M., Zecchina, R.: Threshold values of random k-sat from the cavity method. Random Struct. Algorithms 28(3), 340–373 (2006). https://doi.org/10.1002/rsa.20090
Mézard, M., Montanari, A.: Reconstruction on trees and spin glass transition. J. Stat. Phys. 124, 1317–1350 (2006). https://doi.org/10.1007/s10955-006-9162-3
Mézard, M., Parisi, G.: The bethe lattice spin glass revisited. Eur. Phys. J. B 20, 217 (2001). https://doi.org/10.1007/PL00011099
Mézard, M., Parisi, G., Zecchina, R.: Analytic and algorithmic solution of random satisfiability problems. Science 297, 812–815 (2002). https://doi.org/10.1126/science.1073287
Molloy, M.: The freezing threshold for k-colourings of a random graph. In: Proceedings of the 44th symposium on Theory of Computing, p. 921. ACM (2012). https://doi.org/10.1145/2213977.2214060
Monasson, R., Zecchina, R., Kirkpatrick, S., Selman, B., Troyansky, L.: 2+p-sat: relation of typical-case complexity to the nature of the phase transition. Random Struct. Algorithms 15, 414 (1999). https://doi.org/10.1002/(SICI)1098-2418(199910/12)15:3/4<414::AID-RSA10>3.0.CO;2-G
Montanari, A., Semerjian, G.: Rigorous inequalities between length and time scales in glassy systems. J. Stat. Phys. 125, 23 (2006). https://doi.org/10.1007/s10955-006-9175-y
Montanari, A., Ricci-Tersenghi, F., Semerjian, G.: Clusters of solutions and replica symmetry breaking in random k-satisfiability. J. Stat. Mech. P04004, (2008). https://doi.org/10.1088/1742-5468/2008/04/p04004
Montanari, A., Restrepo, R., Tetali, P.: Reconstruction and clustering in random constraint satisfaction problems. SIAM J. Discrete Math. 25(2), 771–808 (2011). https://doi.org/10.1137/090755862
Mossel, E., Peres, Y.: Information flow on trees. Ann. Appl. Probab. 13(3), 817–844 (2003). https://doi.org/10.1214/aoap/1060202828
Nishimori, H.: Statistical Physics of Spin Glasses and Information Processing: An Introduction. Oxford University Press, Oxford (2001)
Papadimitriou, C.H.: Computational Complexity. Addison-Wesley, Reading (1994)
Richardson, T., Urbanke, R.: Modern Coding Theory. Cambridge University Press, Cambridge (2007)
Sato, K.I.: Lévy Processes and Infinitely Divisible Distributions. Cambridge University Press, Cambridge (2013)
Semerjian, G.: On the freezing of variables in random constraint satisfaction problems. J. Stat. Phys. 130, 251 (2008). https://doi.org/10.1007/s10955-007-9417-7
Sly, A.: Reconstruction of random colourings. Commun. Math. Phys. 288(3), 943–961 (2009). https://doi.org/10.1007/s00220-009-0783-7
Sly, A., Zhang, Y.: Reconstruction of colourings without freezing. arXiv preprint arXiv:1610.02770 (2016)
Watanabe, T., Yamamuro, K.: Ratio of the tail of an infinitely divisible distribution on the line to that of its lévy measure. Electron. J. Probab. 15, 44–74 (2010). https://doi.org/10.1214/EJP.v15-732
Zdeborová, L., Krzakala, F.: Phase transitions in the coloring of random graphs. Phys. Rev. E 76, 031131 (2007). https://doi.org/10.1103/PhysRevE.76.031131
Acknowledgements
GS is part of the PAIL grant of the French Agence Nationale de la Recherche, ANR-17-CE23-0023-01.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they have no conflict of interest.
Additional information
Communicated by Federico Ricci-Tersenghi.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
A The Graph Coloring Case
A The Graph Coloring Case
This Appendix is devoted to the graph coloring problem. We shall present computations that are the counterparts of the ones explained in Sect. 2 and 3 of the main text for the hypergraph bicoloring problem, namely the definition of the reconstruction problem and its study in the limit where the number of colors q and the degree c diverge simultaneously according to (1), and show that the very same equations summarized at the beginning of Sect. 4 arise in this limit. As the computations are quite similar we will be more succinct than in the main text and concentrate on the specificities of the coloring problem.
