Split Graphs, Unigraphs, and Tyshkevich Decompositions
Rebecca Whitman
Submitted in Partial Fulfillment of the Prerequisite for Honors in the Wellesley College Department of Mathematics under advisor Ann Trenk May 2020
c 2020 Rebecca Whitman Abstract
This thesis examines several topics connected to split graphs and unigraphs. We first study a partially ordered set representation of split graphs defined by [13], connecting properties of poset elements, called blocks, to those of split graphs and vice versa. We also find an equivalence relation between poset elements, equivalent to changing the leaf vertices of a vertex of highest degree in a split graph, and provide results for when equivalence classes are closed under various conditions. In the second part of the thesis, we investigate an inverse operation on split graphs, giving proofs for previously stated results and establishing new propositions about the relationship between the inverse operation and unigraphs, induced subgraphs, and hereditary classes. In the third part of the thesis, we use Tyshkevich de- composition [17] to give a structural characterization of the hereditary closure of the set of unigraphs. This characterization implies several results about the nature of forbidden induced subgraphs for the hereditary class.
i Acknowledgements
I cannot offer enough thanks to my thesis advisor and professor Ann Trenk, for her guidance, confidence in my work, and constructive criticism throughout this project. I have learned so much from her over the past two years about how to create and communicate mathematics. I also would like to thank Michael Barrus for his advice and for suggesting the problem that became Chapters 4 and 5 of this thesis. I am grateful to my mathematics professors Megan Kerr, Jonathan Tannenhauser, Karen Lange, Clara Chan, Ismar Voli´c,Helen Wang, and Andy Schultz for mentoring and challenging me and sharing their enthusiasm for the subject. Thank you as well to H´el`ene Bilis for graciously agreeing to serve as the honors visitor. I am fortunate and thankful to have friends who encourage my work, laugh with me, and have helped make my time at Wellesley incredible. Last but not least, I would like to thank my family for their love, kindness, and support.
ii Preface
This thesis studies split graphs in two different contexts: as elements of a poset in their own right and as the building blocks of a graph composition operation. The goal of Chapter 1 is to introduce the reader to the graph theoretic definitions, objects, and operations under consideration. We provide definitions for degree sequences, unigraphs, induced subgraphs, and hereditary classes. Chapter 1 also discusses split graphs, with the class’s forbidden induced subgraph and degree sequence characterizations, and includes background results on split partitions and threshold graphs. The final section introduces the canonical decomposition of a graph as defined by Tyshkevich [17] (hereafter Tyshkevich decomposition). Chapter 2 presents a fourth characterization of split graphs. All degree sequences cor- responding to graphs can be reshaped into elements that form a poset under a majorization operation. The elements corresponding to split graphs take a recognizable, characterizable form [13]. The chapter also introduces an equivalence relation on these poset elements. We give results on the extent to which various properties of split graphs can be identified by the poset representations of split graphs, and are preserved within equivalence classes. In Chapter 3, we examine an inverse operation on split graphs that is used in [17] to identify all indecomposable unigraphs. Tyshkevich states a number of properties about the inverse, and we provide proofs for completeness. Additionally, we give results about the interaction of the inverse with split graph properties, induced subgraphs, and hereditary classes. The final two chapters together state and prove a structural characterization of the hereditary closure of the class of unigraphs, that is, the class of all graphs that are induced in some unigraph. We consider nine graphs and graph families of indecomposable unigraphs
iii more closely in Chapter 4, where, family by family, we find all non-unigraphs induced in an indecomposable unigraph. These graphs, together with the indecomposable unigraphs, are the indecomposable components of graphs in the hereditary closure. Chapter 5 contains the statement and proof of this theorem. We also present a lemma regarding the Tyshkevich decomposition of induced subgraphs, and preliminary work towards a forbidden induced subgraph characterization of the class.
iv Contents
Abstract i
Acknowledgements ii
Preface iii
1 An Introduction to Split Graphs and Tyshkevich Decompositions 1 1.1 Graph Theory ...... 1 1.2 Split Graphs ...... 5 1.3 Tyshkevich Decompositions ...... 8
2 Posets of Split Graphs 12 2.1 Introduction ...... 12 2.2 Partition and Block Posets ...... 12 2.3 Split Graphs ...... 15 2.4 Block Similarity ...... 17 2.5 Graph Properties ...... 20 2.6 Poset Properties ...... 23
3 The Inverse Operation on Split Graphs 28 3.1 Introduction ...... 28 3.2 Inverses Depend on the KS-Partition ...... 29 3.3 Involutions ...... 31 3.4 Inverses of Tyshkevich Compositions ...... 32
v 3.5 Properties Preserved by the Inverse ...... 34 3.6 Hereditary Classes and the Inverse ...... 35
4 Non-Unigraphs Induced in Indecomposable Unigraphs 38 4.1 Introduction ...... 38 4.2 Non-Split Graphs ...... 39
4.2.1 The cycle C5 ...... 39
4.2.2 The family mK2 ...... 40
4.2.3 The family U2(m, n)...... 40
4.2.4 The family U3(m)...... 40 4.3 Split Graphs ...... 42
4.3.1 The graph K1 ...... 43 4.3.2 The family S(p, q)...... 43
4.3.3 The family S2(p1, q1, . . . , pr, qr)...... 44
4.3.4 The family S3(p, q1, q2)...... 44
4.3.5 The family S4(p, q)...... 48
5 A Structural Characterization of the Hereditary Closure of the Unigraphs 53 5.1 Introduction ...... 53 5.2 Characterization Theorem ...... 54 5.3 Properties of the Hereditary Closure ...... 57
6 Open Questions 60
References 62
vi Chapter 1
An Introduction to Split Graphs and Tyshkevich Decompositions
1.1 Graph Theory
In the first section of this chapter, we give necessary background in graph theory for the topics covered in this thesis. For more information, consult Douglas West’s Introduction to Graph Theory [18], a standard graph theory reference.
