arXiv++ Combinatorics

Given a digraph $D$, we denote by $\vecα(D)$ the maximum size of an acyclic set of $D$ (i.e. a set of vertices which induces a subdigraph with no directed cycles), and by $\vecχ(D)$ the minimum number of acyclic sets into which $V(D)$ can be partitioned. In this paper, we study $\vecα(D)$ and $\vecχ(D)$ from various perspectives, including restrictions on degrees and cycle lengths. A main result is that, if $D$ is a random $r$-regular digon-free simple digraph of order $n$, then $\vecα(D) = Θ(n \log r /r)$ with high probability. This corresponds to a result of Spencer and Subramanian on the Erdős--Rényi random digraph model. Along the way, we derive some related results and propose some conjectures. An example of this is an analogue of the theorem of Bondy which bounds the chromatic number of a graph by the circumference of any strong orientation.

Edge expansion is a parameter indicating how well-connected a graph is. It is useful for designing robust networks, analysing random walks or information flow through a network and is an important notion in theoretical computer science. Modularity is a measure of how well a graph can be partitioned into communities and is widely used in clustering applications. We study these two parameters in two commonly considered models of random preferential attachment graphs, with $h \geq 2$ edges added per step. We establish new bounds for the likely edge expansion for both random models. Using bounds for edge expansion of small subsets of vertices, we derive new upper bounds also for the modularity values for small $h$.

This work will appear as a chapter in a forthcoming volume titled `Topics in Probabilistic Graph Theory'. For a given graph $G$, each partition of the vertices has a modularity score, with higher values indicating that the partition better captures community structure in $G$. The modularity $q^*(G)$ of $G$ is the maximum over all vertex-partitions of the modularity score, and satisfies $0\leq q^*(G)< 1$. Modularity lies at the heart of the most popular algorithms for community detection. In this chapter we discuss the behaviour of the modularity of various kinds of random graphs, starting with the binomial random graph $G_{n,p}$ with $n$ vertices and edge-probability $p$.

We relate two important notions in graph theory: expanders which are highly connected graphs, and modularity a parameter of a graph that is primarily used in community detection. More precisely, we show that a graph having modularity bounded below 1 is equivalent to it having a large subgraph which is an expander. We further show that a connected component $H$ will be split in an optimal partition of the host graph $G$ if and only if the relative size of $H$ in $G$ is greater than an expansion constant of $H$. This is a further exploration of the resolution limit known for modularity, and indeed recovers the bound that a connected component $H$ in the host graph~$G$ will not be split if~$e(H)<\sqrt{2e(G)}$.

It is known that complete graphs and complete multipartite graphs have modularity zero. We show that the least number of edges we may delete from the complete graph $K_n$ to obtain a graph with non-zero modularity is $\lfloor n/2\rfloor +1$. Similarly we determine the least number of edges we may delete from or add to a complete bipartite graph to reach non-zero modularity. We give some corresponding results for complete multipartite graphs, and a short proof that complete multipartite graphs have modularity zero. We also analyse the modularity of very dense random graphs, and in particular we find that there is a transition to modularity zero when the average degree of the complementary graph drops below 1.

Given a class $\mathcal G$ of graphs, let ${\mathcal G}_n$ denote the set of graphs in $\mathcal G$ on vertex set $[n]$. For certain classes $\mathcal G$, we are interested in the asymptotic behaviour of a random graph $R_n$ sampled uniformly from ${\mathcal G}_n$. Call $\mathcal G$ smooth if $ n |{\mathcal G}_{n-1}| / |{\mathcal G}_n|$ tends to a limit as $n \to \infty$. Showing that a graph class is smooth is a key step in an approach to investigating properties of $R_n$, in particular the asymptotic probability that $R_n$ is connected, and more generally the asymptotic behaviour of the fragment of $R_n$ outside the largest component. The composition method of Bender, Canfield and Richmond shows that the class of graphs embeddable in a given surface is smooth; and similarly we have smoothness for any minor-closed class of graphs with 2-connected excluded minors. Here we develop the approach further, and give results encompassing both these cases and much more. We see that, under quite general conditions, our graph classes are smooth and we can describe for example the limiting distribution of the fragment of $R_n$ and the size of the core; and we obtain similar results for the graphs in the class with minimum degree at least 2.

Suppose that there is an unknown underlying graph $G$ on a large vertex set, and we can test only a proportion of the possible edges to check whether they are present in $G$. If $G$ has high modularity, is the observed graph $G'$ likely to have high modularity? We see that this is indeed the case under a mild condition, in a natural model where we test edges at random. We find that $q^*(G') \geq q^*(G)-\varepsilon$ with probability at least $1-\varepsilon$, as long as the expected number edges in $G'$ is large enough. Similarly, $q^*(G') \leq q^*(G)+\varepsilon$ with probability at least $1-\varepsilon$, under the stronger condition that the expected average degree in $G'$ is large enough. Further, under this stronger condition, finding a good partition for $G'$ helps us to find a good partition for $G$. We also consider the vertex sampling model for partially observing the underlying graph: we find that for dense underlying graphs we may estimate the modularity by sampling constantly many vertices and observing the corresponding induced subgraph, but this does not hold for underlying graphs with a subquadratic number of edges. Finally we deduce some related results, for example showing that under-sampling tends to lead to overestimation of modularity.

