arXiv subjects
Publications and source records attributed to David Galvin.
In this expository note we describe a proof due to A. Sapozhenko that the number of independent sets in the discrete $d$-dimensional hypercube $Q_d$ is asymptotically $2 \sqrt{e} 2^{2^{d-1}}$ as $d$ tends to infinity.
The hard-core model has attracted much attention across several disciplines, representing lattice gases in statistical physics and independent sets in discrete mathematics and computer science. On finite graphs, we are given a parameter $λ$, and an independent set $I$ arises with probability proportional to $λ^{|I|}$. On infinite graphs a Gibbs measure is defined as a suitable limit with the correct conditional probabilities, and we are interested in determining when this limit is unique and when there is phase coexistence, i.e., existence of multiple Gibbs measures. It has long been conjectured that on ${\mathbb Z}^2$ this model has a critical value $λ_c \approx 3.796$ with the property that if $λ< λ_c$ then it exhibits uniqueness of phase, while if $λ> λ_c$ then there is phase coexistence. Much of the work to date on this problem has focused on the regime of uniqueness, with the state of the art being recent work of Sinclair, Srivastava, Štefankovič and Yin showing that there is a unique Gibbs measure for all $λ< 2.538$. Here we give the first non-trivial result in the other direction, showing that there are multiple Gibbs measures for all $λ> 5.3506$. There is some potential for lowering this bound, but with the methods we are using we cannot hope to replace $5.3506$ with anything below about $4.8771$. Our proof begins along the lines of the standard Peierls argument, but we add two innovations. First, following ideas of Kotecký and Randall, we construct an event that distinguishes two boundary conditions and always has long contours associated with it, obviating the need to accurately enumerate short contours. Second, we obtain improved bounds on the number of contours by relating them to a new class of self-avoiding walks on an oriented version of ${\mathbb Z}^2$.
Given $R \subseteq \mathbb{N}$ let ${n \brace k}_R$, ${n \brack k}_R$, and $L(n,k)_R$ be the number of ways of partitioning the set $[n]$ into $k$ non-empty subsets, cycles and lists, respectively, with each block having cardinality in $R$. We refer to these as the $R$-restricted Stirling numbers of the second and first kind and the $R$-restricted Lah numbers, respectively. Note that the classical Stirling numbers of the second kind and first kind, and Lah numbers are ${n \brace k} = {n \brace k}_{\mathbb{N}}$, ${n \brack k} = {n \brack k}_{\mathbb{N}} $ and $L(n,k) = L(n,k)_{\mathbb{N}}$, respectively. The matrices $[{n \brace k}]_{n,k \geq 1}$, $[{n \brack k}]_{n,k \geq 1}$ and $[L(n,k)]_{n,k \geq 1}$ have inverses $[(-1)^{n-k}{n \brack k}]_{n,k \geq 1}$, $[(-1)^{n-k} {n \brace k}]_{n,k \geq 1}$ and $[(-1)^{n-k} L(n,k)]_{n,k \geq 1}$ respectively. The inverse matrices $[{n \brace k}_R]^{-1}_{n,k \geq 1}$, $[{n \brack k}_R]^{-1}_{n,k \geq 1}$ and $[L(n,k)_R]^{-1}_{n,k \geq 1}$ exist if and only if $1 \in R$. We express each entry of each of these matrices as the difference between the cardinalities of two explicitly defined families of labeled forests. In particular the entries of $[{n \brace k}_{[r]}]^{-1}_{n,k \geq 1}$ have combinatorial interpretations, affirmatively answering a question of Choi, Long, Ng and Smith from 2006. If $1,2 \in R$ and if for all $n \in R$ with $n$ odd and $n \geq 3$, we have $n \pm 1 \in R$, we additionally show that each entry of $[{n \brace k}_R]^{-1}_{n,k \geq 1}$, $[{n \brack k}_R]^{-1}_{n,k \geq 1}$ and $[L(n,k)_R]^{-1}_{n,k \geq 1}$ is up to an explicit sign the cardinality of a single explicitly defined family of labeled forests. Our results also provide combinatorial interpretations of the $k$th Whitney numbers of the first and second kinds of $Π_n^{1,d}$, the poset of partitions of $[n]$ that have each part size congruent to $1$ mod $d$.