Definition 1.1. A graph G is an ordered pair consisting of a vertex set V (G) and an edge set E(G) whose elements are unordered pairs of distinct vertices.
For the purposes of this thesis, we will restrict our discussion to finite, simple graphs, that is, those whose vertex set is finite, and contain neither edges from a single vertex to itself nor multiple edges between the same pair of vertices. Figure 1.1 gives several examples of important small graphs. Of these graphs, the K in K1 and 2K2 refers to a clique (see
Definition 1.11) on 1 or 2 vertices, the P in P4 designates a path (line), and C4 and C5 are cycles (polygons) of 4 and 5 vertices, respectively. Up to vertex labeling, two graphs are the “same” when they are isomorphic, which we define here.
Definition 1.2. An isomorphism from graph G to graph H is a bijection f : V (G) → V (H) such that uv ∈ E(G) if and only if f(u)f(v) ∈ E(H). If such a function exists, then we say that G and H are isomorphic, denoted G ∼= H.
1 Figure 1.1: Five graphs that appear throughout the thesis
Isomorphisms preserve all graph properties except with respect to vertex labeling. In the remainder of the section, we will introduce definitions related to three types of graph properties: vertex degree, subgraphs, and graph complements.
Definition 1.3. Two vertices u, v ∈ V (G) are adjacent if uv ∈ E(G). The neighbor set N(v) of a vertex v is the set of vertices adjacent to v. The degree of vertex v, denoted degG(v), is the size of its neighbor set N(v) in graph G.
For example, all vertices in C5 have degree 2. In this thesis, we frequently compare the degree of a single vertex v across several graphs, so we are careful to specify the graph in question as a subscript. We extend the notions of vertex adjacencies and degree globally.
Definition 1.4. A graph G is connected if for all pairs of vertices u, v, there exists a sequence of edges ux1, x1x2, . . . , xnv ∈ E(G), called a path, from u to v.
Each of the graphs in Figure 1.1 is connected except 2K2.
Definition 1.5. A degree sequence is a list of non-negative integers written in non-increasing order, as d1 ≥ d2 ≥ ... ≥ dn.
It is worthwhile to examine when a degree sequence is the set of vertex degrees of some graph.
2 Figure 1.2: Two non-unigraphs realizing the degree sequence 3, 2, 2, 2, 1; with an induced P4 subgraph in bold
Definition 1.6. A degree sequence is graphic if its entries are the vertex degrees of some graph G, and we say that G realizes that degree sequence. Furthermore, a degree sequence is unigraphic if it realizes exactly one graph, namely G, up to isomorphism. If so, we call G a unigraph or say it is unigraphic. Otherwise, G is non-unigraphic.
For example, the degree sequence 2, 2, 1, 1 is realized by the graph P4. Though the small graphs of Figure 1.1 are unigraphs, many graphs are non-unigraphic. Figure 1.2 gives an example of two non-isomorphic graphs that realize the degree sequence 3, 2, 2, 2, 1. Graphic sequences are characterized, and [18] includes the statement and proof of this result. We now introduce another group of definitions surrounding graphs contained within other graphs.
Definition 1.7. A subgraph of G is a graph H such that V (H) ⊆ V (G) and E(H) ⊆ E(G). If H is a subgraph of G, then we say that G contains H.
Subgraphs are created by removing vertices and edges from G. Although every edge in H must have both its endpoints present in V (H), it is possible to exclude edges from G whose endpoints are included in V (H). We use the stricter definition of induced subgraph to rule out this behavior.
Definition 1.8. An induced subgraph H of graph G is a subgraph where u, v ∈ V (H) and uv ∈ E(G) implies uv ∈ E(H). If so, we say that H is induced in G, and that H is the subgraph of G induced by its vertex set, V (H).
3 In Figure 1.2, the subgraph P4 is induced by vertices {a, b, c, d}, in bold, but not by {b, c, d, e}, since we cannot omit edge be. Induced subgraphs are determined completely by the choice of vertex subset. From this, we describe sets of graphs closed under taking induced subgraphs.
Definition 1.9. A family G of graphs is hereditary if every induced subgraph of a graph in G is also a graph in G.
All hereditary classes can also be characterized by what they do not include, with a forbidden induced subgraph characterization. Here, a graph G is an element of a hereditary class if and only if no forbidden induced subgraph is induced in G. Since every graph not in the class must contain an induced subgraph outside of the class, a minimal set of forbidden induced subgraphs always exists, although it may be infinite. Many important results in graph theory characterize a hereditary class by finding a minimal set of forbidden induced subgraphs. For example, the bipartite graphs are precisely those containing no induced cycles of odd length. We introduce the graph complement operation, which “reverses” which edges are present in a graph.
