arXiv++ Combinatorics

A foundational theorem of Ajtai, Komlós, Pintz, Spencer, and Szemerédi asserts that every $n$-vertex $k$-uniform uncrowded hypergraph with maximum degree $Δ$ contains an independent set of size $c_k n{\left(\frac{\log Δ}Δ\right)^{\frac{1}{k-1}}}$, for some constant $c_k>0$. Determining the optimal leading constant $c_k$ in this bound is a major open problem. A natural target is the so-called shattering-threshold constant $\left(\frac{1}{k-1}\right)^{\frac{1}{k-1}}$, which appears in the solution-space geometry of random constraint satisfaction problems, in average-case complexity theory, and in statistical physics, among other areas. We prove that uncrowded hypergraphs attain this threshold. More precisely, for every $ε>0$ and $k\geq 2$, every $n$-vertex $k$-uniform uncrowded hypergraph of sufficiently large maximum degree $Δ$ contains an independent set of size at least $(1-ε) n {\left(\frac{1}{k-1}\frac{\log Δ}Δ\right)^{\frac{1}{k-1}}}$. Consequently, we obtain the first pseudorandom class of hypergraphs whose guaranteed independence number matches the shattering threshold, resolving a folklore conjecture. Moreover, as another direct consequence, we resolve a conjecture of Verstraëte and Wilson by proving that there exists a constant $c_k=1-o_k(1)$ such that every $n$-vertex $k$-uniform linear hypergraph of maximum degree $Δ$ has independence number at least $c_k n\left(\frac{\log Δ}Δ\right)^{\frac{1}{k-1}}$. Our techniques are constructive yielding efficient algorithms for both static and distributed settings. Specifically, we provide an $\tilde O(nΔ)$-time randomized static algorithm and an $\tilde O(1)$-round randomized $\textsf{LOCAL}$ algorithm to find an independent set in uncrowded hypergraphs at the shattering threshold. These results extend seamlessly to the the setting of linear hypergraphs.

Consider the following plaquette model from statistical physics: a lamp lies at every vertex of the triangular lattice and a switch lies at every even vertex of the (bipartite) dual hexagonal lattice. Each switch toggles the three lamps on its face. The energy of a configuration is the number of ON lamps. For the Glauber dynamics associated with the Gibbs measure defined by this Hamiltonian at any inverse temperature $β>0$, we show that, in any dimension $d\ge 2$, the infinite volume relaxation time satisfies \[e^{β^2/C}/C \le T_{\mathrm{rel}}\le Ce^{e^{Cβ}}\] for some $C>0$. Our result entails that the Gibbs measure is unique. The $e^{β^2}$ scaling was conjectured by Newman and Moore in 1999 and matches the behaviour of supercritical rooted kinetically constrained models such as the East model, thus recovering fragile glass phenomenology in the absence of kinetic constraints. More precisely, we show that, on a torus of side length $2^k$, when $β\to\infty$ and $k/β\to0$, we have $T_{\mathrm{rel}}=e^{2βk(1+o(1))}$. Quite surprisingly, however, we also prove that, on non-periodic finite domains of size $n\le e^{β/C}$ for large $C>0$, we have the much larger asymptotics $\ln T_{\mathrm{rel}}=βn^{Θ(1)}$. The main ingredients of the proofs are new results in extremal and enumerative combinatorics and rely on renormalisation ideas for the dynamics and its groundstates also known as the Ledrappier subshift. We note consequences of our results to geometric group theory (more precisely to the complexity of the word problem for the Baumslag finitely presented group) and to ergodic theory.

Optimization problems such as the NP-complete 3-SAT provide an important benchmark for the difficult task of finding ground-states in strongly correlated many-body systems with rugged energy landscapes. The study of random 3-SAT problems as Ising spin Hamiltonians in statistical physics has yielded major insights including the existence of a satisfiability phase transition, and the prediction of a critical parameter line of particularly hard instances. Yet, progress on solving those instances has been scarce for several decades. Here, introducing the Target-SAT (TSAT) algorithm, we roughly triple the tractable problem sizes in the hardest regime, with an even greater improvement in a vast range of neighboring regions. By leveraging statistical information hidden in the combinatorial constraints of the problem, TSAT is actively guided in its stochastic local search toward a target within the relevant parameter space. Our analysis also explains why established local search algorithms are limited to relatively small system sizes due to a vast low-energy trap. Furthermore, we characterize the aforementioned critical line in terms of a dominant additional complexity barrier, whose exponential scaling is quickly overcome by TSAT only in the surrounding parameter space. With TSAT, the lead in solving the hardest known random satisfiability problems returns to the realm of stochastic local search algorithms.

