ZERO FORCING AND POWER DOMINATION IN GRAPHS
SUDEEP STEPHEN Master of Science (Mathematics)(University of Madras) School of Mathematical and Physical Sciences The University of Newcastle, Australia
A thesis submitted in total fulfilment of the requirements for the degree of Doctor of Philosophy October 2017
Statements Statement of Originality
The thesis contains no material which has been accepted for the award of any other degree or diploma in any university or other tertiary institution and, to the best of my knowledge and belief, contains no material previously published or written by another person, except where due reference has been made in the text. I give consent to the final version of my thesis being made available worldwide when deposited in the University’s Digital Repository, subject to the provisions of the Copyright Act 1968.
Statement of Authorship
I hereby certify that the work embodied in this thesis contains published papers of which I am a joint author. I have included as part of the thesis a written statement, endorsed by my supervisor, attesting to my contribution to the joint publication.
Signed: SUDEEP STEPHEN
The candidate has six papers, of which he is a joint author, embodied in the thesis. As principal supervisor I can attest that the candidate’s contribution, in all cases were at least 80% of the final article.
Signed: Dr JOE RYAN iv
List of Publications
Publications Arising from this Thesis (1) D. Ferrero, T. Kalinowski and S. Stephen. Zero forcing in iterated line digraphs. Submitted to Discrete Applied Mathematics, 2017 [37]. (2) R. Davila, T. Kalinowski and S. Stephen. Proof of a conjecture of Davila and Kenter regarding a lower bound for the forcing number in terms of girth and minimum degree. Submitted to Discrete Applied Mathematics, 2016 [23]. (3) D. Ferrero, C. Grigorious, T. Kalinowski, J. Ryan and S. Stephen. Minimum rank and zero forcing number for butterfly networks. Submitted to Discrete Applied Mathematics, 2016 [38]. (4) C. Grigorious, T. Kalinowski, J. Ryan and S. Stephen. On the power domination number of de Bruijn and Kautz digraphs. To appear in Lecture Notes in Computer Science, 2017 [45]. (5) S. Stephen, B. Rajan, C. Grigorious and A. William. Resolving-power dominating sets. Applied Mathematics and Computation, 256, 778–785, 2015 [85]. (6) S. Stephen, B. Rajan, J. Ryan, C. Grigorious and A. William. Power domination in certain chemical structures. Journal of Discrete Algorithms, 33, 10–18, 2015 [86].
Other Publications Produced During my Candidature (1) C. Grigorious, T. Kalinowski, J. Ryan and S. Stephen. Metric dimension of Circulant graph C(n, ±{1, 2, 3, 4}). The Australasian Journal of Combinatorics, 69 (3), 417– 441, 2017 [46]. (2) C. Grigorious, S. Stephen, B. Rajan and M. Miller. On the Partition Dimension of Circulant Graphs. The Computer Journal, 60 (2), 180–184, 2016 [48]. (3) S. Klavˇzar, P. Manuel, M. J. Nadjafi-Arani, R. S. Rajan, C. Grigorious and S. Stephen. Average Distance in Interconnection Networks via Reduction Theorems for Vertex-Weighted Graphs. The Computer Journal, 59(12), 1900–1910, 2016 [68]. (4) C. Grigorious, P. Manuel, M. Miller, B. Rajan and S. Stephen. On the metric dimension of circulant and Harary graphs. Applied Mathematics and Computation, 248, 47–54, 2014 [47]. (5) P. Manuel, B. Rajan, C. Grigorious and S. Stephen. On the Strong Metric Dimen- sion of Tetrahedral Diamond Lattice. Mathematics in Computer Science, 9, 201–208, 2015 [76]. (6) C. Grigorious, S. Stephen, B. Rajan, M. Miller and A. William. On the Partition Dimension of a Class of Circulant Graphs. Information Processing Letters, 114, 353– 356, 2014 [49]. To my parents, Stephen P.M and Mary Stephen And in memory of my most beloved supervisor, Mirka Miller.
Acknowledgements
First and foremost, I would like to express my sincere gratitude to my su- pervisors, (L) Prof. Mirka Miller, Dr. Joe Ryan, Dr. Thomas Kalinowski and Dr. Paul Manuel, who have supported me with their encouragement, guidance and patience. My sincere gratitude to Mirka and Joe, without them, I would never have been in Newcastle. Thank you for trusting my research capabilities and above all for the emotional and practical support whenever I needed them. Thank you for teaching how to build collaborations and introducing me to all your collaborators. I can never forget the good times we had spent together. Your motivations and encouragements helped me to be a better person and a researcher. It was indeed a tough phase for me and Joe after Mirka lost her battle with cancer. Thank you Joe for being there always and guiding me through the completion of my candidature. I am happy we did this together for Mirka. I specially like to acknowledge Thomas Kalinowski for his insightful sug- gestions and comments on my research and talks at every stage of my Ph.D. I am always inspired by his work. To be very honest, I have always looked up at him as my role model. I also thank Dr. Paul Manuel for contributing 25% towards my scholarship. My special thanks to Dr. Bharati Rajan, my mentor, for introducing me to graph theory and research. I owe what I am today to you. I am thankful to Dr. Cristina Dalfo Simo, my external supervisor for her inputs and support during my Ph.D candidature. I am grateful to Prof. Indra Rajasingh and Dr. Sundara Rajan for their encouragement and support during my initial stages of research in Chennai. Discussions and chats with Prof. Brian Alspach were always interesting and helpful. Thank you. I am also thankful to Renee Mam and the entire Maths department of St. Joseph’s College, Bangalore for inspiring me to do research in mathematics. I like to thank Prof. Zdenek Ryacek and Prof. Dafik for hosting me during my stay in The University of West Bohemia, Czech Republic and University of Jember, Indonesia respectively. I also like to thank my col- laborators Randy Davila, Prof. Daniela Ferrero and Prof. Sandi Klavzar for accepting my request to work with me and sharing their knowledge. It was indeed wonderful 3 years in Newcastle. I would like to acknowledge Dr. Benny Sudakov for pointing out the argument for Theorem 5.2.3 on his visit to the University of Newcastle during the 40 ACCMCC. Thanks to the University of Newcastle for providing scholarships (UNIPRS and UNRSC 50-25) to pursue my dream. I thank the International Leader- ship Experience and Development (ILEAD) Plus Program for the generous support for participating in various conferences. I acknowledge the Faculty of Sciences for the Conference scholarship. My sincere thanks to the school of mathematical and physical sciences. All the staff, both academic and administrative staff created a healthy and comfortable environment for me to pursue my research. Many thanks to all of them for their supports and efforts. Special thanks to Vicki, Anna Rosa, Irene, Rosemary and Felicity for their administrative help whenever I needed them. To my friend and colleague Cyriac Grigorious, thank you so much for being there with me for the past 8 years. You have influenced my life more than you can imagine. Thank you for the support and friendship during my stay here. To my office mates, Bong, Dushyant, Oudone and Wulan, thank you for bearing me. Thanks to the Malayali community in Newcastle and Sydney for making me welcome among them and making me feel at home away from home. I am indebted to my parents, Stephen P. M and Mary Stephen, my wife, Jinu Joseph, my brother, Suraj Stephen, my sister-in-law, Rakhi Ashok and my nephew, Aswin Suraj, for their love and support during these years. A special word of acknowledgement to Mr. Simon, Saly amme, Ashwin and Ada for considering me a part of their family. Thank you for the unconditional love you have shown towards me. Saly amme, thank you so much for being there like a mother encouraging, supporting and guiding me through various frustrations and disappointments. Last but not least, I thank God Almighty for blessing me with this opportunity to pursue my research and for giving me sound body and mind to complete this successfully.