The game of plates and olives, introduced by Nicolaescu, begins with an empty table. At each step either an empty plate is put down, an olive is put down on a plate, an olive is removed, an empty plate is removed, or the olives on two plates that both have olives on them are combined on one of the two plates, with the other plate removed. Plates are indistinguishable from one another, as are olives, and there is an inexhaustible supply of each. The game derives from the consideration of Morse functions on the $2$-sphere. Specifically, the number of topological equivalence classes of excellent Morse functions on the $2$-sphere that have order $n$ (that is, that have $2n+2$ critical points) is the same as the number of ways of returning to an empty table for the first time after exactly $2n+2$ steps. We call this number $M_n$. Nicolaescu gave the lower bound $M_n \geq (2n-1)!! = (2/e)^{n+o(n)}n^n$ and speculated that $\log M_n \sim n\log n$. In this note we confirm this speculation, showing that $M_n \leq (4/e)^{n+o(n)}n^n$.
For a tree $T$, let $i_T(t)$ be the number of independent sets of size $t$ in $T$. It is an open question, raised by Alavi, Malde, Schwenk and Erdős, whether the sequence $(i_T(t))_{t \geq 0}$ is always unimodal. Here we answer the question in the affirmative for some recursively defined families of trees, specifically paths with auxiliary trees dropped from the vertices in a periodic manner. In particular, extending a result of Wang and B.-X. Zhu, we show unimodality of the independent set sequence of a path on $2n$ vertices with $\ell_1$ and $\ell_2$ pendant edges dropped alternately from the vertices of the path, $\ell_1, \ell_2$ arbitrary. We also show that the independent set sequence of any tree becomes unimodal if sufficiently many pendant edges are dropped from any single vertex, or if $k$ pendant edges are dropped from every vertex, for sufficiently large $k$. This in particular implies the unimodality of the independent set sequence of some non-periodic caterpillars.
For graphs $G$ and $H$, an $H$-coloring of $G$ is an adjacency preserving map from the vertices of $G$ to the vertices of $H$. $H$-colorings generalize such notions as independent sets and proper colorings in graphs. There has been much recent research on the extremal question of finding the graph(s) among a fixed family that maximize or minimize the number of $H$-colorings. In this paper, we prove several results in this area. First, we find a class of graphs ${\mathcal H}$ with the property that for each $H \in {\mathcal H}$, the $n$-vertex tree that minimizes the number of $H$-colorings is the path $P_n$. We then present a new proof of a theorem of Sidorenko, valid for large $n$, that for every $H$ the star $K_{1,n-1}$ is the $n$-vertex tree that maximizes the number of $H$-colorings. Our proof uses a stability technique which we also use to show that for any non-regular $H$ (and certain regular $H$) the complete bipartite graph $K_{2,n-2}$ maximizes the number of $H$-colorings of $n$-vertex $2$-connected graphs. Finally, we show that the cycle $C_n$ maximizes the number of proper colorings of $n$-vertex $2$-connected graphs.
Let $w$ be a word in alphabet $\{x,D\}$ with $m$ $x$'s and $n$ $D$'s. Interpreting "$x$" as multiplication by $x$, and "$D$" as differentiation with respect to $x$, the identity $wf(x) = x^{m-n}\sum_k S_w(k) x^k D^k f(x)$, valid for any smooth function $f(x)$, defines a sequence $(S_w(k))_k$, the terms of which we refer to as the {\em Stirling numbers (of the second kind)} of $w$. The nomenclature comes from the fact that when $w=(xD)^n$, we have $S_w(k)={n \brace k}$, the ordinary Stirling number of the second kind. Explicit expressions for, and identities satisfied by, the $S_w(k)$ have been obtained by numerous authors, and combinatorial interpretations have been presented. Here we provide a new combinatorial interpretation that retains the spirit of the familiar interpretation of ${n \brace k}$ as a count of partitions. Specifically, we associate to each $w$ a quasi-threshold graph $G_w$, and we show that $S_w(k)$ enumerates partitions of the vertex set of $G_w$ into classes that do not span an edge of $G_w$. We also discuss some relatives of, and consequences of, our interpretation, including $q$-analogs and bijections between families of labelled forests and sets of restricted partitions.
We explain the notion of the {\em entropy} of a discrete random variable, and derive some of its basic properties. We then show through examples how entropy can be useful as a combinatorial enumeration tool. We end with a few open questions.