In graph bootstrap percolation, edges of an Erdős-Rényi random graph ${\mathcal G}_{n,p}$ are initially active. Activation spreads to other edges of the complete graph $K_n$ by an iterative process governed by a fixed graph $H$, whereby an edge becomes active whenever it is the only inactive edge in a copy of $H$. If all edges of $K_n$ are eventually activated, we say the process $H$-percolates. The case $H=K_3$ corresponds to the classical sharp threshold for connectivity in ${\mathcal G}_{n,p}$. When $H=K_4$, there are close connections with $2$-neighbor bootstrap percolation from statistical physics. Varying $H$ produces a wide range of behaviors. In this work, for every graph $H$, we locate the critical $H$-percolation threshold $p_c(n,H)$, answering a question of Balogh, Bollobás, and Morris. Our general methods recover and improve several previous results. The location of $p_c(n,H)$ is related to a critical limiting density $ρ(H)$ of graphs that most efficiently activate a given edge. Introducing the parameter $ρ(H)$ raises several questions. For instance, it remains open whether $ρ(H)$ is computable in general, and its expression appears to indicate when the $H$-percolation threshold is sharp.

We introduce the Random Subsequence Model, a spin glass model on pairs of random strings $(X,Y) \in \{0,1\}^N \times \{0,1\}^M$ whose partition function counts subsequence embeddings of $Y$ into $X$. We study two variants: the null model, where $X$ and $Y$ are independent and uniform, and the planted model, where $X$ is uniform and $Y$ is a uniformly-random length-$M$ subsequence of $X$. We connect the Random Subsequence Model to longstanding problems in various fields, including the best rate achievable by uniformly-random codes in the deletion channel, the longest common subsequence problem between two random strings, and models of directed polymers in statistical physics. In the regime where $N,M\to\infty$ at a fixed ratio $α= M/N \in (0,1)$, we exhibit strict asymptotic separations between the null annealed free energy and the quenched free energies of the null and planted models at all values of the density parameter $α$. This suggests that these models are in a spin glass phase at zero temperature throughout the entire dense regime. As a consequence, we show that uniformly-random codes achieve a positive rate in the deletion channel for all deletion probabilities $p\in [0,1),$ settling multiple conjectures of the second author, Isik and Weissman (2024) and proving the first such positive rate result for the regime $p \geq 1/2$. We also give an exact analytic formula for the annealed free energy of the planted model for all values of the density parameter. This implies a corresponding analytic upper bound on the best rate achievable by uniformly-random codes in the deletion channel, complementing the lower bound from our first result. Our upper and lower bounds for the capacity of the deletion channel under uniform codes are far closer to each other than the best known upper and lower bounds for the capacity of the deletion channel.

The study of lattice walks restricted to the first quadrant has shed a lot of interest in the past twenty years. In particular, there has been an important effort to classify models of weighted walks with small steps with respect to the algebraic-differential nature of their generating function. The techniques that were developed in the course of this work are now applied to different extensions of those walks. One of these extensions, called walks with interacting boundaries, consists in accounting for the number of contacts of the walk with the axes, with motivation coming from statistical physics. These contacts are encoded as two additional parameters for the generating function, the Boltzmann weights. For one notable family of models, called genus zero models, we establish in this paper the complete classification of their generating function, for all real values of the parameters. We do this by adapting to this more general case a method due to Dreyfus, Hardouin, Roques and Singer, used in the former classification, and which consists in studying the rational solutions to a $q$-difference equation. In almost all cases, we show that the generating function is hypertranscendental, regardless of the values of the weights. In the remaining cases, we prove that specific algebraic relations between the Boltzmann weights make the generating function $\mathbb{N}$-algebraic or $\mathbb{N}$-rational, contrasting with the interaction-less case.