Sudeep Stephen Contents
List of Figures xi
Nomenclature xv
Chapter 1. Introduction1 1.1. Contributions of the thesis3 1.2. Structure of the thesis3
Chapter 2. Background Theory5 2.1. Graph theory5 2.1.1. Undirected graphs5 2.1.2. Graph operations8 2.1.3. Directed graphs 10 2.2. Zero forcing in graphs 11 2.2.1. Bounds on the zero forcing number 17 2.2.2. Zero forcing in directed graphs 21 2.2.3. Variants of zero forcing problem 23 2.3. Power domination in graphs 24 2.3.1. Bounds on the power domination number 26 2.3.2. Power domination in directed graphs 31 2.3.3. Variants of power domination problem 32 2.4. Zero forcing and power domination 33
Chapter 3. Zero forcing and power domination in iterated line digraphs 35 3.1. An alternate interpretation 35 3.2. Zero forcing and power domination in de Bruijn digraphs 38 3.3. Zero forcing and power domination number in Kautz digraphs 41 3.4. Zero forcing in iterated line digraphs 45 3.5. Power domination in iterated line digraphs 48 3.6. Applications 52
Chapter 4. Zero forcing and power domination in butterfly networks 55 4.1. An alternate interpretation 55 4.2. Zero forcing in butterfly network 57
ix x CONTENTS
4.2.1. The upper bound for Z(BF(r)) 58 4.2.2. The lower bound for Z(BF(r)) 62 4.3. Power domination in butterfly network 71
Chapter 5. Zero forcing number in terms of girth and minimum degree 77 5.1. Proof of the conjecture 77 5.2. Better bounds for large girth and minimum degree 82
Chapter 6. Power domination and resolving-power domination in graphs 85 6.1. Power domination-subgraph relation 85 6.1.1. Silicate networks 88 6.1.2. Rhenium trioxide lattice 90 6.2. Resolving-power dominating set 94 6.2.1. Complexity results of resolving-power domination problem 97 6.2.2. Resolving-power domination in trees 101
Chapter 7. Conclusion 105
Bibliography 111 List of Figures
2.1 A graph G with zero forcing set S = {4, 6} 11 2.2 Exhibition of the colour change rule 12
2.3 Graphs G1 and G2 specified in Theorem 2.2.21 19 2.4 A directed graph with empty set as the zero forcing set 23 2.5 A graph G with power dominating set S = {5, 6} 26 2.6 Exhibition of the power dominating steps 27
3.1 A directed graph G~ . 36 3.2 Illustration of the construction of the power dominating set S for d = 5 and ` = 7. For
the two columns (a1, . . . , a5) = (1, 3, 4, 4, 2) and (a1, . . . , a5) = (3, 1, 0, 2, 4) we show
the elements of S (black squares), and we indicate for the sets X(a1, . . . , a6) (enclosed by rectangles) the elements of S having them as their out-neighbourhood. 40
4.1 The butterfly network BF(4). 58 4.2 The set S(4) indicated by squares. 59 4.3 Labelling of vertices and structure of adjacency matrix in BF(3). 65
4.4 The second level of the recursion for Ar. 67 4.5 The construction of K˜ for r = 4 and i = 1. Here K+ = {2, 6}, K0− = {9, 12}, ˜ − ˜ + ˜ − 00+ K1 = {14, 18, 46, 50}, K1 = {21, 24, 53, 56}, K2 = {76, 77, 80}, K2 = ∅. 69 4.6 The critical sets in different layers of BF(4) . For convenience, we identify the binary r Pr i−1 vector x = (x1, . . . , xr) ∈ {0, 1} with the number i=1 xi2 . 72
6.1 WK(3, 3) containing three copies of WK(3, 2). 87 6.2 Illustration of Step 1 of the algorithm on S(2) 89 6.3 Step 2 and Step 3 of the algorithm 89 6.4 Subgraph X of Theorem 6.1.8 90 6.5 Unit cell of Rhenium trioxide 91 6.6 Square bullets depict (i) Metric basis (ii) Minimum power dominating set (iii) Minimum metric-locating-dominating set and (iv) Minimum resolving-power-dominating set of G. 95
xi xii LIST OF FIGURES
6.7 Variable gadget for each xi 98
6.8 Clause gadget for each Cj 98
6.9 Reduction ofx ¯1 ∨ x2 ∨ x¯3 99 6.10Variable gadget for bipartite graphs 101
7.1 Critical sets in 3 × 3 grid 105 Abstract
This thesis investigates the dynamic colouring of vertices in a graph G. Dy- namic colouring of the vertices in a graph has seen a rise in application and relations to well studied graph theoretic parameters in recent years. By dy- namic colouring, we mean a colouring of the vertices in a graph which may propagate (change the colour of) vertices that were not initially coloured in the graph. Of these dynamic colourings, and of relation to this thesis, we highlight that zero forcing sets and power dominating sets, and the associated graph invariants known as the zero forcing number and power domination number respectively, are of particular interest. Both zero forcing number and power domination number have been shown to relate to well studied graph invariants such as the domination number, the total domination number, the connected domination number, the path cover number, the chromatic number, the independence number, and the minimum rank. Moreover, it has been established that both these problems lie within the class of NP- complete decision problems. Thus, it is desirable to find easily computable bounds for the zero forcing number and power domination number.
In this thesis, we give an alternate interpretation of zero forcing and power domination problems as set covering problems. These interpretations were used in solving both these problems for de Bruijn and Kautz digraphs.
Moreover the same technique was used to extend the results to obtain these graph invariants for iterated line digraphs. The zero forcing problem is solved for butterfly networks and bounds were obtained in the case of power domi- nation number.
