-
On sampling diluted Spin Glasses using Glauber dynamics
Authors:
Charilaos Efthymiou,
Kostas Zampetakis
Abstract:
Spin-glasses are Gibbs distributions that have been studied in CS for many decades. Recently, they have gained renewed attention as they emerge naturally in learning, inference, optimisation etc.
We consider the Edwards-Anderson (EA) spin-glass distribution at inverse temperature $β$ when the underlying graph is an instance of $G(n,d/n)$. This is the random graph on $n$ vertices where each edge…
▽ More
Spin-glasses are Gibbs distributions that have been studied in CS for many decades. Recently, they have gained renewed attention as they emerge naturally in learning, inference, optimisation etc.
We consider the Edwards-Anderson (EA) spin-glass distribution at inverse temperature $β$ when the underlying graph is an instance of $G(n,d/n)$. This is the random graph on $n$ vertices where each edge appears independently with probability $d/n$ and $d=Θ(1)$. We study the problem of approximate sampling from this distribution using Glauber dynamics.
For a range of $β$ that depends on $d$ and for typical instances of the EA model on $G(n,d/n)$, we show that the corresponding Glauber dynamics exhibits mixing time $O(n^{2+\frac{3}{\log^2 d}})$.
The range of $β$ for which we obtain our rapid-mixing results correspond to the expected influence being $<1/d$; we conjecture that this is the best possible.
Unlike the mean-field spin-glasses, where the problem has been studied before, the diluted case has not.
We utilise the well-known path-coupling technique. In the standard Glauber dynamics on $G(n,d/n)$, one has to deal with the so-called effect of high degree vertices. Here, rather than considering degrees, it is more natural to use a different measure on the vertices called aggregate influence.
We build on the block-construction approach proposed by [Dyer et al. 2006] to circumvent the problem of high-degree vertices. Specifically, we first establish rapid mixing for an appropriately defined block-dynamics. We design this dynamics such that vertices of large aggregate influence are placed deep inside their blocks. Then, we obtain rapid mixing for the Glauber dynamics utilising a comparison argument.
△ Less
Submitted 13 March, 2024;
originally announced March 2024.
-
Broadcasting with Random Matrices
Authors:
Charilaos Efthymiou,
Kostas Zampetakis
Abstract:
Motivated by the theory of spin-glasses in physics, we study the so-called reconstruction problem for the related distributions on the tree, and on the sparse random graph $G(n,d/n)$.
Both cases, reduce naturally to studying broadcasting models on the tree, where each edge has its own broadcasting matrix, and this matrix is drawn independently from a predefined distribution. In this context, we…
▽ More
Motivated by the theory of spin-glasses in physics, we study the so-called reconstruction problem for the related distributions on the tree, and on the sparse random graph $G(n,d/n)$.
Both cases, reduce naturally to studying broadcasting models on the tree, where each edge has its own broadcasting matrix, and this matrix is drawn independently from a predefined distribution. In this context, we study the effect of the configuration at the root to that of the vertices at distance $h$, as $h\to\infty$.
We establish the reconstruction threshold for the cases where the broadcasting matrices give rise to symmetric, 2-spin Gibbs distributions. This threshold seems to be a natural extension of the well-known Kesten-Stigum bound which arises in the classic version of the reconstruction problem.
Our results imply, as a special case, the reconstruction threshold for the well-known Edward-Anderson model of spin-glasses on the tree.
Also, we extend our analysis to the setting of the Galton-Watson tree, and the random graph $G(n,d/n)$, where we establish the corresponding thresholds.Interestingly, for the Edward-Anderson model on the random graph, we show that the replica symmetry breaking phase transition, established in [Guerra and Toninelli:2004], coincides with the reconstruction threshold.
Compared to the classical Gibbs distributions, the spin-glasses have a lot of unique features. In that respect, their study calls for new ideas, e.g., we introduce novel estimators for the reconstruction problem. Furthermore, note that the main technical challenge in the analysis is the presence of (too) many levels of randomness. We manage to circumvent this problem by utilising recently proposed tools coming from the analysis of Markov chains.
△ Less
Submitted 13 September, 2023; v1 submitted 22 February, 2023;
originally announced February 2023.
-
The Lovász Local Lemma is Not About Probability
Authors:
Dimitris Achlioptas,
Kostas Zampetakis
Abstract:
Given a collection of independent events each of which has strictly positive probability, the probability that all of them occur is also strictly positive. The Lovász local lemma (LLL) asserts that this remains true if the events are not too strongly negatively correlated. The formulation of the lemma involves a graph with one vertex per event, with edges indicating potential negative dependence.…
▽ More
Given a collection of independent events each of which has strictly positive probability, the probability that all of them occur is also strictly positive. The Lovász local lemma (LLL) asserts that this remains true if the events are not too strongly negatively correlated. The formulation of the lemma involves a graph with one vertex per event, with edges indicating potential negative dependence. The word "Local" in LLL reflects that the condition for the negative correlation can be expressed solely in terms of the neighborhood of each vertex. In contrast to this local view, Shearer developed an exact criterion for the avoidance probability to be strictly positive, but it involves summing over all independent sets of the graph.
In this work we make two contributions. The first is to develop a hierarchy of increasingly powerful, increasingly non-local lemmata for bounding the avoidance probability from below, each lemma associated with a different set of walks in the graph. Already, at its second level, our hierarchy is stronger than all known local lemmata. To demonstrate its power we prove new bounds for the negative-fugacity singularity of the hard-core model on several lattices, a central problem in statistical physics.
Our second contribution is to prove that Shearer's connection between the probabilistic setting and the independent set polynomial holds for \emph{arbitrary supermodular} functions, not just probability measures. This means that all LLL machinery can be employed to bound from below an arbitrary supermodular function, based only on information regarding its value at singleton sets and partial information regarding their interactions. We show that this readily implies both the quantum LLL of Ambainis, Kempe, and Sattath~[JACM 2012], and the quantum Shearer criterion of Sattath, Morampudi, Laumann, and Moessner~[PNAS 2016].
△ Less
Submitted 16 November, 2021;
originally announced November 2021.