In statistical physics, the multivariate hard-core model describes a system of particles, each of which receives its own fugacity. In graph-theoretic language, the partition function of the model translates to the multivariate independence polynomial, i.e., the multiaffine generalisation of the independence polynomial, defined by $Z_G(λ_1,\dots,λ_n) := \sum_{I\in\mathcal{I}(G)} \prod_{v\in I}λ_v$, where $\mathcal{I}(G)$ denotes the set of all independent sets in a graph $G$ on $[n]:=\{1,2,\dots,n\}$. We prove that for every simple graph $G$ on $[n]$ and $λ_1,\dots,λ_n\geq 0$, \[ Z_G(λ_1,\dots,λ_n) \geq \prod_{i=1}^n (1+(d_i+1)λ_i)^{1/(d_i+1)}, \] where $d_1,\dots,d_n$ is the degree sequence of $G$. This generalises a result of Sah, Sawhney, Stoner, and Zhao, who proved the univariate case $λ_1=\dots=λ_n=λ$. We further conjecture that our inequality should generalise to other antiferromagnetic models and give some evidence in support of it. In particular, for $λ_i,μ_i\geq 0$, $1\leq i\leq n$, we obtain a stronger inequality \[ \sum_{\substack{I,J\in \mathcal{I}(G) \\ I\cap J=\emptyset}} \prod_{v\in I}λ_v\prod_{u\in J}μ_u \geq \prod_{i=1}^n \left(1+(d_i+1)(λ_i+μ_i)+d_i(d_i+1)λ_iμ_i\right)^{1/(d_i+1)}, \] which proves our conjecture for a multiaffine generalisation of the semiproper colouring partition function with two proper colours. Our key technical steps for both theorems are obtained by using a custom mathematical research agent built on top of Gemini Deep Think, which can be seen as a benchmark demonstrating that the current state-of-the-art language models can, in part, assist with mathematical research.

We provide precise asymptotics for the number of antichains in the poset $\{0,1,2\}^n$, answering a question of Sapozhenko. Finding improved estimates for this number was also a problem suggested by Noel, Scott, and Sudakov, who obtained asymptotics for the logarithm of the number. Key ingredients for the proof include a graph-container lemma to bound the number of expanding sets in a class of irregular graphs, isoperimetric inequalities for generalizations of the Boolean lattice, and methods from statistical physics based on the cluster expansion.

Given a graph $G$ and a target graph $H$, an $H$-coloring of $G$ is an adjacency-preserving vertex map from $G$ to $H$. The number of $H$-colorings of $G$, $\hom(G,H)$, has been studied for many classes of $G$ and $H$. In particular, extremal questions of maximizing and minimizing $\hom(G,H)$ have been considered when $H$ is a clique or $G$ is a tree. In this paper, we develop a new technique using automorphisms of $H$ to show that $\hom(T,H)$ is minimized by paths as $T$ varies over trees on a fixed number of vertices. We introduce the term Hoffman-London to refer to graphs that are minimal in this sense. In particular, we define an automorphic similarity matrix which is used to compute $\hom(T,H)$ and give matrix conditions under which $H$ is Hoffman-London. We then apply this technique to identify several families of graphs that are Hoffman-London, including loop threshold graphs and some with applications in statistical physics (e.g. the Widom-Rowlinson model). By combining our approach with a few other observations, we fully characterize the minimizing trees for all graphs $H$ on three or fewer vertices.

Two directions in algorithms and complexity involve: (1) classifying which optimization problems can be solved in polynomial time, and (2) understanding which computational problems are hard to solve \emph{on average} in addition to the worst case. For many average-case problems, there does not currently exist strong evidence via reductions that they are hard. However, we can still attempt to predict their polynomial time tractability by proving lower bounds against restricted classes of algorithms. Geometric approaches to predicting tractability typically study the \emph{optimization landscape}. For optimization problems with random objectives or constraints, ideas originating in statistical physics suggest we should study the \emph{overlap} between approximately-optimal solutions. Formally, properties of \emph{Gibbs measures} and the \emph{Franz--Parisi potential} imply lower bounds against natural local search algorithms, such as Langevin dynamics. A related theory, the \emph{Overlap Gap Property (OGP)}, proves rigorous lower bounds against classes of algorithms which are stable functions of their input. A remarkable recent work of Li and Schramm showed that the shortest path problem in random graphs admits lower bounds against a class of stable algorithms, via the OGP. Yet this problem is polynomial time tractable. We further investigate this. We find that both the OGP and the Franz--Parisi potential predict that: (1) local search will fail in the optimization landscape of shortest paths, but (2) local search should succeed in the optimization landscape for shortest path \emph{trees}, which is true. Using the Franz--Parisi potential, we explain an analogy with results from combinatorial optimization -- submodular minimization is tractable via local search on the Lovász extension, even though ``naive'' local search over sets or the multilinear extension provably fails.