Let us consider a rooted tree with spins \(\sigma _i\) placed on its vertices. These spins can take q values, interpreted as colors, \(\sigma _i\in \{1,\dots ,q \}\). A proper coloring of the tree is a configuration of the spins such that no edge is monochromatic (i.e. no pair of adjacent vertices are given the same color). A uniform proper coloring can be drawn in a broadcast fashion, by choosing the color \(\sigma \) of the root uniformly at random among the q possible ones, then each descendent of the root is assigned a color uniformly at random among the \(q-1\) colors distinct from \(\sigma \), and this is repeated recursively down to the n-th generation of the tree. In the reconstruction problem an observer is then provided with the colors on the vertices of the n-th generation of the tree only, and asked to guess the color of the root. The optimal strategy is to compute \(\eta \), the posterior probability of the root given the observations, which is a distribution over \(\{1,\dots ,q\}\). A moment of thought reveals that the probability distribution of \(\eta \), with respect to a broadcast process conditioned on the root value \(\sigma \), and with respect to a random choice of the tree as a Galton-Watson branching process with Poisson offspring distribution of mean c, is a measure \(P_{\sigma ,n}(\eta )\) that can be determined recursively through the induction relation:
where the Belief Propagation recursion function \(f_{\mathrm{c}}\) is defined here in such a way that \(\eta =f_\mathrm{c}(\eta _1,\dots ,\eta _l)\) means
The initial condition of this recursive computation is given by \(P_{\sigma ,0}=\delta (\eta - \delta _\sigma )\), where \(\delta _\sigma \) is the measure concentrated on the color \(\sigma \), i.e. \(\delta _\sigma (\tau ) = {\mathbb {I}}(\sigma =\tau )\), which corresponds to the colors being revealed on the leaves of the tree. These equations correspond to (9,12,13) for the hypergraph bicoloring problem.
We separate now the contribution of “hard fields”, or frozen variables, namely the configurations of boundary variables that determine unambiguously the root in the naive reconstruction procedure. This forced value can only be the correct one the root had in the broadcast process, hence we shall write :
where \(Q_{\sigma ,n}\) has no atom on \(\delta _\sigma \); this mimicks the decomposition (22) of the main text. Plugging this decomposition in (86) yields the evolution equation for the weight of the hard fields:
Indeed \(f_{\mathrm{c}}(\eta _1,\dots ,\eta _l) = \delta _\sigma \) if and only if for each color \(\sigma '\ne \sigma \) at least one of the arguments \(\eta _i\) is equal to \(\delta _{\sigma '}\), thus forbidding all colors except \(\sigma \). The number of hard fields of the color \(\sigma ' \ne \sigma \) is easily seen from (86) to be Poisson distributed with average \(\frac{c w_n}{q-1}\), independently from one color \(\sigma '\) to another, from which (89) follows.
The recursion (89) has a bifurcation at \(c_{\mathrm{r}}(q)\), in the sense that \(w_n \rightarrow 0\) as \(n \rightarrow \infty \) if and only if \(c<c_{\mathrm{r}}(q)\). Writing down the equations fixing \(c_{\mathrm{r}}\) and \(w_{\mathrm{r}}\) at the bifurcation, which are similar to (25), and then expanding them for large q one finds the asymptotic expansion \(c_{\mathrm{r}}=q (\ln q + \ln \ln q + 1 + o(1))\). We shall thus study the large q limit with c getting also large, on the scale \(c=q (\ln q + \ln \ln q + \gamma )\) with \(\gamma \) finite. In this limit one finds that for n finite
where \(x_n\) is of order 1 and obeys exactly the same recursion as in the bicoloring case, namely \(x_0=0\) and \(x_{n+1}=e^{-\gamma + x_n}\).