A distance graph is an undirected graph on the integers where two integers are adjacent if their difference is in a prescribed distance set. The independence ratio of a distance graph $G$ is the maximum density of an independent set in $G$. Lih, Liu, and Zhu [Star extremal circulant graphs, SIAM J. Discrete Math. 12 (1999) 491--499] showed that the independence ratio is equal to the inverse of the fractional chromatic number, thus relating the concept to the well studied question of finding the chromatic number of distance graphs. We prove that the independence ratio of a distance graph is achieved by a periodic set, and we present a framework for discharging arguments to demonstrate upper bounds on the independence ratio. With these tools, we determine the exact independence ratio for several infinite families of distance sets of size three, determine asymptotic values for others, and present several conjectures.
For a simple finite graph G denote by {G \brace k} the number of ways of partitioning the vertex set of G into k non-empty independent sets (that is, into classes that span no edges of G). If E_n is the graph on n vertices with no edges then {E_n \brace k} coincides with {n \brace k}, the ordinary Stirling number of the second kind, and so we refer to {G \brace k} as a graph Stirling number. Harper showed that the sequence of Stirling numbers of the second kind, and thus the graph Stirling sequence of E_n, is asymptotically normal --- essentially, as n grows, the histogram of ({E_n \brace k})_{k \geq 0}, suitably normalized, approaches the density function of the standard normal distribution. In light of Harper's result, it is natural to ask for which sequences (G_n)_{n \geq 0} of graphs is there asymptotic normality of ({G_n \brace k})_{k \geq 0}. Do and Galvin conjectured that if for each n, G_n is acylic and has n vertices, then asymptotic normality occurs, and they gave a proof under the added condition that G_n has no more than o(\sqrt{n/\log n}) components. Here we settle Do and Galvin's conjecture in the affirmative, and significantly extend it, replacing "acyclic" in their conjecture with "co-chromatic with a quasi-threshold graph, and with negligible chromatic number". Our proof combines old work of Navon and recent work of Engbers, Galvin and Hilyard on the normal order problem in a Weyl algebra, and work of Kahn on the matching polynomial of a graph.
In the hard-core model on a finite graph we are given a parameter lambda>0, and an independent set I arises with probability proportional to lambda^|I|. On infinite graphs a Gibbs distribution is defined as a suitable limit with the correct conditional probabilities. In the infinite setting we are interested in determining when this limit is unique and when there is phase coexistence, i.e., existence of multiple Gibbs states. On finite graphs we are interested in determining the mixing time of local Markov chains. On Z^2 it is conjectured that these problems are related and that both undergo a phase transition at some critical point lambda_c approx 3.79. For phase coexistence, much of the work to date has focused on the regime of uniqueness, with the best result being recent work of Restrepo et al. showing that there is a unique Gibbs state for all lambda < 2.3882. Here we give the first non-trivial result in the other direction, showing that there are multiple Gibbs states for all lambda > 5.3646. Our proof adds two significant innovations to the standard Peierls argument. First, building on the idea of fault lines introduced by Randall, we construct an event that distinguishes two boundary conditions and always has long contours associated with it, obviating the need to accurately enumerate short contours. Second, we obtain vastly improved bounds on the number of contours by relating them to a new class of self-avoiding walks on an oriented version of Z^2. We extend our characterization of fault lines to show that local Markov chains will mix slowly when lambda > 5.3646 on lattice regions with periodic (toroidal) boundary conditions and when lambda > 7.1031 with non-periodic (free) boundary conditions. The arguments here rely on a careful analysis that relates contours to taxi walks and represent a sevenfold improvement to the previously best known values of λ.
We show that for all sufficiently large d, the uniform proper 3-coloring model (in physics called the 3-state antiferromagnetic Potts model at zero temperature) on Z^d admits multiple maximal-entropy Gibbs measures. This is a consequence of the following combinatorial result: if a proper 3-coloring is chosen uniformly from a box in Z^d, conditioned on color 0 being given to all the vertices on the boundary of the box which are at an odd distance from a fixed vertex v in the box, then the probability that v gets color 0 is exponentially small in d. The proof proceeds through an analysis of a certain type of cutset separating v from the boundary of the box, and builds on techniques developed by Galvin and Kahn in their proof of phase transition in the hard-core model on Z^d. Building further on these techniques, we study local Markov chains for sampling proper 3-colorings of the discrete torus Z^d_n. We show that there is a constant ρ\approx 0.22 such that for all even n \geq 4 and d sufficiently large, if M is a Markov chain on the set of proper 3-colorings of Z^d_n that updates the color of at most ρn^d vertices at each step and whose stationary distribution is uniform, then the mixing time of M (the time taken for M to reach a distribution that is close to uniform, starting from an arbitrary coloring) is essentially exponential in n^{d-1}.