Definition 1.10. The complement G of a graph G is the graph with vertex set V (G) and with uv ∈ E(G) if and only if uv 6∈ E(G).
For example, 2K2 and C4, shown in 1.1, are complements of one another, and P4 =
P4. We state without proof the standard result that induced subgraphs commute with complementation.
Proposition 1.1. If H is an induced subgraph of G, then H is the subgraph of G induced by V (H).
We define two types of vertex subsets and induced subgraphs, based on adjacency.
Definition 1.11. A clique, denoted K, is a subset of V (G) whose elements are pairwise adjacent. In contrast, a stable set, denoted S, is a subset of V (G) whose elements are pairwise non-adjacent. The clique number ω(G) is the maximum size of a clique in G, and the independence number α(G) is the maximum size of a stable set in G.
4 Whereas a clique contains all edges, a stable set contains no edges. For example, both
the clique and stable set numbers of P4 are 2. Hence, the complement of a subgraph induced on a clique is a stable set, and vice versa. Cliques and stable sets are essential for defining split graphs, which will be introduced in the next section.
1.2 Split Graphs
F¨oldesand Hammer [10] first defined split graphs in 1977 and they have been studied ever since. As a subclass of perfect graphs, split graphs are presented in detail in Golumbic’s book Algorithmic Graph Theory and Perfect Graphs [11]. Updated results, including on balanced graphs, can be found in Collins and Trenk’s forthcoming chapter “Split Graphs” in Topics in Algorithmic Graph Theory [9]. The reader can consult either text for proofs to theorems and more information.
Definition 1.12. A graph G is a split graph if there exists a partition of the vertex set of G into a clique K and a stable set S. We call this partition a KS-partition.
A split graph may have multiple KS-partitions, with different clique sizes. Figure 1.3 shows two split graphs, one of which is depicted twice, with different KS-partitions. Since any induced subgraph of a split graph can inherit its KS-partition, the class of split graphs is hereditary. We provide its forbidden induced subgraph characterization.
Figure 1.3: Two split graphs; the solid vertices are the clique and the hollow vertices are the stable set of the given partitions.
5 Theorem 1.2. [10] A graph G is a split graph if and only if G contains no induced 2K2,
C4, or C5.
These three forbidden induced subgraphs are shown in Figure 1.1. The vertices in K are filled-in circles, whereas the vertices in S are hollow. Unlike the majority of graph classes, split graphs can also be identified from their degree sequences. We recount that theorem here.
Theorem 1.3. [12] A graph G on n vertices is split if and only if its degree sequence d =
(d1, d2, . . . , dn) satisfies the following equation for h = ω(G):
h n X X di = h(h − 1) + di i=1 i=h+1 Furthermore, it is possible to deduce the clique size in all possible KS-partitions of a split graph from its degree sequence, and so all split graphs realizing the same degree sequence have the same clique number. In [8], Collins and Trenk introduce terminology for split graphs with a unique KS-partition.
Definition 1.13. A split graph is balanced if there exists a unique partition of its vertices into a clique and a stable set, and unbalanced otherwise.
We can determine whether a split graph has one or multiple KS-partitions based on the presence of swing vertices, which we define here.
Definition 1.14. Given a split graph G and any KS-partition K ∪ S of G, a swing vertex v is a vertex of G whose neighbor set is exactly K − {v}.
Because swing vertices can “swing” back and forth between the clique and the stable set of a KS-partition, we give a remark about balanced graphs.
Remark 1.4. A split graph G is balanced if and only if it contains no swing vertices.
Furthermore, as a result of Hammer and Simeone’s work in [12], all KS-partitions of a split graph take one of three forms. We classify partitions by comparing |K| with ω(G), the clique number of G (see Definition 1.11), and |S| with α(G), the independence number. We call a KS-partition K-max when |K| = ω(G) and S-max when |S| = α(G); for an example, see Figure 1.3.
6 Theorem 1.5. Let G be a split graph with partition K ∪ S. Exactly one of the following holds:
1. |K| = ω(G) and |S| = α(G), in which case G is balanced.
2. |K| = ω(G) and |S| = α(G) − 1, in which case G is unbalanced and the partition is K-max.
3. |K| = ω(G) − 1 and |S| = α(G), in which case G is unbalanced and the partition is S-max.
Thus, a given split graph has at most two partitions, up to isomorphism. In addition to swing vertices, we assign names to vertices of particular degrees.
Definition 1.15. In any graph G, a vertex v ∈ V (G) is an isolated vertex if degG(v) = 0, a leaf vertex if degG(v) = 1, and a dominating vertex if degG(v) = |V (G)| − 1.
Hence, isolated vertices have no neighbors, leaf vertices (leaves) are adjacent to exactly one vertex, and dominating vertices are adjacent to every other vertex in G. For example, the top vertex of the left-hand graph in Figure 1.3 is a leaf vertex, and the single vertex in
K1 shown in Figure 1.1 is trivially both an isolated and dominating vertex. We conclude this introduction to split graphs with a discussion of one subclass of split graphs, the threshold graphs. Threshold graphs were first defined by Chvat´aland Hammer [4] in 1973.
