Abstract
We consider mean field ferromagnetic spin models on dilute random graphs and prove that, with suitable one-body infinitesimal perturbations added to the Hamiltonian, the multi-overlaps concentrate for all temperatures, both with respect to the thermal Gibbs average and the quenched randomness. Results of this nature have been known only for the lowest order overlaps, at high temperature or on the Nishimori line. Here we treat all multi-overlaps by a non-trivial application of Griffiths–Kelly–Sherman correlation inequalities. Our results apply in particular to the pure and mixed p-spin ferromagnets on random dilute Erdoes–Rényi hypergraphs. On physical grounds one expects that multi-overlap concentration is an important ingredient for the validity of the cavity (or replica-symmetric) formula for the pressure of mean field models. However rigorously establishing this formula for the p-spin ferromagnet on a random dilute hypergraph remains an open problem.
Similar content being viewed by others
Notes
By infinitesimal perturbations we mean perturbations that do not change the thermodynamic limit of the pressure when we take the limit of zero perturbation after the thermodynamic limit.
With minor adjustments in the formulation of the models we can also cover ferromagnets on dense graphs.
It is an open problem to assess if these can be dropped and the fluctuations controlled only thanks to the homogeneous perturbation.
Interpolating from \(t=0\) to 1 over \(\ln \sum _{{\varvec{\sigma }}\in \{\pm 1\}^n}\exp (-{\mathcal H}_t({\varvec{\sigma }}))\) with \({\mathcal H}_t({\varvec{\sigma }})\equiv -\sum _{X:J_X\in S_1\cup S_2} J_X \sigma _X-t\sum _{X:J_X\in S_{12}} J_X \sigma _X\) shows that \(\ln {\mathcal {Z}}(S) = \ln {\mathcal {Z}}(S_1) + \ln {\mathcal {Z}}(S_2)+\sum _{X:J_X\in S_{12}}J_X\int _0^1dt\langle \sigma _X\rangle _t\) (using that \(S_1\) and \(S_2\) are disjoint). Then the first GKS inequality gives \(\langle \sigma _X\rangle _t\ge 0\). As \(J_X\ge 0\) too, we obtain the result.
References
Mézard, M., Parisi, G., Virasoro, M.-A.: Spin Glass Theory and Beyond. World Scientific, Singapore (1987)
Pastur, L.A., Shcherbina, M.V.: Absence of self-averaging of the order parameter in the sherrington-kirkpatrick model. J. Stat. Phys. 62(1), 1–19 (1991)
Shcherbina, M.V.: More about absence of selfaverageness of the order parameter in the sherrington-kirkpatrick model. CARR Reports in Mathematical Physics, n. 3/91, Department of Mathematics, University of Rome “La Sapienza”, (1991)
Ghirlanda, S., Guerra, F.: General properties of overlap probability distributions in disordered spin systems. towards parisi ultrametricity. J. Phys. A Math. Gen. 31(46), 9149 (1998)
Guerra, F.: An introduction to mean field spin glass theory: methods and results. Les Houches Summer School Proceedings 83, 12 (2006)
Guerra, F.: About the overlap distribution in mean field spin glass models. Int. J. Mod. Phys. B 10, 12 (2012)
Guerra, F., Toninelli, F.L.: The thermodynamic limit in mean field spin glass models. Commun. Math. Phys. 230(1), 71–79 (2002)
Guerra, F., Toninelli, F.L.: Quadratic replica coupling in the sherrington-kirkpatrick mean field spin glass model. J. Math. Phys. 43(7), 3704–3716 (2002)
Aizenman, M., Contucci, P.: On the stability of the quenched state in mean-field spin-glass models. J. Stat. Phys. 92(5), 765–783 (1998)
Chatterjee, S.: Superconcentration and Related Topics. Springer Monographs in Mathematics. Springer, New York (2014)
Talagrand, M.: Mean Field Models for Spin Glasses: Volume I: Basic Examples, vol. 54. Springer, New York (2010)
Talagrand, M.: Mean Field Models for Spin Glasses: Volume II: Advanced Replica-Symmetry and Low Temperature, vol. 55. Springer, New York (2011)
Panchenko, D.: The Sherrington-Kirkpatrick Model. Springer Monographs in Mathematics. Springer, New York (2013)
Chatterjee, S.: Absence of replica symmetry breaking in the random field ising model. Commun. Math. Phys. 337(1), 93–102 (2015)
Barra, A., De Sanctis, L.: Overlap fluctuations from the boltzmann random overlap structure. J. Math. Phys. 47(10), 103305 (2006)
Barra, A., De Sanctis, L.: Stability properties and probability distributions of multi-overlaps in dilute spin glasses. J. Stat. Mech. Theory Exp. 2007(08), P08025 (2007)