Another important result in this thesis is settling the conjecture of Davila and Kenter connecting zero forcing number, girth and minimum degree. A general lower bound technique for power domination number by looking at subgraphs satisfying certain conditions is described and used effectively in
finding the power domination number of some specific graphs. In the last section of the thesis, we introduce a new variant of power domination problem called resolving-power domination problem. We show that the problem is
NP-complete. Nomenclature
Table 0.1. General Notation
N The natural numbers Z The integers Zd The cyclic group of order d R The real numbers [j] {1, 2, . . . , j}
G An undirected graph mrF (G) The minimum rank of G over field F MF (G) The maximum nullity of G over field F P(G) The path number of G Z(G) The zero forcing number of G
γp(G) The power domination number of G γ(G) The domination number of G dim(G) The metric dimension of G
ηp(G) The resolving-power domination number of G G~ A directed graph mrF (G~ ) The minimum rank of G~ over field F MF (G~ ) The maximum nullity of G~ over field F P(G~ ) The path number of G~ Z(G~ ) The zero forcing number of G~
γp(G~ ) The power domination number of G~
xv
CHAPTER 1
Introduction
Many real world situations can conveniently be described by means of a diagram consisting of a set of points (nodes) together with lines (edges) joining certain pairs of these points. Points joined by a line are referred to as neighbours. For example, the points might be communication centres, with lines representing communication links between the centres. Note that we are interested mainly in whether or not two given points are joined by a line; the manner in which they are joined is immaterial. A mathematical abstraction of situations of this type gives rise to the concept of a graph. Graph theory is the study of graphs. The wide variety of applications has increased the popularity of graph theory in many fields like natural sciences, engineering, genetics and even in soft sciences like psychology and sociology.
This thesis deals with the study of two minimisation problems - zero forcing problem and power domination problem. In the zero forcing problem, we are given a graph G and the goal is to find a minimum-sized set of nodes S that covers all of the nodes, where a node v is covered if
(1) v is an element of the set S, or
(2) v has a neighbour u such that u and all of its neighbours except v are covered.
Rule (2) is called the colour change rule and is applied iteratively. The cardinality of any such minimum sized set is called the zero forcing number.
The power domination problem can be thought as a variant of zero forcing problem. In this case, we are given a graph G and the goal is to find a minimum-sized set of nodes S that covers all of the nodes, where a node v is covered if
(1) v is in the closed neighbourhood of the set S, or
(2) v has a neighbour u such that u and all of its neighbours except v are covered.
The cardinality of any such minimum sized set is called the power domination number.
1 2 1. INTRODUCTION
Although the power domination and the zero forcing problem are closely related, they were introduced independently and have been studied by different groups of researchers using different approaches until Benson et al. [11] connected the concepts and combined techniques and results from both to obtain new results on several graph products.
Both the zero forcing problem and the power domination problem are modelled as graph theory problems from a real world scenario.
The zero forcing process was introduced in [16] and used in [15] as a criterion for quantum controllability of a system. Independently, the authors in [3] have introduced it to bound the minimum rank, or equivalently, the maximum nullity of a graph G. Given an n-vertex graph
G, let M(G) denote the maximum nullity over all symmetric real-valued matrices A whose zero-nonzero pattern of the off-diagonal entries is described by the graph G. This means that for i 6= j, the entry Aij is non-zero if and only if ij is an edge in G, whereas the diagonal entries are chosen freely. The minimum rank of G is n − M(G). This parameter has been extensively studied in the last fifteen years, largely due to its connection to inverse eigenvalue problems for graphs, singular graphs, biclique partitions and other problems. Among several tools introduced to study the minimum rank, the zero forcing number has the advantage that its definition is purely combinatorial.
The motivation for the power domination problem comes from modelling electrical power systems. Electric power companies need to continually monitor their system state as defined by a set of state variables (for example, the voltage magnitude at loads and the machine phase angle at generators). One method of monitoring these variables is to place phase measurement units (PMUs) at selected locations in the system. Because of the high cost of a PMU, it is desirable to minimize their number while maintaining the ability to monitor (observe) the entire system. A system is said to be observed if all of the state variables of the system can be determined from a set of measurements (e.g., voltages and currents). This was modelled as a graph theory problem by Haynes et al. in [55].
Both zero forcing problem and power domination problem have been extended to digraphs in [9] and [1] respectively. 1.2. STRUCTURE OF THE THESIS 3
1.1. Contributions of the thesis
The main objective of this research is to exploit the similarity between zero forcing and power domination problem in order to obtain an efficient technique of lower bound for both these parameters and use them effectively in obtaining the zero forcing number and power domination number in several classes of graphs. To this end the following results were obtained:
• Zero forcing number and power domination number for de Bruijn and Kautz digraphs.
• Bounds on the zero forcing number and power domination number of iterated line
digraphs.
• Zero forcing number and minimum rank of butterfly networks.
• Proof of the conjecture connecting zero forcing number, girth and minimum degree.
• A lower bound technique for power domination number in relation to subgraphs of
the graph satisfying certain condition.
• A new variant of power domination called resolving-power domination.
1.2. Structure of the thesis
This thesis is structured as follows.
Chapter 2 on page 5: Background Theory- This chapter contains relevant and useful con- cepts from graph theory and an extended literature review of zero forcing and power domination problems needed for the thesis.
Chapter 3 on page 35: Zero forcing and power domination in iterated line digraphs-
In this chapter, we obtain the zero forcing and power domination numbers of de Bruijn and
Kautz digraphs by interpreting them as set cover problems. The later sections of the chapter extends the results obtained for de Bruijn and Kautz digraphs to iterated line digraphs.
Chapter 4 on page 55: Zero forcing and power domination in butterfly networks-
In this chapter, we obtain the zero forcing number of butterfly networks. A lower bound and upper bound for power domination number for butterfly networks is provided in the second section.
Chapter 5 on page 77: Zero forcing number in terms of girth and minimum degree
- In this chapter we resolve the conjecture connecting zero forcing number, girth and minimum degree posed by Davila and Kenter in [24]. 4 1. INTRODUCTION
Chapter 6 on page 85: Power domination and resolving-power domination in graphs- In the first section of this chapter, a lower bound technique for power domination is introduced and is used effectively in finding the power domination numbers of certain chemi- cal structures. In the second section, a new variant of power domination called resolving-power domination is introduced. We show that the problem is NP-complete and obtain bounds for trees.
Chapter 7 on page 105: Conclusion- In this chapter, we summarise the results and provide a list of open problems that arise from this thesis.
In this thesis, all original results are indicated by the symbol ♦ and end of proofs are marked by symbol . CHAPTER 2
Background Theory
This chapter presents some basic definitions, concepts and facts from graph theory that we would use in this thesis. For terms not defined here, readers may refer to [14, 50, 70]. We also give a literature review of zero forcing and power domination problems.