Let us now simplify the evolution equation for the distributions \(Q_{\sigma ,n}\) of the soft fields. First of all we insert the decomposition (88) in the right hand side of (86) and obtain, without any approximation,
where l is the number of neighbors of the root that receive a soft field, \(\sigma _1,\dots ,\sigma _l\) the colors these vertices have in the broadcast, and the indicator variables \(p_{\sigma '}\) are equal to 1 if and only if no neighbor of the root assigned the color \(\sigma '\) in the broadcast is perfectly recovered (i.e. receives a hard field). Hence the colors \(\sigma '\ne \sigma \) with \(p_{\sigma '}=0\) are precisely the ones forbidden for the root, as at least one of its neighbors is forced to this value. The relation \(\eta ={\widetilde{f}}_\mathrm{c}(\sigma ,\{p_{\sigma '}\}_{\sigma ' \ne \sigma };\eta _1,\dots ,\eta _l)\) is obtained by specializing \(f_{\mathrm{c}}\) of (87) to this pattern for the presence of hard fields in its arguments, and thus reads
with the convention \(p_{\sigma }=1\).
Let us call \(p =\underset{\sigma ' \ne \sigma }{\sum } p_{\sigma '}\) the number of colors that satisfy the condition explained above; in the equation (91) it corresponds to a random variable with a binomial distribution of parameters \((q-1,e^{-\frac{c w_n}{q-1}})\). According to (90) the product of these parameters go to zero as \(1/\ln q\) in the limit we are considering, we shall thus truncate (91) on the smallest possible values of p. As \(p=0\) yields a hard field in the left hand side of (91), the distribution of the soft fields is dominated in this limit by the case \(p=1\). As a consequence the fields \(\eta \) in the support of \(Q_{\sigma ,n}\) have non-zero values on two colors only, \(\sigma \) and another one \(\sigma '\) uniformly distributed on the \(q-1\) possibilities. Let us parametrize this type of distributions via a distribution \(Q_n(h)\) on real random variables \(h\in [-1,1]\),
with
Let us also denote \({\widehat{f}}_{\mathrm{c}}(\sigma ,\sigma ';\eta _1,\dots ,\eta _l)\) the function \({\widetilde{f}}_{\mathrm{c}}(\sigma ,\{p_{\sigma ''}\}_{\sigma '' \ne \sigma };\eta _1,\dots ,\eta _l)\) with \(p_{\sigma ''}=\delta _{\sigma '',\sigma '}\), in such a way that the only two non-zero components of \({\widehat{f}}_\mathrm{c}(\sigma ,\sigma ';\eta _1,\dots ,\eta _l)\) correspond to the colors \(\sigma \) and \(\sigma '\). Using this notation, and the parametrization (93), one can deduce from (91) the evolution equation for \(Q_{\sigma ,n}(\eta )\) at lowest order:
To put this expression under the form (93), and hence close the recursion on \(Q_n(h)\), it remains to notice that
where f is the function defined for Ising spins in (9), and
For a fixed choice of \(\sigma \) and \(\sigma ' \ne \sigma \), one can see that \({\widehat{u}}(\sigma ,\sigma ';s(\sigma _i,\sigma '_i;h_i))\) is a random variable with respect to the uniform choices of \(\sigma _i \ne \sigma \) and \(\sigma '_i \ne \sigma _i\), that takes the following values (recall the definition of the function \(g_1\) from (39)):
As \(c(1-w_n)\frac{q-2}{(q-1)^2} \rightarrow x_n \) in the regime we are considering, the number of occurrences of the first three cases in (95) is Poissonian of mean \(x_n\); on the other hand the number of times the fourth case happens vanishes when q diverges (as 1/q), while the fifth does not contribute to (95), because \(f(u_1,\dots ,u_l,0,\dots ,0)=f(u_1,\dots ,u_l)\). We thus see by comparison with (42,43) that the probability distribution \(Q_n(h)\) defined in (93) obeys exactly the same recursion equations as the one derived in the main text for the hypergraph bicoloring model, the initial condition \(Q_1(h)=\delta (h)\) being also valid here.
Rights and permissions
About this article
Cite this article
Budzynski, L., Semerjian, G. The Asymptotics of the Clustering Transition for Random Constraint Satisfaction Problems. J Stat Phys 181, 1490–1522 (2020). https://doi.org/10.1007/s10955-020-02635-8
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10955-020-02635-8