Franz, S., Leone, M., Toninelli, F.L.: Replica bounds for diluted non-poissonian spin systems. J. Phys. A 36(43), 10967 (2003)
De Sanctis, L., Franz, S.: Self-averaging identities for random spin systems. In: de Monvel, A.B., Bovier, A. (eds.) Spin Glasses: Statics and Dynamics, pp. 123–142. Birkhäuser Basel, Basel (2009)
Tanaka, T.: A statistical-mechanics approach to large-system analysis of cdma multiuser detectors. IEEE Trans. Inf. Theory 48(11), 2888–2910 (2002)
Korada, S.B., Macris, N.: Tight bounds on the capacity of binary input random cdma systems. IEEE Trans. Inf. Theory 56(11), 5590–5613 (2010)
Montanari, A.: Tight bounds for ldpc and ldgm codes under map decoding. IEEE Trans. Inf. Theory 51(9), 3221–3246 (2005)
Kudekar, S., Macris, N.: Sharp bounds for optimal decoding of low-density parity-check codes. IEEE Trans. Inf. Theory 55(10), 4635–4650 (2009)
Abbe, E.: Community detection and stochastic block models. Found. Trends® Commun. Inf. Theory 14(1–2), 1–162 (2018)
Guerra, F.: An introduction to mean field spin glass theory: methods and results. Math. Stat. Phys. 2005, 243–271 (2005)
Barbier, J., Macris, N.: The adaptive interpolation method: a simple scheme to prove replica formulas in bayesian inference. Probab. Theory Relat. Fields 174, 1133–1185 (2018)
Barbier, J., Macris, N., Miolane, L.: The layered structure of tensor estimation and its mutual information. In 55th Annual Allerton Conference on Communication, Control, and Computing (Allerton), September (2017)
Barbier, J., Krzakala, F., Macris, N., Miolane, L., Zdeborová, L.: Optimal errors and phase transitions in high-dimensional generalized linear models. Proc. Natl Acad. Sci. 116(12), 5451–5460 (2019)
Barbier, J., Macris, N., Maillard, A., Krzakala, F.: The Mutual Information in Random Linear Estimation Beyond i.i.d. Matrices. In IEEE International Symposium on Information Theory (ISIT), (2018)
Aubin, B., Maillard, A., Barbier, J., Krzakala, F., Macris, N., Zdeborová, L.: The committee machine: Computational to statistical gaps in learning a two-layers neural network. In Advances in Neural Information Processing Systems, vol. 31, pp. 3226–3237. Curran Associates, Inc., (2018)
Gabrié, M., Manoel, A., Luneau, C., Barbier, J., Macris, N., Krzakala, F., Zdeborová, L.: Entropy and mutual information in models of deep neural networks. In Advances in Neural Information Processing Systems, vol. 31, pp 1824–1834. Curran Associates, Inc. (2018)
Dembo, A., Montanari, A.: Ising models on locally tree-like graphs. Ann. Appl. Probab. 20, 565–592 (2010)
Coja-Oghlan, A., Krzakala, F., Perkins, W., Zdeborová, L.: Information-theoretic thresholds from the cavity method. Adv. Math. 333, 694–795 (2018)
Aizenman, M., Sims, R., Starr, S.L.: Extended variational principle for the Sherrington-Kirkpatrick spin-glass model. Phys. Rev. B 68(21), 214403 (2003)
Franz, S., Leone, M.: Replica bounds for optimization problems and diluted spin systems. J. Stat. Phys. 111(3), 535–564 (2003)
Panchenko, D., Talagrand, M.: Bounds for diluted mean-fields spin glass models. Prob. Theory Relat. Fields 130(3), 319–336 (2004)
Bayati, M., Gamarnik, D., Tetali, P.: Combinatorial approach to the interpolation method and scaling limits in sparse random graphs. In Proceedings of the forty-second ACM symposium on Theory of computing, ACM, Cambridge, pp 105–114 (2010)
Salez, J.: The interpolation method for random graphs with prescribed degrees. Comb. Prob. Comput. 25(3), 436–447 (2016)
Montanari, A.: Estimating random variables from random sparse observations. Eur. Trans. Telecommun. 19, 385–403 (2008)
Liggett, T.M.: T. E. Harris’ contributions to interacting particle systems and percolation. Ann. Probab. 39(03), 407–416 (2011)
Acknowledgements
This work was done while J.B. was affiliated with EPFL. J.B. and C.L.C. acknowledge funding from the Swiss National Science Foundation Grant 200021-156672. J.B. would like to particularly thank Amin Coja-Oghlan for suggesting him to work on this problem and for the very many interesting discussions they had when he visited Dr. Coja-Oghlan’s group in Francfurt, as well as Florent Krzakala for insightful comments, Adriano Barra for pointing relevant references and Silvio Franz for clarifications on some of his works. J.B. also deeply thanks Nadia Bersier for her support.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Michael Aizenman.