A connected set in a graph is a non-empty set of vertices that induces a connected subgraph. In an infinite lattice, a connected set is often referred to as a lattice animal, whose enumeration up to isomorphism is a classical problem in both combinatorics and statistical physics. In this paper, we focus on the enumeration of connected sets in finite lattice graphs, providing a link between combinatorial counting and structural connectivity in the system. For any positive integers $m,n$, let $N(P_m\times P_n)$ and $N(C_m\times P_n)$ denote the number of all connected sets in the $(m\times n)$-lattice graph $P_m\times P_n$ and $(m\times n)$-cylindrical lattice graph $C_m\times P_n $, respectively. In 2020, Vince derived enumeration formulas for $N(P_m\times P_2)$ and $N(C_m\times P_2)$, and highlighted the increasing difficulty of extending these calculation results to larger (cylindrical) lattice graphs. Recently, the authors of this paper have developed a method based on multi-step recurrence formulas to obtain the enumeration formula for $N(P_m\times P_n)$ with $m\le 4$. In this article, we apply a similar approach to derive the enumeration formula for $N(C_m\times P_n)$ with $m\le 7$. Further, for the general case, we establish an explicit and tight lower bound on the number of connected sets in the Cartesian product graph $G\times P_n $ for any connected graph $G$, by employing the transfer matrix method on a subclass of connected sets. Based on this, we perform an asymptotic analysis on several lattice graphs and show that $O(N(P_3\times P_n))=1.6694^{3n}$, $O(N(C_4\times P_n))=1.8014^{4n}$, and $O(N(C_5\times P_n))=1.7877^{5n}$.

We study graph bootstrap percolation on the Erdős-Rényi random graph ${\mathcal G}_{n,p}$. For all $r \ge 5$, we locate the sharp $K_r$-percolation threshold $p_c \sim (γn)^{-1/λ}$, solving a problem of Balogh, Bollobás and Morris. The case $r=3$ is the classical graph connectivity threshold, and the threshold for $r=4$ was found using strong connections with the well-studied $2$-neighbor dynamics from statistical physics. When $r \ge 5$, such connections break down, and the process exhibits much richer behavior. The constants $λ=λ(r)$ and $γ=γ(r)$ in $p_c$ are determined by a class of $\left({r\choose2}-1\right)$-ary tree-like graphs, which we call $K_r$-tree witness graphs. These graphs are associated with the most efficient ways of adding a new edge in the $K_r$-dynamics, and they can be counted using the Fuss-Catalan numbers. Also, in the subcritical setting, we determine the asymptotic number of edges added to ${\mathcal G}_{n,p}$, showing that the edge density increases only by a constant factor, whose value we identify.

The independence polynomial of a graph $G$ is the generating polynomial corresponding to its independent sets of different sizes. More formally, if $a_k(G)$ denotes the number of independent sets of $G$ of size $k$ then \[I(G,z) \as \sum_{k}^{} (-1)^k a_k(G) z^k.\] The study of evaluating $I(G,z)$ has several deep connections to problems in combinatorics, complexity theory and statistical physics. Consequently, the roots of the independence polynomial have been studied in detail. In particular, many works have provided regions in the complex plane that are devoid of any roots of the polynomial. One of the first such results showed a lower bound on the absolute value of the smallest root $β(G)$ of the polynomial. Furthermore, when $G$ is connected, Goldwurm and Santini established that $β(G)$ is a simple real root of $I(G,z)$ smaller than one. An alternative proof was given by Csikvári. Both proofs do not provide a gap from $β(G)$ to the smallest absolute value amongst all the other roots of $I(G,z)$. In this paper, we quantify this gap.

Generalising the Heilman-Lieb Theorem from statistical physics, Chudnovsky and Seymour [J. Combin. Theory Ser. B, 97(3):350--357] showed that the univariate independence polynomial of any claw-free graph has all of its zeros on the negative real line. In this paper, we show that for any fixed subdivded claw $H$ and any $Δ$, there is an open set $F \subseteq \mathbb{C}$ containing $[0, \infty)$ such that the independence polynomial of any $H$-free graph of maximum degree $Δ$ has all of its zeros outside of $F$. We also show that no such result can hold when $H$ is any graph other than a subdivided claw or if we drop the maximum degree condition. We also establish zero-free regions for the multivariate independence polynomial of $H$-free graphs of bounded degree when $H$ is a subdivided claw. The statements of these results are more subtle, but are again best possible in various senses.