Definition 1.16. A graph G is a threshold graph if there exist a threshold t > 0 and a positive weight ai assigned to each vertex vi so that S ⊆ V (G) is a stable set if and only if P i∈S ai ≤ t.
An example of a threshold graph is given in Figure 1.4. There are many characteri- zations of threshold graphs, including the following. We will give further characterizations in Propositions 1.11 and 2.8.
Theorem 1.6. The following are equivalent for a graph G:
1. G is a threshold graph.
7 Figure 1.4: An example of a threshold graph
2. G contains no induced 2K2,C4, or P4.
3. G can be constructed from the empty graph by repeatedly adding an isolated vertex or a dominating vertex.
We omit the proof, but note that since P4 is an induced subgraph of C5, it follows that all threshold graphs are split graphs, by Theorem 1.2. The forbidden induced subgraphs of threshold graphs are shown in Figure 1.1. Additionally, threshold graphs are unbalanced by Remark 1.4, since the first vertex v in the sequence of isolated and dominating vertices is a swing vertex. Every dominating vertex that is subsequently added is adjacent to v, and no isolated vertex added is adjacent to v. For example, in Figure 1.4, one possible vertex order to construct the graph is a (isolated/swing), c (isolated), b (dominating), and d (dominating). Split graphs are not necessarily unigraphic (see Figure 4.5 for a non-example), but threshold graphs are all unigraphic by the construction in Theorem 1.6. The next section, on a particular method of graph decomposition, includes an important result on identifying unigraphs.
1.3 Tyshkevich Decompositions
Across graph theory, numerous operations have been defined to combine and separate graphs. Here, we introduce a relatively recent one that decomposes any graph into split graph com- ponents and one free component. This graph decomposition operation, first introduced by Tyshkevich [16], [17], acts like factorization, where indecomposable split graphs stand in for
8 prime numbers, the fundamental building blocks. We define the composition of two graphs, then extend to n graphs and decomposition.
Definition 1.17. Let G be a split graph with KS-partition K ∪ S, and let H be any graph. Then, the composition of G and H, denoted G ◦ H, is the graph on V (G) ∪ V (H) with edge set E(G) ∪ E(H) ∪ {uv|u ∈ K, v ∈ V (H)}.
For example, consider G = P4 ◦ C4, where the clique vertices in the unique KS- partition of P4 are filled in:
=
Thus, G ◦ H is the disjoint union of the two graphs, together with all edges from the designated clique in G to vertices in H. While H may not be a split graph, the composition G ◦ H is split if and only if H is split. Changing the KS-partition of G always changes the composition, so in the case of unbalanced graphs, it is essential to specify the KS-partition of split graphs under composition. It is straightforward to extend the composition to n split graphs and one graph chosen freely. We define indecomposable graphs as those that cannot be the result of a non-trivial graph composition.
Definition 1.18. A graph G is decomposable if there exist non-empty graphs G1,G0, where
G1 is a split graph and G = G1 ◦ G0, and indecomposable otherwise.
We state the existence and uniqueness of Tyshkevich decomposition; a proof is found in [17].
Theorem 1.7. Every graph G can be written as a composition of indecomposable compo-
nents, where each of Gn,...,G1 has a fixed KS-partition:
G = Gn ◦ ... ◦ G1 ◦ G0.
9 Furthermore, when each Gi is non-empty, this decomposition is unique up to isomorphism, where neither the order of the components nor the choice of partitions can vary.
Finding a graph’s unique Tyshkevich decomposition is not easily done on sight. There is a degree sequence characterization of indecomposable graphs in [17], but the characteri- zation by Barrus and West [3] of indecomposable graphs and indecomposable components within a graph is much easier to use. We present that theorem in terms of induced subgraphs, rather than hypergraphs.
Theorem 1.8. Create a graph A4(G) with V (A4(G)) = V (G) and with edge uv if and only if
some P4,C4, or 2K2 induced in G contains both u and v. Then, a graph G is indecomposable
if and only if A4(G) is connected. More generally, the vertex sets of the components of the
decomposition are the vertex sets of the connected components of A4(G).
Figure 1.1 depicts the induced subgraphs P4, C4, and 2K2. We can say more about indecomposable graphs with respect to balance and KS-partitions.
Proposition 1.9. [9] A split graph G is unbalanced if and only if G = H ◦ K1 for some split graph H.
Proof. Let H be a split graph, so H ◦ K1 is also split. Since the single vertex k ∈ K1 is
adjacent to exactly the clique vertices of H in the composition, k is a swing vertex in H ◦K1.
By Remark 1.4, H ◦ K1 is unbalanced. For the converse, consider an unbalanced graph G with a K-max partition K ∪ S. Since G is unbalanced, it has a swing vertex k ∈ V (G), whose neighbor set NG(k) = K − {k}. Let H be the subgraph of G induced on V (G) − {k}, 0 0 0 0 with an inherited partition K ∪ S , so K = K − {k}. Thus NG(k) = K , so k is adjacent to exactly the clique part of H, and G can be decomposed as G = H ◦ K1.
Thus, swing vertices are represented in decompositions with K1 as the rightmost term. As a corollary to Proposition 1.9, indecomposable graphs cannot be unbalanced with more than one vertex, or they can be decomposed as above.