Given a `genus' function $g=g(n)$, we let $\mathcal{E}^g$ be the class of all graphs $G$ such that if $G$ has order $n$ (that is, has $n$ vertices) then it is embeddable in a surface of Euler genus at most $g(n)$. Let the random graph $R_n$ be sampled uniformly from the graphs in $\mathcal{E}^g$ on vertex set $[n]=\{1,\ldots,n\}$. Observe that if $g(n)$ is 0 then $R_n$ is a random planar graph, and if $g(n)$ is sufficiently large then $R_n$ is a binomial random graph $G(n,\tfrac12)$. We investigate typical properties of $R_n$. We find that for \emph{every} genus function $g$, with high probability at most one component of $R_n$ is non-planar. In contrast, we find a transition for example for connectivity: if $g$ is non-decreasing and $g(n) = O(n/\log n)$ then $\liminf_{n \to \infty} \mathbb{P}(R_n \mbox{ is connected}) < 1$, and if $g(n) \gg n$ then with high probability $R_n$ is connected. These results also hold when we consider orientable and non-orientable surfaces separately. We also investigate random graphs sampled uniformly from the `hereditary part' or the `minor-closed' part of $\mathcal{E}^g$, and briefly consider corresponding results for unlabelled graphs.

We consider random graphs sampled uniformly from a structured class of graphs, such as the class of graphs embeddable in a given surface. We sharpen and extend earlier results on pendant appearances, concerning for example numbers of leaves; and obtain results on the asymptotic distribution of components other than the giant component, under quite general conditions.

Given a function $g=g(n)$ we let ${\mathcal E}^g$ be the class of all graphs $G$ such that if $G$ has order $n$ (that is, has $n$ vertices) then it is embeddable in some surface of Euler genus at most $g(n)$, and let ${\widetilde{\mathcal E}}^g$ be the corresponding class of unlabelled graphs. We give estimates of the sizes of these classes. For example we show that if $g(n)=o(n/\log^3n)$ then the class ${\mathcal E}^{g}$ has growth constant $γ_{\mathcal P}$, the (labelled) planar graph growth constant; and when $g(n) = O(n)$ we estimate the number of n-vertex graphs in ${\mathcal E}^{g}$ and ${\widetilde{\mathcal E}}^g$ up to a factor exponential in $n$. From these estimates we see that, if ${\mathcal E}^g$ has growth constant $γ_{\mathcal P}$ then we must have $g(n)=o(n/\log n)$, and the generating functions for ${\mathcal E}^g$ and ${\widetilde{\mathcal E}}^g$ have strictly positive radius of convergence if and only if $g(n)=O(n/\log n)$. Such results also hold when we consider orientable and non-orientable surfaces separately. We also investigate related classes of graphs where we insist that, as well as the graph itself, each subgraph is appropriately embeddable (according to its number of vertices); and classes of graphs where we insist that each minor is appropriately embeddable. In a companion paper [43], these results are used to investigate random $n$-vertex graphs sampled uniformly from ${\mathcal E}^g$ or from similar classes.

Let the class A of graphs be bridge-addable; that is, whenever a graph G in A has vertices u and v in different components then the graph G+uv is in A. For a random graph sampled uniformly from the graphs in A on vertex set {1,..,n}, there are known lower bounds on the probability of being connected (for example, the probability is always at least 1/e). We ask here about similar results when the random graph is sampled uniformly from the unlabelled n-vertex graphs in A.

For a given graph $G$, each partition of the vertices has a modularity score, with higher values indicating that the partition better captures community structure in $G$. The modularity $q^*(G)$ of the graph $G$ is defined to be the maximum over all vertex partitions of the modularity score, and satisfies $0\leq q^*(G) < 1$. Modularity is at the heart of the most popular algorithms for community detection. We investigate the behaviour of the modularity of the Erdős-Rényi random graph $G_{n,p}$ with $n$ vertices and edge-probability $p$. Two key findings are that the modularity is $1+o(1)$ with high probability (whp) for $np$ up to $1+o(1)$ and no further; and when $np \geq 1$ and $p$ is bounded below 1, it has order $(np)^{-1/2}$ whp, in accord with a conjecture by Reichardt and Bornholdt in 2006. We also show that the modularity of a graph is robust to changes in a few edges, in contrast to the sensitivity of optimal vertex partitions.

Given $p \in (0,1)$, we let $Q_p= Q_p^d$ be the random subgraph of the $d$-dimensional hypercube $Q^d$ where edges are present independently with probability $p$. It is well known that, as $d \rightarrow \infty$, if $p>\frac12$ then with high probability $Q_p$ is connected; and if $p<\frac12$ then with high probability $Q_p$ consists of one giant component together with many smaller components which form the `fragment'. Here we fix $p \in (0,\frac12)$, and investigate the fragment, and how it sits inside the hypercube. In particular we give asymptotic estimates for the mean numbers of components in the fragment of each size, and describe their asymptotic distributions and indeed their joint distribution, much extending earlier work of Weber.