2.1. Graph theory
In mathematics, graph theory is the study of graphs, which are mathematical structures used to model pairwise relations between objects. A graph consists of a finite set of vertices
(points, nodes), a finite set of edges (lines, linkages) and a rule that defines which edge joins which pair of vertices. It is customary to represent a graph by means of a diagram and refer to the diagram as the graph. Graphs in graph theory portray topological quantities, patterns and relationships. Structures that can be represented as graphs exist in nature and many problems of practical interest can be represented by graphs. For example, the vertices could be different cities in a country and the edges could be the roads joining them.
2.1.1. Undirected graphs. An undirected graph (or simply graph) G is a finite non- empty set of objects called vertices together with a set of unordered pairs of distinct vertices of G, called edges. The vertex set and the edge set of G are denoted by V (G) and E(G) respectively. The order of a graph G is the number of vertices in G, denoted by n = |V (G)|; the number of edges in G is called the size of G and denoted by m = |E(G)|. If e = (u, v) is an edge of G, we say that u and v are adjacent and that each vertex is incident with e. We may also say that two vertices u and v are neighbours, if they are adjacent. The open neighbourhood of a vertex v, denoted N(v), is given by N(v) = {u ∈ V :(u, v) ∈ E}. The closed neighbourhood of a vertex v, denoted N[v], is given by N[v] = N(v) ∪ {v}. The idea of neighbourhood can be generalized to include set of vertices as follows. For a set S, let N(S) = ∪v∈SN(v) − S and N[S] = N(S) ∪ S denote the open and close neighbourhood of S respectively.
5 6 2. BACKGROUND THEORY
An edge with identical ends is called a loop and an edge with distinct ends a proper edge.
A multi-edge is a collection of two or more edges having identical endpoints. A graph is simple if it has no loops or multi-edges. A simple graph on n vertices, in which each pair of distinct vertices is joined by an edge is called a complete graph, denoted by Kn. A graph is finite if both its vertex set and edge set are finite. We call a graph with just one vertex trivial and all other graphs nontrivial.A null graph is a graph whose vertex and edge sets are empty.
The degree of a vertex v in G, denoted by degG(v), is the number of edges incident with v. The maximum degree of G, denoted by ∆(G), is the maximum degree over all vertices of G, while the minimum degree of G, denoted by δ(G), is defined as the minimum degree over all vertices of G. A graph G is called r-regular if degG(v) = r, for each v ∈ V (G). A vertex v is called a pendant vertex if degG(v) = 1 and an isolated vertex if degG(v) = 0. An edge e in a graph G is called a pendant edge if it is incident with a pendant vertex.
A walk W of length ` in a graph G is an alternating sequence of vertices and edges W = v0, v0v1, v1, v1v2, v2, . . . , v`−2, v`−2v`−1, v`−1 containing ` vertices and ` − 1 edges. A walk in which no edges are repeated is called a trail and a walk of length ` in which no vertices are repeated is called a path P`. A walk is closed if it has positive length and its origin and terminus are the same. A closed path is a cycle. A cycle of length ` is called a `-cycle, denoted C`; a `-cycle is odd or even according as ` is odd or even. A 3-cycle is often called a triangle. The length of the smallest cycle in a graph G is called the girth of G. If the graph G contains no triangles, alternatively, if G has girth g ≥ 4, then G is called triangle-free graph.
A trail in G is called an Eulerian trail if it includes every edge of G.A tour of G is a closed walk of G which includes every edge of G at least once. An Eulerian tour of G is a tour which includes each edge of G exactly once. A graph G is said to be Eulerian if it has an Eulerian tour. A graph is Eulerian if and only if it has no vertex of odd degree.
A Hamiltonian path in a graph G is a path which contains every vertex of G.A Hamiltonian cycle in a graph G is a cycle which contains every vertex of G. A graph G is called Hamiltonian if it has a Hamiltonian cycle.
The distance distG(u, v) ( or simply dist(u, v) if the graph G is clear in the context) between two vertices u and v in a graph G is the length of the shortest path between u and v. The 2.1. GRAPH THEORY 7 eccentricity of a vertex v, denoted by e(v), in a graph G, is the longest distance from a vertex v to any other vertex in G, that is, e(v) = max{distG(v, u): u ∈ V (G)}. The diameter, denoted by λ (or D) of a graph G is the maximum eccentricity of all the vertices in G, that is,
λ(G) = max{e(v): v ∈ V }. The radius of G, written as rad(G), is the minimum eccentricity of any vertex in G, that is, rad(G) = min{e(v): v ∈ V }. A vertex whose eccentricity is equal to the radius is called a central vertex or simply a center.
A graph S is a subgraph of G if V (S) ⊆ V (G) and E(S) ⊆ E(G). If V (S) = V (G) then
S is a spanning subgraph of G. If V (S) is a proper subset of V (G) or E(S) is a proper subset of E(G) then S is called a proper subgraph of G. If U is a nonempty subset of the vertex set
V (G) of a graph G, then the subgraph hUi of G induced by U is the graph having vertex set U and whose edge set consists of those edges of G incident with two elements of U. A subgraph
H of G is called vertex-induced or simply induced if H = hUi for some subset U of V (G). By deleting from G all loops and, for every pair of adjacent vertices, all but one edge joining them, we obtain a simple spanning subgraph of G, called the underlying simple graph of G.
A vertex u is said to be connected to a vertex v in a graph G if there exists a u−v path in G.
A graph G is connected if every two of its vertices are connected. A graph that is not connected is called disconnected. Connection is an equivalence relation on the vertex set V . Thus there is a partition of V into nonempty subsets V1,V2 ...Vω such that two vertices u and v are connected if and only if both u and v belong to the same set Vi. The subgraphs hV1i, hV2i ... hVωi are called the components of G. If G has exactly one component, G is connected; otherwise G is disconnected. The number of components of G is denoted by ω(G). A cut-vertex is a vertex whose removal increases the number of components. A cut-edge is an edge whose removal increases the number of components.
A graph G1 is isomorphic to a graph G2 if there exists a one-to-one mapping φ, called an isomorphism from V (G1) onto V (G2) such that φ preserves adjacency and non-adjacency; that is, uv ∈ E(G1) if and only if φ(u)φ(v) ∈ E(G2). As structural properties of graphs are determined by their adjacency relations, we can prove that each G1 and G2 are not isomorphic by finding some structural properties of one that are not true of the other. If they have different vertex degrees, different sizes of the largest clique or smallest cycle, and so on, then they cannot 8 2. BACKGROUND THEORY be isomorphic, because these properties are preserved by isomorphism. On the other hand, no known list of common structural properties implies that G1 is isomorphic to G2. An automorphism of G is a permutation of V (G), that is, an isomorphism from G into G.