Let $i_t(G)$ be the number of independent sets of size $t$ in a graph $G$. Alavi, Erdős, Malde and Schwenk made the conjecture that if $G$ is a tree then the independent set sequence $\{i_t(G)\}_{t\geq 0}$ of $G$ is unimodal; Levit and Mandrescu further conjectured that this should hold for all bipartite $G$. We consider the independent set sequence of finite regular bipartite graphs, and graphs obtained from these by percolation (independent deletion of edges). Using bounds on the independent set polynomial $P(G,λ):=\sum_{t \geq 0} i_t(G)λ^t$ for these graphs, we obtain partial unimodality results in these cases. We then focus on the discrete hypercube $Q_d$, the graph on vertex set $\{0,1\}^d$ with two strings adjacent if they differ on exactly one coordinate. We obtain asymptotically tight estimates for $i_{t(d)}(Q_d)$ in the range $t(d)/2^{d-1} > 1-1/\sqrt{2}$, and nearly matching upper and lower bounds otherwise. We use these estimates to obtain a stronger partial unimodality result for the independent set sequence of $Q_d$.
For a graph $G$ and a positive integer $k$, the {\em graphical Stirling number} $S(G,k)$ is the number of partitions of the vertex set of $G$ into $k$ non-empty independent sets. Equivalently it is the number of proper colorings of $G$ that use exactly $k$ colors, with two colorings identified if they differ only on the names of the colors. If $G$ is the empty graph on $n$ vertices then $S(G,k)$ reduces to $S(n,k)$, the familiar Stirling number of the second kind. In this note we first consider Stirling numbers of forests. We show that if $(F^{c(n)}_n)_{n\geq 0}$ is any sequence of forests with $F^{c(n)}_n$ having $n$ vertices and $c(n)=o(\sqrt{n/\log n})$ components, and if $X^{c(n)}_n$ is a random variable that takes value $k$ with probability proportional to $S(F^{c(n)}_n,k)$ (that is, $X^{c(n)}_n$ is the number of classes in a uniformly chosen partition of $F^{c(n)}_n$ into non-empty independent sets), then $X^{c(n)}_n$ is asymptotically normal, meaning that suitably normalized it tends in distribution to the standard normal. This generalizes a seminal result of Harper on the ordinary Stirling numbers. Along the way we give recurrences for calculating the generating functions of the sequences $(S(F^c_n,k))_{k \geq 0}$, show that these functions have all real zeroes, and exhibit three different interlacing patterns between the zeroes of pairs of consecutive generating functions. We next consider Stirling numbers of cycles. We establish asymptotic normality for the number of classes in a uniformly chosen partition of $C_n$ (the cycle on $n$ vertices) into non-empty independent sets. We give a recurrence for calculating the generating function of the sequence $(S(C_n,k))_{k \geq 0}$, and use this to give a direct proof of a log-concavity result that had previously only been arrived at in a very indirect way.
For graphs $G$ and $H$, an {\em $H$-colouring} of $G$ (or {\em homomorphism} from $G$ to $H$) is a function from the vertices of $G$ to the vertices of $H$ that preserves adjacency. $H$-colourings generalize such graph theory notions as proper colourings and independent sets. For a given $H$, $k \in V(H)$ and $G$ we consider the proportion of vertices of $G$ that get mapped to $k$ in a uniformly chosen $H$-colouring of $G$. Our main result concerns this quantity when $G$ is regular and bipartite. We find numbers $0 \leq a^-(k) \leq a^+(k) \leq 1$ with the property that for all such $G$, with high probability the proportion is between $a^-(k)$ and $a^+(k)$, and we give examples where these extremes are achieved. For many $H$ we have $a^-(k) = a^+(k)$ for all $k$ and so in these cases we obtain a quite precise description of the almost sure appearance of a randomly chosen $H$-colouring. As a corollary, we show that in a uniform proper $q$-colouring of a regular bipartite graph, if $q$ is even then with high probability every colour appears on a proportion close to $1/q$ of the vertices, while if $q$ is odd then with high probability every colour appears on at least a proportion close to $1/(q+1)$ of the vertices and at most a proportion close to $1/(q-1)$ of the vertices. Our results generalize to natural models of weighted $H$-colourings, and also to bipartite graphs which are sufficiently close to regular. As an application of this latter extension we describe the typical structure of $H$-colourings of graphs which are obtained from $n$-regular bipartite graphs by percolation, and we show that $p=1/n$ is a threshold function across which the typical structure changes. The approach is through entropy, and extends work of J. Kahn, who considered the size of a randomly chosen independent set of a regular bipartite graph.