The q-state Potts model is a fundamental framework in statistical physics and graph theory, with its partition function encoding rich information about spin configurations. The multivariate Tutte polynomial (known as the partition function of the Potts model) can be defined on an arbitrary finite graph $G$ and encodes a lot of important combinatorial information about the graph. As a special case, it contains the familiar Tutte polynomial with two variables and, consequently, its specialization with one variable, such as the chromatic polynomial, the flow polynomial and the reliability polynomial. The main goal of this paper is to present an efficient algorithm for computing the Potts model partition function on SP-graphs (series-parallel graphs) with arbitrary weights. The algorithm for SP-graphs is based on simplifying the graph by replacing several edges with a single edge of equivalent weight, which significantly reduces computational complexity. In this paper, we present a linear-time algorithm for exactly computing the Potts model partition function on series-parallel graphs (SP-graphs).

The robustness of properties of a statistical physics model to slight perturbations in the exact local interactions of the model is a very relevant philosophical question, considering real-life measurements on which we base some models can only ever reach a finite precision. In this article, we will discuss this topic in a formal mathematical setting, and notably exhibit a family of models for which the low-temperature behaviour is highly non-robust.

In the context of random regular graphs, the size of the maximum cut is probably the second most studied graph parameter after the independence ratio. Zdeborová and Boettcher used the cavity method, a non-rigorous statistical physics technique, to predict one-step replica symmetry breaking (1-RSB) formulas. Coja-Ohglan et al. confirmed these predictions as rigorous upper bounds using the interpolation method. While these upper bounds were not expected to be exact, they may be very close to the true values. In this paper, we establish 2-RSB upper bounds and fine-tune their parameters to beat the aforementioned 1-RSB bounds.

An edge-weighted graph $G$, possibly with loops, is said to be antiferromagnetic if it has nonnegative weights and at most one positive eigenvalue, counting multiplicities. The number of graph homomorphisms from a graph $H$ to an antiferromagnetic graph $G$ generalises various important parameters in graph theory, including the number of independent sets and proper vertex-colourings, as well as their relaxations in statistical physics. We obtain homomorphism inequalities for various graphs $H$ and antiferromagnetic graphs~$G$ of the form \[ \lvert\operatorname{Hom}(H,G)\rvert^2 \leq \lvert\operatorname{Hom}(H\times K_2,G)\rvert, \] where $H\times K_2$ denotes the tensor product of $H$ and $K_2$. Firstly, we show that the inequality holds for any $H$ obtained by blowing up vertices of a bipartite graph into complete graphs and any antiferromagnetic $G$. In particular, one can take $H=K_{d+1}$, which already implies a new result for the Sah--Sawhney--Stoner--Zhao conjecture on the maximum number of $d$-regular graphs in antiferromagnetic graphs. Secondly, the inequality also holds for $G=K_q$ and those $H$ obtained by blowing up vertices of a bipartite graph into complete multipartite graphs, paths or even cycles. Both results can be seen as the first progress towards Zhao's conjecture on $q$-colourings, which states that the inequality holds for any $H$ and $G=K_q$, after his own work. Our method leverages on the emerging theory of Lorentzian polynomials due to Brändén and Huh and log-concavity of the list colourings of bipartite graphs, which may be of independent interest.

Model counting is a fundamental problem that consists of determining the number of satisfying assignments for a given Boolean formula. The weighted variant, which computes the weighted sum of satisfying assignments, has extensive applications in probabilistic reasoning, network reliability, statistical physics, and formal verification. A common approach for solving weighted model counting is to reduce it to unweighted model counting, which raises an important question: {\em What is the minimum number of terms (or clauses) required to construct a DNF (or CNF) formula with exactly $k$ satisfying assignments?} In this paper, we establish both upper and lower bounds on this question. We prove that for any natural number $k$, one can construct a monotone DNF formula with exactly $k$ satisfying assignments using at most $O(\sqrt{\log k}\log\log k)$ terms. This construction represents the first $o(\log k)$ upper bound for this problem. We complement this result by showing that there exist infinitely many values of $k$ for which any DNF or CNF representation requires at least $Ω(\log\log k)$ terms or clauses. These results have significant implications for the efficiency of model counting algorithms based on formula transformations.

We present two new contributions to the study of the independence polynomial $Z_G(z)$ of a finite simple graph $G = (V,E)$. First, we provide an improved lower bound for the zero-free region of $Z_G(z)$ for the important class of claw-free graphs. Our bound exceeds the classical Shearer radius and it is derived through a refined application of the Fernández-Procacci criterion using properties of the local neighborhood structure in claw-free graphs. Second, we establish a novel combinatorial expression for $Z_G(z)$, inspired by the connection with the abstract polymer gas models in statistical mechanics, which offers a new structural interpretation of the polynomial and may be of independent interest. These results strengthen the connection between statistical physics, combinatorics, and graph theory, and suggest new approaches for analytic exploration.