Given a minor-closed class $\mathcal{A}$ of graphs, let $β_{\mathcal{A}}$ denote the supremum over all graphs in $\mathcal{A}$ of the ratio of edges to vertices. We investigate the set $B$ of all such values $β_{\mathcal{A}}$, taking further the project begun by Eppstein. Amongst other results, we determine the small values in $B$ (those up to 2); we show that $B$ is `asymptotically dense'; and we answer some questions posed by Eppstein.

A clique colouring of a graph is a colouring of the vertices such that no maximal clique is monochromatic (ignoring isolated vertices). The least number of colours in such a colouring is the clique chromatic number. Given $n$ points $x_1, \ldots,x_n$ in the plane, and a threshold $r>0$, the corresponding geometric graph has vertex set $\{v_1,\ldots,v_n\}$, and distinct $v_i$ and $v_j$ are adjacent when the Euclidean distance between $x_i$ and $x_j$ is at most $r$. We investigate the clique chromatic number of such graphs. We first show that the clique chromatic number is at most 9 for any geometric graph in the plane, and briefly consider geometric graphs in higher dimensions. Then we study the asymptotic behaviour of the clique chromatic number for the random geometric graph $RG$ in the plane, where $n$ random points are independently and uniformly distributed in a suitable square. We see that as $r$ increases from 0, with high probability the clique chromatic number is 1 for very small $r$, then 2 for small $r$, then at least 3 for larger $r$, and finally drops back to 2.

A clique colouring of a graph is a colouring of the vertices so that no maximal clique is monochromatic (ignoring isolated vertices). The smallest number of colours in such a colouring is the clique chromatic number. In this paper, we study the asymptotic behaviour of the clique chromatic number of the random graph G(n,p) for a wide range of edge-probabilities p=p(n). We see that the typical clique chromatic number, as a function of the average degree, forms an intriguing step function.

Given a graph $H$ with at least one edge, let $\operatorname{gap}_{H}(n)$ denote the maximum difference between the numbers of edges in two $n$-vertex edge-maximal graphs with no minor $H$. We show that for exactly four connected graphs $H$ (with at least two vertices), the class of graphs with no minor $H$ is pure, that is, $\operatorname{gap}_{H}(n) = 0$ for all $n \geq 1$; and for each connected graph $H$ (with at least two vertices) we have the dichotomy that either $\operatorname{gap}_{H}(n) = O(1)$ or $\operatorname{gap}_{H}(n) = Θ(n)$. Further, if $H$ is 2-connected and does not yield a pure class, then there is a constant $c>0$ such that $\operatorname{gap}_{H}(n) \sim cn$. We also give some partial results when $H$ is not connected or when there are two or more excluded minors.

We prove that for every surface $Σ$ of Euler genus $g$, every edge-maximal embedding of a graph in $Σ$ is at most $O(g)$ edges short of a triangulation of $Σ$. This provides the first answer to an open problem of Kainen (1974).

Clustering algorithms for large networks typically use modularity values to test which partitions of the vertex set better represent structure in the data. The modularity of a graph is the maximum modularity of a partition. We consider the modularity of two kinds of graphs. For $r$-regular graphs with a given number of vertices, we investigate the minimum possible modularity, the typical modularity, and the maximum possible modularity. In particular, we see that for random cubic graphs the modularity is usually in the interval $(0.666, 0.804)$, and for random $r$-regular graphs with large $r$ it usually is of order $1/\sqrt{r}$. These results help to establish baselines for statistical tests on regular graphs. The modularity of cycles and low degree trees is known to be close to 1: we extend these results to `treelike' graphs, where the product of treewidth and maximum degree is much less than the number of edges. This yields for example the (deterministic) lower bound $0.666$ mentioned above on the modularity of random cubic graphs.

We present a tight extremal threshold for the existence of Hamilton cycles in graphs with large minimum degree and without a large ``bipartite hole`` (two disjoint sets of vertices with no edges between them). This result extends Dirac's classical theorem, and is related to a theorem of Chvátal and Erdős. In detail, an $(s, t)$-bipartite-hole in a graph $G$ consists of two disjoint sets of vertices $S$ and $T$ with $|S|= s$ and $|T|=t$ such that there are no edges between $S$ and $T$; and $\widetildeα(G)$ is the maximum integer $r$ such that $G$ contains an $(s, t)$-bipartite-hole for every pair of non-negative integers $s$ and $t$ with $s + t = r$. Our central theorem is that a graph $G$ with at least $3$ vertices is Hamiltonian if its minimum degree is at least $\widetildeα(G)$. From the proof we obtain a polynomial time algorithm that either finds a Hamilton cycle or a large bipartite hole. The theorem also yields a condition for the existence of $k$ edge-disjoint Hamilton cycles. We see that for dense random graphs $G(n,p)$, the probability of failing to contain many edge-disjoint Hamilton cycles is $(1 - p)^{(1 + o(1))n}$. Finally, we discuss the complexity of calculating and approximating $\widetildeα(G)$.