A graph is vertex transitive if for every pair u, v ∈ V (G) there is an automorphism that maps u to v. A graph G is edge transitive if for every pair e, f ∈ E(G) there is an automorphism that maps e to f. For example, every cycle of size n is not only vertex transitive but also edge transitive.
An acyclic graph is one that contains no cycles. An acyclic graph is also called a forest.A tree is a connected acyclic graph, denoted T . Every tree T has at least two vertices of degree one; indeed, two end-vertices of any longest path in a tree are of degree one. A tree in which one vertex (called the root), is distinguished is called a rooted-tree. In a rooted-tree any vertex of degree one, unless it is the root, is called a leaf and the other vertices are called internal vertices.A complete binary tree is defined as a tree in which there is exactly one vertex of degree two, and each of the remaining vertices is of degree one or three. Since the vertex of degree two is distinct from all other vertices, this vertex serves as a root. Thus every complete binary tree is a rooted tree.
A bipartite graph is one whose vertex set can be partitioned into two subsets X and Y , so that each edge has one end in X and the other end in Y ; such a partition (X,Y ) is called a bipartition of the graph. A graph is bipartite if and only if it contains no odd cycle. A complete bipartite graph is a bipartite graph G(X ∪ Y,E) in which each vertex of X is joined by an edge to each vertex of Y . If |X| = s and |Y | = t, a complete bipartite graph is denoted by Ks,t.
2.1.2. Graph operations. Let G = (V,E) be a graph. The operation of adding the vertex u, u∈ / V to a graph G, yields a new graph G ∪ {u} with vertex set V ∪ {u} and edge set
E. Note that the new vertex u has no neighbours.
The operation of deleting the vertex v from a graph G, not only removes the vertex v but also removes every edge incident to v. The resulting graph is denoted G − v. The operation of adding an edge e (or uv) to a graph G = (V,E) joining the vertices u and v yields a new graph with vertex set V and edge set E ∪ {e} (or E ∪ {uv}), which is denoted G ∪ {e} (or G ∪ {uv}). 2.1. GRAPH THEORY 9
The operation of deleting an edge e (or uv) from a graph G = (V,E) removes only that edge.
The resulting graph is denoted as G − e (or G − uv).
A subdivision of an edge e = uv in a graph is the graph obtained from G by replacing e by the path uwv where w is a new vertex of degree 2. The resulting graph is denoted as Ge. The operation of edge contraction produces a graph with edge set E \{e} but with a vertex set obtained by identifying the vertices defining e in G, thus creating a new single vertex where the latter inherits all the adjacencies of the pair of replaced vertices, without introducing loops or multiple edges. The resulting graph is denoted as G/e.
The graph union of two graphs G and H is the graph G ∪ H whose vertex-set and edge-set are the disjoint unions, respectively, of the vertex-sets and the edge-sets of G and H. The complement G of a graph G also has V (G) as its vertex set, and two vertices are adjacent in
G if and only if they are not adjacent in G. For disjoint graphs G and H, G + H is the graph obtained from G and H by joining every vertex of G with every vertex of H.
The Cartesian product of two graphs G and H, denoted G × H is defined as the graph with vertex set V (G) × V (H) and edge set {((u, v), (x, y)) : u = x and vy ∈ E(H) or ux ∈ E(G) and v = y}. The rth hypercube, Qr is defined inductively by Q1 = K2 and Qr+1 = Qr × K2.
The Cartesian product of Ps × Pt, Ps × Cs and Cs × Ct are often called s × t grid, cylinder and torus respectively.
The direct product of two graphs G and H, denoted G · H is defined as the graph with vertex set V (G) × V (H) and edge set {((u, v), (x, y)) : ux ∈ E(G) and vy ∈ E(H)}.
The strong product of two graphs G and H, denoted G ⊗ H, is the graph with vertex set
V (G) × V (H) and edge set {((u, v), (x, y)) : u = x and vy ∈ E(H) or ux ∈ E(G) and v = y or ux ∈ E(G) and vy ∈ E(H)} .
The lexicographic product of two graphs G and H, denoted G ∗ H, is the graph with vertex set V (G) × V (H) and edge set {((u, v), (x, y)) : ux ∈ E(G) or u = x and vy ∈ E(H)} .
The corona of G with H, denoted G ◦ H, is the graph of order |G||H| + |G| obtained by taking one copy of G and |G| copies of H, and joining all the vertices in the ith copy of H to the ith vertex of G. 10 2. BACKGROUND THEORY
2.1.3. Directed graphs. An edge between two vertices creates a connection in two op- posite senses at once. Assigning a direction makes one of these senses forward and the other backward. Viewing direction as an edge attribute is partly motivated by its impact on com- puter implementations of graph algorithms. A directed edge (or arc) is an edge e, one of whose endpoints is designed as the tail, and whose other endpoint is designed as the head. A directed edge is said to be directed from its tail to its head. A digraph (or directed graph) is a graph each of whose edges is directed, denoted by G~ = (V,A) with vertex set V = V (G~ ) and arc set
A = A(G~ ). An arc in the form (u, u) is called a loop.A simple digraph is a directed graph that does not allow loops and multiple edges, and a loop digraph is a directed graph that allow loops but not multiple edges. For each vertex v the open out-neighbourhood of v is N out(v) = {u ∈ G~ V (G~ ): vu ∈ A(G~ )} and the open in-neighbourhood of v is N in(v) = {w ∈ V (G~ ): wv ∈ A(G~ )}. G~ The closed out-neighbourhood and the closed in-neighbourhood of v are respectively defined as
N out[v] = N out(v) ∪ {v} and N in[v] = N in(v) ∪ {v}. Analogously, for each vertex v, Aout(v) = G~ G~ G~ G~ G~ {vw ∈ A(G~ ): w ∈ N out(v)} and Ain(v) = {uv ∈ A(G~ ): u ∈ N in(v)}. For a vertex v, the out- G~ G~ G~ degree is degout(v) = |N out(v)| = |Aout(v)| and the in-degree is degin(v) = |N in(v)| = |Ain(v)|. G~ G~ G~ G~ G~ G~ ~ ~ out in A digraph G is r-regular if for every vertex v in G, degG~ (v) = degG~ (v) = r. The maximum ~ out ~ out ~ out-degree of G is denoted as ∆ (G) and it is defined as max{degG~ (v): v ∈ V (G)}. Anal- ~ in ~ in ~ ogously, the maximum in-degree of G is ∆ (G) = max{degG~ (v): v ∈ V (G)}. The minimum ~ out ~ out ~ out-degree of G is denoted as δ (G) and it is defined as min{degG~ (v): v ∈ V (G)}. Analo- ~ in ~ in ~ ~ gously, the minimum in-degree of G is δ (G) = min{degG~ (v): v ∈ V (G)}. When the graph G can be deduced from the context we will omit the sub-indices.