Dedicated to Joel Lebowitz.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Some Technicalities
1.1 Proof of the Approximation Inequality (7)
Note that
We have \(| \frac{d p_{n}(h_0, h_1, \alpha )}{d{h_0}}| = |{\mathbb {E}}\langle Q_1\rangle | \le 1\) and from (23) we also have \(| \frac{d p_n(0, h_1, \alpha )}{d\alpha }| \le h_1 n^{-(1-\theta )}\). Thus by the mean value theorem we obtain (7), i.e. \(|p_n(h_0,h_1,\alpha ) - p_n(0,0,0)| \le h_0 + \alpha h_1n^{-(1-\theta )}\).
1.2 A Property of the Poisson Distribution
Any function \(g: \mathbb {N} \rightarrow \mathbb {R}\) of a random variable \(X\sim \mathrm{Poi}(\nu )\) with Poisson distribution and mean \(\nu \), and such that both \(\mathbb {E}\,g(X)\) and \(\mathbb {E}\,g(X+1)\) exist, satisfies
1.3 Multivariate Harris Inequality
For completeness we provide here a simple proof of the multivariate version of the Harris inequality. We refer to [39] for more information.
Lemma 12
(Multivariate version of the Harris inequality) Let \(g,\tilde{g}:\mathbb {R}^n\mapsto \mathbb {R}\) be two functions of the random vector \(\mathbf{x }= (x_1, \ldots , x_n)\) where all components are independent random variables. If for all \(i \in \{1, \ldots , n\}\)g and \(\tilde{g}\) are both monotone w.r.t. \(x_i\) with same monotonicity, i.e. \(\partial _{x_i}g(\mathbf{x })\,\partial _{x_i}\tilde{g}(\mathbf{x })\ge 0 \ \forall \ i\), then \(\mathbb {E}[g(\mathbf{x })\, \tilde{g}(\mathbf{x })] - \mathbb {E}\,g(\mathbf{x }) \, \mathbb {E}\,\tilde{g}(\mathbf{x }) \ge 0\).
Proof
Let \(\mathbf{x }_i^j \equiv (x_i, x_{i+1}, \ldots , x_{j})\). The monotonicity w.r.t. \(x_1\) implies
which by expanding the product can be simplified to
The proof then proceeds by induction. Suppose
Again, the monotonicity w.r.t. \(x_i\) implies
which can be simplified to
The induction is ended by noting that with the hypothesis (46) the identity (47) can further be relaxed to
This ends the induction argument and the proof. \(\square \)
On the Concentration and Existence of the Pressure
We consider Hamiltonian (1) with independent random couplings \(J_X\), \(X\subset \{1,\ldots , n\}\) and prove the following generic result used in (2). We then discuss a simple argument and condition that guarantees the existence of the thermodynamic limit using the first GKS inequality. We verify that these results apply to Example 1.
Proposition 13
(Concentration of the pressure) Let \(J_X\), \(X\subset \{1,\ldots , n\}\) be independent random variables such that \(\sum _{X\subset \{1,\ldots , n\}} \mathrm {Var}(J_X) \le C_P n\) for some numerical constant \(C_P>0\). Then we have \(\mathbb {E}[(P_n - p_n)^2] \le C_P/n\).