Proposition 1.10. If G is an indecomposable split graph, then G is balanced (and hence
has a unique KS-partition) or is isomorphic to K1.
10 The converse does not hold, and many balanced split graphs are decomposable. We can similarly show that isolated and dominating vertices are represented in decompositions
with K1 as the leftmost term, contingent on partition. Based on the construction of threshold graphs in Theorem 1.6, we derive the following corollary.
Corollary 1.11. A graph G is threshold if and only if every element of its Tyshkevich
decomposition is K1, with either partition.
Although the class of unigraphs is not hereditary and therefore has no forbidden induced subgraph characterization, it can be characterized via Tyshkevich decomposition. In addition to stating and proving the following theorem, Tyshkevich [17] provides a list of all indecomposable unigraphs, completing the characterization.
Theorem 1.12. A graph G is unigraphic if and only if the components of its unique decom- position are indecomposable unigraphs.
Equivalently, if a collection of split graphs G1,...,Gn and a graph G0 are all uni- graphic, then so too is their composition Gn ◦ ... ◦ G0. The indecomposable unigraphs are the focus of Chapter 4. We now set aside Tyshkevich decomposition and turn to a focused discussion of split graphs and subclasses in Chapter 2.
11 Chapter 2
Posets of Split Graphs
2.1 Introduction
Here, we consider a partially ordered set whose elements are ordered pairs of partitions of n into distinct integers. Each element of this poset will correspond to a degree sequence of a split graph. We will investigate the poset representation of split graph properties, and the graph representation of poset elements’ properties. For additional results and proofs, Merris’ Graph Theory [13] is an excellent resource.
2.2 Partition and Block Posets
We begin by constructing the well-known poset of partitions of a natural number into distinct parts. Let α = α1α2 . . . αm be a partition of n ∈ N into m distinct parts, such that α1 >
... > αm. Denote the length of α by len(α). As an example, consider α = 321, which is a partition of 6 into distinct parts, with len(α) = 3. We will continue this example throughout this chapter.
Definition 2.1. Denote the set of partitions of n ∈ N into distinct parts by Dis(n). Let Dis(n, m) be the (induced) subposet of Dis(n) comprising elements of length m.
Thus, α = 321 is an element of Dis(6) and in fact is the unique element of Dis(6, 3). We define an ordering on elements of Dis(n).
12 Definition 2.2. If α and β are two partitions of n into distinct parts, then β majorizes α, denoted by β α, when for all k ≤ len(β),
k k X X βi ≥ αi. i=1 i=1 Continuing our example, β = 42 is a partition of 6 into distinct parts, and 42 321. For k = 1, we have 4 > 3, and for k = 2, the sums of the first two terms of β and α yield that 6 > 5. We provide a proof that the majorization operation induces a poset.
Proposition 2.1. The set Dis(n) is a poset ordered by majorization.
Proof. It is easy to see that for all α ∈ Dis(n), α α. The relation is anti-symmetric, since Pk Pk whenever β 6= α, there exists k ≤ len(β) with i=1 βi > i=1 αi. Since the sum of the first k terms of α is strictly less than that of β, it follows that α 6 β. Lastly, majorization is transitive since the relation ≥ is transitive. Hence Dis(n) is a poset under majorization.
We can construct Hasse diagrams of these posets; for example, the Hasse diagram of Dis(6) is shown in Figure 2.1.
6
51
42
321
Figure 2.1: The Hasse diagram of Dis(6)
Next, we combine elements of Dis(n) into ordered pairs, called blocks.
Definition 2.3. A block is an ordered pair [α|β] where α, β ∈ Dis(n), β α, and len(β) = len(α) or len(β) = len(α) − 1. Let Block(n) denote the set of blocks with α, β ∈ Dis(n).
For example, [321|42] is a block, because 42 and 321 are each distinct partitions of 6, 42 321, and len(42) = len(321) − 1. The length condition in the block definition is
13 inspired by work in [14] and allows us to restrict our definition to blocks of split graphs, as we will see in Proposition 2.5. We next define what it means for one block to majorize another, and create a poset of blocks of partitions of n into distinct parts. Note that two types of majorization occur simultaneously, both within and between blocks.
Definition 2.4. If [α|β] and [α0|β0] are blocks with α, β, α0, and β0 ∈ Dis(n), then [α0|β0] majorizes [α|β] when β β0 and α0 α.
For example, [42|42] [321|42]: we compare the α terms and 42 321. Again, we confirm that this majorization operation defines a poset.
Proposition 2.2. The set Block(n) is a poset ordered by block majorization.
Proof. Let [α|β], [α0|β0], and [α00|β00] ∈ Block(n). Since α α and β β, it follows that block majorization is reflexive. Suppose that [α0|β0] [α|β] and [α0|β0] 6= [α|β]. Thus, β β0 and α0 α. Both of these conditions are antisymmetric, so it follows that whenever β 6= β0, then β0 6 β, and when α 6= α0, then α 6 α0. Since the two blocks are not equal, we conclude that [α|β] 6 [α0|β0]. Suppose that [α00|β00] [α0|β0] [α|β]. Thus β β0 β00 and α00 α0 α. By transitivity of partition majorization, it follows that [α00|β00] [α|β]. Hence Block(n) is a poset ordered by majorization.
The Hasse diagram of Block(6) is shown in Figure 2.2.