Let u, v be two vertices in a digraph G~ = (V,A). A walk of length n from u to v is a sequence of vertices u = x1, . . . , xn = v such that xixi+1 ∈ A(G~ ) for every i = 1, . . . n − 1. A trail is a walk that does not repeat arcs, and a path is a trail that does not repeat vertices. A walk is a circuit if u = v and a circuit that does not repeat vertices is a cycle. The distance between two different vertices u and v in a digraph G~ = (V,A) is the length of a shortest path ~ from u to v and it is denoted as distG~ (u, v) or simply dist(u, v) if the graph G is clear in the context. The maximum distance between any two different vertices in G~ is the diameter of G~ and it is denoted as D(G~ ). 2.2. ZERO FORCING IN GRAPHS 11
For a digraph G~ = (V,A), the line digraph of G~ is the digraph L(G~ ) where V (L(G~ )) = A(G~ ) and there is an arc from a vertex x to a vertex y in L(G~ ) if and only if x = uv and y = vw for some vertices v, w in V (G~ ). Iterated line graphs are naturally defined by the following rules:
L0(G~ ) = G~ and for every integer i > 0, Li(G~ ) = L(Li−1(G~ )). Two famous examples of line digraphs are de Bruijn and Kautz digraphs, discussed in detail in Chapter3.
2.2. Zero forcing in graphs
The notion of a zero forcing set, as well as the associated zero forcing number, of a simple undirected graph was introduced by the “AIM Minimum Rank Special Graphs Work Group” in [3] to bound the minimum rank of associated matrices for numerous families of graphs.
For a red/blue colouring of the vertex set of a simple undirected graph G = (V,E) consider the following colour-change rule: a red vertex u is converted to blue if it is the only red neighbour of some blue vertex v. We call such a blue vertex v a forcing vertex or in other words, we say v forces u, and write v → u. Given a two-colouring of G, the derived set is the set of blue vertices obtained by applying the colour-change rule until no more changes are possible. A zero forcing set for G is a subset of vertices S ⊆ V such that if initially the vertices in S are coloured blue and the remaining vertices are coloured red, then the derived set is the complete vertex set V .
The minimum cardinality of a zero forcing set for the undirected graph G is called the zero forcing number of G, denoted by Z(G).
5 6
4 3
1 2
Figure 2.1. A graph G with zero forcing set S = {4, 6}
In Figure 2.1, the set S = {4, 6} is a zero forcing set because at the first time step, vertex
6 forces vertex 3, vertex 3 forces vertex 2 in the second time step, vertex 2 forces vertex 1 at third time step and lastly, vertex 4 forces vertex 5 in the fourth time step. See Figure 2.2. 12 2. BACKGROUND THEORY
5 6 5 6
4 3 4 3
1 2 1 2 (a) First time step (b) Second time step 5 6 5 6
4 3 4 3
1 2 1 2 (c) Third time step (d) Fourth time step
Figure 2.2. Exhibition of the colour change rule
The term “zero forcing” is based on an algebraic property of these sets. Consider a vector with the entries corresponding to the vertices of a graph. Further, assume the entries corre- sponding to a set of vertices in a zero forcing set for the graph are equal to zero. The zero forcing property of the set guarantees that such a vector is in the kernel of the adjacency matrix of the graph only if the vector is the zero vector. The term “zero forcing” refers to the fact that the remaining entries of the vector are forced to be zero for the vector to be in the kernel of the adjacency matrix [3, 12].
Let F be a field, and denote by Sn(F ) the set of symmetric n × n matrices over F . For a simple undirected graph G(V,E) with vertex set V = {1, . . . , n}, let S(F,G) be the set of matrices in Sn(F ) whose non-zero off-diagonal entries correspond to edges of G, i.e.,
S(F,G) = {A ∈ Sn(F ): i 6= j =⇒ (ij ∈ E(G) ⇐⇒ aij 6= 0)}. 2.2. ZERO FORCING IN GRAPHS 13
The minimum F -rank of an undirected graph G is defined as the minimum rank over all matrices
A in S(F,G):
mrF (G) = min {rank(A): A ∈ S(F,G)} .
The maximum F -corank of an undirected graph G is defined as the maximum corank over all matrices A in S(F,G):
MF (G) = max {corank(A): A ∈ S(F,G)} .
If the index F is omitted then it is understood that F = R. The minimum rank problem and maximum corank problem for an undirected graph G is to determine mr(G) and M(G) respectively (and more generally, mrF (G) and MF (G)), and has been studied intensively for more than ten years. See [35, 36] for surveys of known results and an extensive bibliography.
It is a crucial observation that for a zero forcing set S and a matrix A ∈ S(F,G), the rows of A that correspond to the vertices in V \S must be linearly independent, so rank(A) > n−|S|, F and consequently mr (G) > n − Z(G), or equivalently,
F Theorem 2.2.1 ([3]). For any graph G and field F , Z(G) > M (G).
Based on this insight, the authors of [3] determined MF (G) and established equality in
Theorem 2.2.1 for many interesting graph classes. In [63], this equality is proved for block- clique graphs and unit interval graphs. See Table 2.1 for zero forcing number of various graph classes obtained. The American Institute for Mathematics maintains the minimum rank graph catalogue [59] in order to collect the information about minimum rank problem for various graph classes.
In general, characterising equality for Theorem 2.2.1 is difficult. Addressing this issue, the authors in [3] posed the following question which is still open.
Question 1 ([3]). What is the class of graphs G for which Z(G) = MF (G) for some field F ?
Prior to the development of zero forcing number, another graph parameter was studied in conjunction with maximum nullity. A path cover of an undirected graph G is a set of vertex disjoint induced paths that cover all the vertices of G. The path cover number of an undirected 14 2. BACKGROUND THEORY graph G, P(G), is the minimum size of a path cover. An undirected graph G is a graph of two parallel paths if there exist two independent induced paths of G that cover all the vertices of G and such that G can be drawn in the plane in such a way that the two paths are parallel and the edges (drawn as segments, not curves) between the two paths do not cross. A simple path is not considered to be such a graph. A graph that consists of two connected components, each of which is a path, is considered to be such a graph.
The relation between zero forcing number and path cover was brought out in [7] where the concept of forcing chain, a sequence of forcing vertices obtained through iterative application of the colour-change rule, was introduced. The vertices in a forcing chain induce a path in G because the forcing process in a forcing chain occur chronologically in the order of the chain.
This implies that if S ⊂ V (G) is a zero forcing set of G, then S induces a path cover for G.
Theorem 2.2.2 ([7]). For any graph G, Z(G) ≥ P(G).