For graphs $G$ and $H$, an $H$-coloring of $G$ is a function from the vertices of $G$ to the vertices of $H$ that preserves adjacency. $H$-colorings encode graph theory notions such as independent sets and proper colorings, and are a natural setting for the study of hard-constraint models in statistical physics. We study the set of $H$-colorings of the even discrete torus ${\mathbb Z}^d_m$, the graph on vertex set ${0, ..., m-1}^d$ ($m$ even) with two strings adjacent if they differ by 1 (mod $m$) on one coordinate and agree on all others. This is a bipartite graph, with bipartition classes ${\mathcal E}$ and ${\mathcal O}$. In the case $m=2$ the even discrete torus is the discrete hypercube or Hamming cube $Q_d$, the usual nearest neighbor graph on ${0,1}^d$. We obtain, for any $H$ and fixed $m$, a structural characterization of the space of $H$-colorings of ${\mathbb Z}^d_m$. We show that it may be partitioned into an exceptional subset of negligible size (as $d$ grows) and a collection of subsets indexed by certain pairs $(A,B) \in V(H)^2$, with each $H$-coloring in the subset indexed by $(A,B)$ having all but a vanishing proportion of vertices from ${\mathcal E}$ mapped to vertices from $A$, and all but a vanishing proportion of vertices from ${\mathcal O}$ mapped to vertices from $B$. This implies a long-range correlation phenomenon for uniformly chosen $H$-colorings of ${\mathbb Z}^d_m$ with $m$ fixed and $d$ growing. Our proof proceeds through an analysis of the entropy of a uniformly chosen $H$-coloring, and extends an approach of Kahn, who had considered the special case of $m=2$ and $H$ a doubly infinite path. All our results generalize to a natural weighted model of $H$-colorings.
For graphs $G$ and $H$, a {\em homomorphism} from $G$ to $H$, or {\em $H$-coloring} of $G$, is an adjacency preserving map from the vertex set of $G$ to the vertex set of $H$. Writing ${\rm hom}(G,H)$ for the number of $H$-colorings admitted by $G$, we conjecture that for any simple finite graph $H$ (perhaps with loops) and any simple finite $n$-vertex, $d$-regular, loopless graph $G$ we have $$ {\rm hom}(G,H) \leq \max{{\rm hom}(K_{d,d},H)^{\frac{n}{2d}}, {\rm hom}(K_{d+1},H)^{\frac{n}{d+1}}} $$ where $K_{d,d}$ is the complete bipartite graph with $d$ vertices in each partition class, and $K_{d+1}$ is the complete graph on $d+1$ vertices. Results of Zhao confirm this conjecture for some choices of $H$ for which the maximum is achieved by ${\rm hom}(K_{d,d},H)^{n/2d}$. Here we exhibit infinitely many non-trivial triples $(n,d,H)$ for which the conjecture is true and for which the maximum is achieved by ${\rm hom}(K_{d+1},H)^{n/(d+1)}$. We also give sharp estimates for ${\rm hom}(K_{d,d},H)$ and ${\rm hom}(K_{d+1},H)$ in terms of some structural parameters of $H$. This allows us to characterize those $H$ for which ${\rm hom}(K_{d,d},H)^{1/2d}$ is eventually (for all sufficiently large $d$) larger than ${\rm hom}(K_{d+1},H)^{1/(d+1)}$ and those for which it is eventually smaller, and to show that this dichotomy covers all non-trivial $H$. Our estimates also allow us to obtain asymptotic evidence for the conjecture in the following form. For fixed $H$, for all $d$-regular $G$ we have $$ {\rm hom}(G,H)^{\frac{1}{|V(G)|}} \leq (1+o(1))\max{{\rm hom}(K_{d,d},H)^{\frac{1}{2d}}, {\rm hom}(K_{d+1},H)^{\frac{1}{d+1}}} $$ where $o(1)\rightarrow 0$ as $d \rightarrow \infty$. More precise results are obtained in some special cases.
It is shown that the hard-core model on ${\bf Z}^d$ exhibits a phase transition at activities above some function $λ(d)$ which tends to zero as $d\rightarrow \infty$