[321|321][42|42] [51|51] [6|6]
[321|42] [42|51] [51|6]
[321|51] [42|6]
Figure 2.2: The Hasse diagram of Block(6)
14 2.3 Split Graphs
Now that we have shown Block(n) is a poset, we will argue that its elements correspond to degree sequences of split graphs. To begin, we consider a combinatorial tool, the Ferrers diagram, which can be used to represent degree sequences.
Definition 2.5. A Ferrers diagram is an array of boxes in left-justified rows. Rows are counted from the top, and the number of boxes in each row is non-increasing.
In a Ferrers diagram F , let αi be the number of boxes in the ith row starting at the
main diagonal, and βj be the number of boxes in the jth column starting below the main
diagonal. Figure 2.3 shows a Ferrers diagram with each αi and βj labeled. In [13], Merris
showed that each block is uniquely represented with a Ferrers diagram when {βj} majorizes
{αi}. We state this proposition and give a sketch of the proof. For more details, see [13].
Figure 2.3: The Ferrers diagram of block [321|42]
Proposition 2.3. There is a bijection between Block(n) and the set of Ferrers diagrams of
2n boxes where the sequence {βj} of column lengths below the main diagonal majorizes the
sequence {αi} of row lengths to the right of and including the main diagonal.
Proof. Let F be a Ferrers diagram under the given conditions. For the majorization condition
from Definition 2.2 to hold, α = {αi} and β = {βj} must partition the same integer; since the Ferrers diagram has 2n boxes, this integer is n. It can be shown that α and β are each distinct partitions of n if and only if the Ferrers diagram has non-increasing row lengths. Lastly, F has no blocks missing in the middle of its rows and has non-increasing row lengths
15 exactly when len(β) = len(α) or len(β) = len(α) − 1. We conclude that α, β ∈ Dis(n) and F corresponds uniquely to the block [α|β].
For instance, the Ferrers diagram of [321|42] is shown in Figure 2.3. There is a straightforward correspondence between a Ferrers diagram and a degree sequence d1, . . . , dt.
Namely, di is equal to the number of boxes in row i. Ruch and Gutman [15] show when such a degree sequence is graphic.
Proposition 2.4. The degree sequence d = d1, . . . , dn is graphic exactly when the sequences
{βj} and {αi} of its Ferrers diagram fulfill a weaker form of majorization, called weak sub- majorization.
Under weak sub-majorization, it is also possible to study larger posets of blocks corresponding to all graphs, not just split graphs. Both Merris [13] and Collins and Trenk [9] provide definitions of weak sub-majorization and are useful resources on these larger posets. For example, from the Ferrers diagram for [321|42] in Figure 2.3, we derive the degree sequence (3, 3, 3, 2, 1). The graph shown in Figure 2.4 is the unique graph realizing this degree sequence.
Figure 2.4: The split graph associated to the block [321|42]
Thus far, we have shown that blocks under majorization correspond to both Ferrers diagrams and degree sequences under some conditions. Our next result is that the elements of these posets correspond to split graphs, giving an additional perspective from which to study split graphs. We state the proposition from [13] and give a sketch of the proof.
16 Proposition 2.5. Each block [α|β] corresponds uniquely to a degree sequence of a split graph.
Proof. The proof is in two steps: first, we must show that the degree sequence of [α|β] is graphic. This follows from Proposition 2.4 since β α implies that β weakly majorizes α. To verify that this graphic degree sequence indeed gives rise to split graph(s), we apply Theorem 1.3, the degree sequence characterization of split graphs. The calculation details can be found in multiple sources, including in [9].
There could be several split graph realizations of the same degree sequence, and consequently the same Ferrers diagram. We introduce notation to count these.
Definition 2.6. The realization number r([α|β]) is the number of non-isomorphic graphs arising from the degree sequence associated with block [α|β].
Thus, a block and its associated degree sequence are unigraphic if r([α|β]) = 1. For example, the realization number r([321|42]) is 1, since the three vertices of degree three must be adjacent to one another, and then each has exactly one more adjacency to one of the vertices of degrees one or two. Consider a second example: let α0 = 521 and β0 = 62, from Dis(8). Then [521|62] ∈ Block(8), and its Ferrers diagram is shown in Figure 2.5. We derive the degree sequence 5, 3, 3, 2, 1, 1, 1 from the row lengths of the Ferrers diagram. Since two non-isomorphic graphs, given in Figure 2.5 realize this degree sequence, it follows that r([521|62]) ≥ 2.
2.4 Block Similarity
In order to trace properties between elements of Block(n) for different natural numbers n, we define a notion of similarity between blocks, which forms an equivalence relation. Our equivalence relation will preserve many properties of both blocks and split graphs. As with the definitions above, we will begin at the level of partitions and extrapolate to blocks. We will then consider numerous properties of blocks or split graphs, establish parallel definitions in terms of the other category, and show, except in the cases of unigraphs and minimal blocks, that the properties are invariant among graphs in a shared equivalence class.
17 Figure 2.5: The two non-isomorphic split graphs whose degree sequence corresponds to the block [521|62].
Definition 2.7. Partition α0 ∈ Dis(n) is a shift of α ∈ Dis(m) by the integer n − m if
0 0 0 α1 − α1 = n − m and αk = αk for all 2 ≤ k ≤ len(α) = len(α ).