Proof
The proof is a simple application of the Efron-Stein inequality. Set \(\varvec{J}\equiv (J_X, X\subset \{1,\ldots ,n\})\). Let \(\varvec{J}^{(X)}\) be a vector such that \(\varvec{J}^{(X)}\) differs from \(\varvec{J}\) only at the Xth component which becomes \(J_X^\prime \) drawn independently from the same distribution as the one of \(J_X\) (note that the random variables \(J_X\) for different X do not necessarily have the same distribution). Efron Stein’s inequality tells us that
An elementary interpolation gives
Replacing in (48) (and recalling \(p_n\equiv \mathbb {E}\, P_n\)) gives
With the hypothesis on \(\mathrm {Var}(J_X)\) the proof is complete. \(\square \)
An easy and more or less standard superadditivity argument proves that the thermodynamic limit exists for the ferromagnetic model (1). We give the argument for completeness. For simplicity we consider that there exists a maximal size \(x_\mathrm{max}\) independent of n such that \(\vert X\vert \le x_\mathrm{max}\). We suppose furthermore that all \(J_X\) are independent with a distribution that depends only on the cardinalities \(\vert X\vert \) (in other words given a cardinality they are i.i.d.) and also
where \(m({\vert X\vert }) \equiv \mathbb {E}\,J_X\) and C a positive constant independent of n.
Proposition 14
(Existence of the thermodynamic limit of the pressure) Let \(J_X, X \subset \{1,\ldots ,n\}\) be independent random variables with a probability distribution supported on \(\mathbb {R}_{\ge 0}\) depending only on \(\vert X\vert \). Moreover assume \(J_X=0\) for \(\vert X\vert > x_\mathrm{max}\) independent of n. Let (49) be satisfied. Then \(\lim _{n \rightarrow +\infty } p_n\) exists and is finite.
Proof
Fix non-zero integers \(n_1\), \(n_2\) both greater than \(x_\mathrm{max}\) and \(n \equiv n_1 + n_2\). Consider a set of realizations \(S\equiv \{J_X, X\subset \{1,\ldots ,n\}\}\). This set can be split in three disjoint sets \(S=S_1\cup S_2\cup S_{12}\) with \(S_1 \equiv \{J_X, X\subset \{1,\ldots , n_1\}\}\), \(S_2 \equiv \{J_X, X\subset \{n_1+1,\ldots , n\}\}\) and \(S_{12} \equiv \{J_X, X\cap \{1,\ldots , n_1\} \ne \emptyset , X\cap \{n_1+1,\ldots , n\} \ne \emptyset \}\). Let \(\ln {\mathcal {Z}}(S)/n\) the pressure corresponding to the Hamiltonian with couplings in S and \(\ln {\mathcal {Z}}(S_1)/n_1\) and \(\ln {\mathcal {Z}}(S_2)/n_2\) the pressures corresponding to the Hamiltonians with couplings from \(S_1\) and \(S_2\) only. One can show, using the first GKS inequality, thatFootnote 4
Then averaging over all coupling constants in S, using that they are independent with distributions depending only on the cardinality \(\vert X\vert \) and that all cardinalities are contained in S, \(S_1\) and \(S_2\), we obtain
which is equivalent to \(np_n \ge n_1 p_{n_1} + n_2 p_{n_2}\) (for \(n_1\), \(n_2\) greater than \(x_\mathrm{max}\)). This means that the function \(n\mapsto np_n\) is a superadditive sequence and therefore by Fekete’s lemma the limit \(\lim _{n\rightarrow +\infty } p_n\) equals \(\sup _n p_n\). To show that \(\sup _n p_n\) is finite note that
using \(J_X\ge 0\) and condition (49). This ends the proof. \(\square \)
Consider now Example 1 for n large and p fixed. We have \(J_X = 0\) for all subsets with cardinalities \(\vert X\vert \) different from 1 and p. For \(\vert X\vert =1\) the coupling constants \(J_X=H\) are deterministic so obviously \(\mathrm {Var}(J_X) =0\). For \(\vert X\vert =p\) the couplings \(J_X\) are independent Bernoulli variables taking value J with probability \(\gamma n \left( {\begin{array}{c}n\\ p\end{array}}\right) ^{-1}\) and 0 with complementary probability, so \(\mathrm {Var}(J_X) = J^2 \gamma n \left( {\begin{array}{c}n\\ p\end{array}}\right) ^{-1} \big ( 1 - \gamma n \left( {\begin{array}{c}n\\ p\end{array}}\right) ^{-1} \big )\). Thus
Therefore Proposition 13 applies. Similarly the condition for the existence of the thermodynamic limit of the pressure is also met because the left hand side of (49) equals
The mixed p-spin models can be treated similarly.
Rights and permissions
About this article
Cite this article
Barbier, J., Chan, C.L. & Macris, N. Concentration of Multi-overlaps for Random Dilute Ferromagnetic Spin Models. J Stat Phys 180, 534–557 (2020). https://doi.org/10.1007/s10955-019-02470-6
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10955-019-02470-6