The most important family of graphs that satisfy Z(G) = P(G) are trees [3] and block-cycle graphs [87]. However, the difference between the two parameters could be arbitrarily large, for example,
P(Kn) = dn/2e < n − 1 = Z(Kn).
Since its introduction the zero forcing number has been studied for its own sake as an interesting graph invariant [7,8, 32, 73]. Similarly in this thesis, we treat zero forcing problem as a graph theory problem in its own right and not in relation with minimum rank problem. The zero forcing problem is closely related to a number of topics of current interest. Physicists have independently studied the zero forcing parameter, referring to it as the graph infection number, in conjunction with the control of quantum systems [82]. Current applied and theoretical research on social networks has focussed on diffusion processes (spread of influence) and their characteristics that led to the introduction of models such as threshold model [66], very closely related to zero forcing model.
It is very clear that the zero forcing set of an undirected graph G should have at least δ vertices otherwise the forcing process cannot begin. 2.2. ZERO FORCING IN GRAPHS 15
Graph G order Z(G) Reference r r−1 Qr (hypercube) 2 2 [3] Ks × Pt st s [3] Ps × Pt st min{s, t} [3] Cs × Pt st min{s, 2t} [3] Ks × Kt st st − s − t + 2 [3] Cs × Kt, s ≥ 4 st 2t [3] Ks × Kt, t ≥ 2 st + t st − 1 [3] Cs × Ct, s = t and s is odd st 2s − 1 [11] Cs × Ct, s 6= t or s = t and s is even st 2s [11] Ps ⊗ Pt st s + t − 1 [3] P2s+1 · Kt 2st + t (2s + 1)t − 4s [63] Ks · Kt st st − 4 [63] Cn, n ≥ 5 n n − 3 [3] n(n−1) n2−3n+4 L(Kn) 2 2 [3] L(T ), T a tree and ` = # pendant vertices of T |T | − 1 ` − 1 [3] Tree T |T | P(T )[3] Block-cycle graph G |G| P(G)[87] Unicyclic graph G |G| P(G)[87] Block clique graphs G |G| M(G)[63] Unit interval graphs G |G| M(G)[63] Table 2.1. Zero forcing number of various graph classes
Theorem 2.2.3 ([3]). For any graph G of order n ≥ 2,
Z(G) ≥ δ.
Graphs that satisfy equality in Theorem 2.2.3 are cycles, paths and complete graphs amongst others.
Theorem 2.2.4. Let G be a connected graph of order n ≥ 2. Then
(1) [3, 33] Z(G) = 1 if and only if G = Pn. (2) [79] Z(G) = 2 if and only if G is a graph of two parallel paths.
(3) [3, 33] Z(G) = n − 1 if and only if G = Kn.
Table 2.1 gives the summary of the results that has been obtained for various graph classes.
In [7], it was proved that there exists no graph of order greater than one that has a unique minimum zero forcing set. 16 2. BACKGROUND THEORY
Theorem 2.2.5 ([7]). For any connected graph G of order more than one, no vertex is in every optimal zero forcing set of G.
The interesting consequence of Theorem 2.2.5 is the following result.
Theorem 2.2.6 ([7]). If G is a non trivial graph, then G does not have a unique minimal zero forcing set.
To validate the above theorem, in Figure 2.2, both {3, 5} and {4, 6} are zero forcing sets.
Theorem 2.2.7 ([3]). For any graphs G and H,
Z(G × H) ≤ min{Z(G)|H|, Z(H)|G|}.
Theorem 2.2.8 ([3]). For any graphs G and H,
Z(G ◦ H) ≤ Z(G)|H| + Z(H)|G| − Z(G) Z(H).
Recently, the authors in [44] characterised all connected graphs on n vertices with zero forcing number at least n − 2. A graph G is complement reducible (cograph) if every induced subgraph of G with at least two vertices is either disconnected or is the complement of a disconnected graph.
Theorem 2.2.9 ([44]). A graph G with Z(G) = n − 2 is a cograph.
Theorem 2.2.10 ([44]). Every connected graph G with Z(G) ≥ n − 2 can be constructed as follows. Pick two, not necessarily distinct, graphs from the list below and connect all vertices from one to the other.
(1) Kn;
(2) The complement of Kn; (3) Disjoint union of two complete graphs;
(4) A connected graph H on n vertices with zero forcing number n − 2.
In [17], a relationship between zero forcing number of graph G and number of edges of G is found. 2.2. ZERO FORCING IN GRAPHS 17
Theorem 2.2.11 ([17]). For a graph G with n vertices, if Z(G) ≤ k then,
k + 1 |E(G)| ≤ kn − . 2
The bound is tight in the case of path Pn as Z(Pn) = 1 with E(Pn) = n − 1. The bound is also tight if G is an outerplanar graph. An interesting consequence that follows from the above theorem is the following result.
m Theorem 2.2.12 ([17]). For a graph G with n vertices and m edges, Z(G) > n , in particular Z(G) is greater than half of the average degree of the graph.
This bound works best for graphs with highly irregular degree sequences.
2.2.1. Bounds on the zero forcing number. Theorem 2.2.1, Theorem 2.2.2 and The- orem 2.2.3 are the best general lower bounds we have on zero forcing number in literature.
However, there have been many attempts to improve the lower bounds by imposing certain conditions on the graph G.
Theorem 2.2.13 ([3]). If G is a strongly regular graph, then
j|V (G)|k Z(G) ≥ . 2
Theorem 2.2.14 ([13]). If G is bipartite and k-regular, then
Z(G) ≥ 2(k − 1).
Theorem 2.2.15 ([24]). Let G be a triangle-free graph with minimum degree δ ≥ 3. Then
Z(G) ≥ δ + 1.
Theorem 2.2.16 ([24]). Let G be a graph with girth g and minimum degree δ ≥ 2. Then,
Z(G) ≥ 2δ − 2. 18 2. BACKGROUND THEORY
Theorem 2.2.17 ([24]). Let G be a graph with δ ≥ 3 and a cut vertex v such that G \ v has a component with girth at least 5. Then,
Z(G) ≥ 3δ − 6.
Theorem 2.2.18 ([24]). Let G be a graph with girth g ≥ 5, minimum degree δ ≥ 3 and a cut edge e. Then,
Z(G) ≥ 4δ − 9.
Davila and Kenter in [24] conjectured that for graphs G with girth g ≥ 3 and minimum degree δ ≥ 2,
(2.2.1) Z(G) ≥ (g − 3)(δ − 2) + δ.
Gentner, Penso, Rautenbach, and Souzab [42], Gentner and Rautenbach [41] and Davila and Henning [22] have shown that inequality (2.2.1) is true for small girth g ≤ 10, while Davila and Kenter [24] have proven that inequality (2.2.1) is true for girth g ≥ 7 and sufficiently large minimum degree.