Thus, α and α0 differ only in their first element.
Definition 2.8. Let [α|β] ∈ Block(m) and [α0|β0] ∈ Block(n), with len(α) = len(α0) and len(β) = len(β0). These blocks are similar, denoted by [α|β] ∼ [α0|β0], if α0 is a shift of α by some integer k, and β0 is a shift of β by the same integer k.
For example, [321|42] ∼ [521|62], since 521 is a shift of 321 by 2, and 62 is a shift of 42 also by 2. We can also describe block similarity in terms of the graphs realized from each block: when two blocks are similar, each graph realizing one block differs from a graph realizing the other block only by the number of leaves adjacent to a specified vertex of highest degree. In other words, if [α|β] ∼ [α0|β0] by a shift of k > 0, we can take a graph G whose degree sequence corresponds to [α|β] and attach k additional leaves to a maximum degree vertex to get a graph G0 that realizes [α0|β0]. For example, Figure 2.4 shows a graph G corresponding to the block [321|42], and Figure 2.5 shows two graphs H and H0 which realize the block [521|62]. Observe that G can be obtained from H and H0 by removing 2 leaves from a maximum degree vertex, and H and H0 can each be obtained by adding 2 leaves to a maximum degree vertex of G. Additionally, if two graphs differ only by the number of leaves adjacent to a highest degree vertex, then their blocks are similar.
18 We now show that these shifts induce an equivalence relation.
Proposition 2.6. Block similarity is an equivalence relation.
0 0 Proof. Since α1 = α1 + 0 and β1 = β1 + 0, it follows that block similarity is reflexive. Let [α|β] ∼ [α0|β0] by a shift of k. Then [α0|β0] ∼ [α|β] by a shift of −k. To show transitivity,
0 0 00 00 0 0 00 0 let [α |β ] ∼ [α |β ] by a shift of l. Then α1 = α1 + k and α1 = α1 + l, and β1 = β1 + k and 0 00 00 00 00 00 β1 = β1 +l. Hence α1 = α1 +l+k and β1 = β1 +l+k, so we conclude that [α|β] ∼ [α |β ].
Since we can shift among similar blocks, we will focus on those in smallest possible form, which are useful for counting graph realizations.
Definition 2.9. A partition α ∈ Dis(n) is in root form if α1 = α2 + 1.
Definition 2.10. We similarly define a block [α|β] ∈ Block(n) to be in root form if either
α1 = α2 + 1 or β1 = β2 + 1.
This is the smallest possible form since we cannot shift α or β by a negative integer, to make their entries smaller. If we were to do so, then the entries of α or β would no longer be distinct and in strictly decreasing order. For example, α = 321 is in root form, but (42), (521), and (62) are not in root form. Since one of its partitions is in root form, [321|42] is in root form. The block [521|62] is not in root form, but it is similar to the root form block [321|42]. Root form blocks are of interest because they are the canonical representatives of their equivalence classes. Additionally, only graphs realizing root form blocks may have two or more vertices of highest degree. For instance, in Figure 2.4, vertices a and b are both of highest degree (3). Since no automorphism of the graph realizing [321|42] maps a to b, there are two non-isomorphic realizations of [321|42] shifted by 1, that is, the block [421|52], since we can append a leaf vertex to either a or b. Should we wish to shift [421|52] up 1 to [521|62], there are still exactly two non-isomorphic realizations of [521|62] since each of the two graphs realizing [421|52] has a unique vertex of highest degree. We summarize this process in the following proposition.
0 0 0 0 0 Proposition 2.7. If [α|β] and [α |β ] are similar and α1 ≥ α1, then r([α |β ]) ≥ r([α|β]). Furthermore, r([α0|β0]) = r([α|β]) when [α|β] is not in root form.
19 0 0 0 Proof. Shifting [α|β] to [α |β ] is equivalent to appending k = α1 −α1 leaf vertices to a vertex of highest degree in each realization of [α|β]. Let r([α|β]) = r and let G1,...,Gr be the r 0 non-isomorphic realizations of [α|β]. From each graph Gi, create graph Gi by identifying a vertex v of highest degree and append k leaves to v. Since Gi and Gj are pairwise non- 0 0 0 0 isomorphic, so too are Gi and Gj. Thus, G1,...,Gr are r non-isomorphic realizations of [α0|β0], so r([α0|β0]) ≥ r. When [α|β] is not in root form, it has already been shifted at least once from a root form block, so all graphs realizing [α|β] have a unique vertex of highest degree. Therefore,
0 each graph Gi has exactly one graph Gi corresponding to it. We can also remove k leaf ver- 0 0 0 0 tices from any graph realizing [α |β ] to get a graph realizing [α|β]. Therefore, {G1,...,Gr} is the set of graphs realizing [α0|β0], so r([α0|β0]) = r.
We conclude this section with several open questions on the relationship between similarity and the realization number. First and foremost, is it possible to deduce r([α|β]) from α and β from the block? We also hope to improve Proposition 2.7 and give an exact count of the realizations of [α0|β0], given a similar [α|β] in root form. Both are included on our list of open questions in Chapter 6.
2.5 Graph Properties
We now have the tools to address various properties of split graphs and connect them to the block representation of a graph. Throughout this section, we will refer to blocks, sequences, and graphs equally with properties normally attributed to graphs (e.g., a block is balanced when...) We begin with threshold graphs and state a result from Ruch and Gutman [15], then describe the effect of similarity on threshold graphs.
Proposition 2.8. A block [α|β] corresponds to threshold graphs if and only if α = β.
Proposition 2.9. Similar blocks both correspond to threshold graphs, or neither does.
Proof. Let [α|β] be a threshold graph, so α = β. We shift [α|β] by k with the result that
0 0 0 0 α = α1 + k, α2, . . . , αm = β , so [α |β ] is also a threshold graph by Proposition 2.8.
20 The other direction is similar.
Figure 2.6 shows the Ferrers diagram and threshold graph corresponding to [32|32].
Figure 2.6: The Ferrers diagram and threshold graph realizing block [32|32]
Recall that a split graph is balanced if there exists a unique partition of its vertices into a clique and a stable set, and unbalanced otherwise. We state a characterization of balanced split graphs in terms of blocks that appears in [8].
Proposition 2.10. A block [α|β] is unbalanced if and only if len(α) = len(β), and conse- quentially, [α|β] is balanced exactly when len(α) = len(β) + 1.
Since shifting does not change the length of either α or β, similarity preserves balance, which we state as a corollary.
Corollary 2.11. Similar blocks are both balanced or both unbalanced.
Another type of graphs, Nordhaus-Gaddum graphs, generalize split graphs. They were initially characterized by Collins and Trenk in [7], through limitations on the chromatic number of the graph and its complement. Further work by Cheng, Collins, and Trenk, including applications to split graphs, is found in [6]. Nordhaus-Gaddum graphs are the graphs whose vertices can be partitioned into a clique (B), a stable set (C), and a nonempty third set (A) that is either a clique, a stable set, or C5. All possible edges between A and B are present, and there are no edges between A and C. Based on the options for A, there are three types of Nordhaus-Gaddum graphs.
21 Definition 2.11. A graph is an NG-1 graph if A is a clique, NG-2 graph if A is a stable
set, and NG-3 graph if A is C5. A graph is both NG-1 and NG-2 if and only if A is a single vertex.
We summarize results about split graphs from [6]. If a graph G is NG−1, then A ∪ B is a clique and C a stable set, so G is a split graph. If G is NG-2, then A ∪ C is a stable set,
and B a clique, so G is again a split graph. Since C5 is a forbidden induced subgraph of split graphs, it follows that NG-3 graphs are not split graphs. In both NG-1 and NG-2 graphs, all vertices of A are swing vertices. Since A is nonempty, these graphs are unbalanced. We will connect blocks to Nordhaus-Gaddum graphs via another characterization of the latter.
Let d1 ≥ d2 ≥ ... ≥ dn be a degree sequence satisfying the split graph criterion in Theorem 1.3, and let h be the value defined in Theorem 1.3. The degree sequence corresponds to an
NG-1 graph if dh > dh+1 and to an NG-2 graph if dh−1 > dh. Should both conditions occur, the graph is both NG-1 and NG-2. For instance, the graph in Figure 1.4 which realizes the block [32|32] has degree sequence 3, 3, 2, 2, with h = 3, so this graph is NG-2. It may be helpful to refer to this example when reading the proof of Proposition 2.12 for NG-2 graphs. We classify unbalanced blocks as NG-1 or NG-2 based on the final elements of α and β.
Proposition 2.12. Let [α|β] be a block and let len(α) = len(β) = m. Then [α|β] is NG-1
if and only if βm = 1, and [α|β] is NG-2 if and only if αm ≥ 2. Furthermore, all unbalanced split graphs exhibit one or both of these properties.
Proof. Consider the unique degree sequence d corresponding to [α|β], and fix h as defined
in Theorem 1.3. Since [α|β] is unbalanced by Proposition 2.10, it is known that dh = dm+1,
and that the vertex with degree dh will be a swing vertex, with m squares in its row of the
Ferrers diagram. Then [α|β] has βm = 1 if and only if dh+1 < dh, since the single Ferrers
diagram square implied by βm = 1 occurs in row h, which is equivalent to [α|β] being NG-1.
On the other hand, [α|β] has αm ≥ 2 exactly when dh−1 contains h − 2 + αm ≥ h = m + 1
squares, but dh has exactly m squares. Hence dh−1 > dh and [α|β] is NG-2.
For a contradiction, if neither property occurs for a block [α|β], then bm > am and so Pm−1 Pm−1 i=1 βi < i=1 αi and β 6 α.
22 We note further that if these properties occur simultaneously, then [α|β] is both NG-
1 and NG-2. Since these conditions are irrespective of α1 and β1, they are invariant under similarity. We state this as a corollary.
Corollary 2.13. For any two similar blocks, both correspond to NG-1 graphs or neither does, and both correspond to NG-2 graphs or neither does.
2.6 Poset Properties
We now return to the posets Dis(n) and Block(n), and translate poset properties to the generated split graphs. More precisely, the minimum and maximum elements of Dis(n, m) will give rise to elements of Block(n) and classes of split graphs with certain properties, including a hereditary class. We begin with results about maximum elements, which are easier to prove, then proceed to minimum elements. The following result is well-known; we provide a proof for completeness.