One of the contributions of this thesis is proving that inequality (2.2.1) is true for all graphs with girth g ≥ 11 and minimum degree δ ≥ 2. We shall discuss this in detail in Chapter4.
It is a well known fact that Z(Kn) = n − 1 and thus we have
Observation 1. Let G be a graph with non-empty components and order n. Then,
Z(G) ≤ n − 2,
whenever G 6= Kn.
Theorem 2.2.19 ([5]). For any graph G with maximum degree ∆ ≥ 1,
n∆ Z(G) ≤ . ∆ + 1
Theorem 2.2.20 ([5]). If G is connected with maximum degree ∆ ≥ 2, then
(∆ − 2)n + 2 Z(G) ≤ . ∆ − 1 2.2. ZERO FORCING IN GRAPHS 19
(a) Graph G1 (b) Graph G2
Figure 2.3. Graphs G1 and G2 specified in Theorem 2.2.21
It was shown that the only extremal graph satisfying the inequality in Theorem 2.2.19 is the complete graph K∆+1 of order ∆ + 1 [42] and that the only extremal graphs satisfying inequality in Theorem 2.2.20 are Cn, Kn and K∆,∆ [42, 73]
2 The authors in [41] have shown that the additive term in the inequality of ∆−1 in Theo- rem 2.2.20 could be done away with.
Theorem 2.2.21 ([41]). If G is a connected graph of order n and maximum degree ∆ ≥ 3, then (∆ − 2)n Z(G) ≤ ∆ − 1 if and only if G/∈ {K∆+1,K∆,∆,K∆−1,∆,G1,G2} where G1 and G2 are the two specific graphs illustrated in Figure 2.3.
The authors in [41] have been able to improve the upper bound stated in Theorem 2.2.20 by some lower order term for sub cubic graphs of girth at least 5.
Theorem 2.2.22 ([41]). If G is a connected graph of order n, maximum degree ∆ ≥ 3, and girth g ≥ 5, then n n Z(G) ≤ − + 2. 2 24 log2(n) + 6
Some upper bounds are obtained for special classes of graphs. In [34], zero forcing in line graphs were studied in which the authors introduced an edge equivalent notion of zero forcing, edge zero forcing. They made an observation that each edge zero forcing set of G corresponds to a zero forcing set of line graph of G, L(G). Using this observation, they proved the following theorem. 20 2. BACKGROUND THEORY
Theorem 2.2.23 ([34]). For a connected graph G,
Z(G) ≤ 2 Z(L(G)).
A connected graph G is called a cycle-tree if it consist of q vertex disjoint cycles that are connected by q − 1 edges. Thus, a cycle-tree with n vertices will have m = n + q − 1 edges and each edge between two cycles is a cut-edge. In [5], they obtained the following results.
Theorem 2.2.24 ([5]). Let G be a cycle-tree with q ≥ 1 cycles. Then,
Z(G) ≤ 2q.
Theorem 2.2.25 ([5]). Let G be a Hamiltonian graph with t ≥ 1 chords and n ≥ 4. Then,
Z(G) ≤ t + 1.
In general, the study of the zero forcing number is challenging for many reasons. First, it is difficult to compute exactly, as it is NP-hard [20]. Further, many of the known bounds leave a wide gap for graphs in general. For example, given the minimum and maximum degree of a graph, δ and ∆ respectively, the zero forcing number on a graph with n vertices can be as low
n∆ as δ (Theorem 2.2.3) and as high as ∆+1 (Theorem 2.2.19). It is thus important to devise a good lower bound technique that would enable us to compute zero forcing number of graphs.
We have made our attempts towards this end in Chapter5.
The next couple of results examine what happens to the zero forcing number under certain graph operations.
Theorem 2.2.26 ([32]). If G − e is the graph obtained from G by deleting an edge e = uv, then
Z(G) − 1 ≤ Z(G − e) ≤ Z(G) + 1.
Theorem 2.2.27 ([32]). If G/e is the graph obtained from G by contracting an edge e = uv, then
Z(G) − 1 ≤ Z(G/e) ≤ Z(G) + 1. 2.2. ZERO FORCING IN GRAPHS 21
Theorem 2.2.28 ([32]). If Ge is the graph obtained from G by subdividing an edge e = uv, then
Z(G) ≤ Z(Ge) ≤ Z(G) + 1.
It is evident that if the subdivision is carried out on an edge incident to degree 1 or 2 vertex, then there will be no change in the zero forcing number.
Theorem 2.2.29 ([32]). If G − v is the graph obtained from G by deleting a vertex v, then
Z(G − v) − 1 ≤ Z(G) ≤ Z(G − v) + 1.
2.2.2. Zero forcing in directed graphs. This notion was extended to digraphs in [9] with the same motivation i.e., to bound the minimum rank of a digraph. Recall that a simple digraph is a pair G~ = (V,A) where A ⊆ (V × V ) \{(i, i): i ∈ V }, and a loop digraph is a pair G~ = (V,A) where A ⊆ V × V . In fact, the definition of zero forcing in directed graphs is identical to definition in undirected graphs except for the colour-change rule. For a red/blue colouring of the vertex set of a digraph G~ consider the following colour-change rule:
(1) For a simple digraph G~ , a red vertex w is converted to blue if it is the only red
out-neighbour of some blue vertex u.
(2) For a loop digraph G~ , a red vertex w is converted to blue if it is the only red out-
neighbour of some vertex u.
We say u forces w and denote this by u → w. A vertex set S ⊆ V is called zero forcing if, starting with the vertices in S blue and the vertices in the complement V \ S red, all the vertices can be converted to blue by repeatedly applying the colour-change rule. The minimum cardinality of a zero forcing set for the digraph G is called the zero forcing number of G~ , denoted by Z(G~ ).
Remark 2.2.30. We note that Z(G) and Z(G~ ) is used to denote the zero forcing number of an undirected and a directed graph respectively.
Note that there is a difference in the forcing process in simple and loop directed graphs.
While in simple digraphs, only a blue vertex can force a colour change, in the loop digraphs, a red or a blue vertex can force a colour change. 22 2. BACKGROUND THEORY
The motivation of defining this way comes from bringing about a relation between zero forcing in directed graphs and minimum rank in directed graphs.
For a digraph G~ = (V,A) of order n, the qualitative class of G~ is the set of matrices n×n ~ {X ∈ R : i 6= j =⇒ (Xi,j 6= 0 ⇐⇒ (i, j) ∈ A)} if G does not have any loops, Q(G~ ) = n×n ~ {X ∈ R : Xi,j 6= 0 ⇐⇒ (i, j) ∈ A} if G has at least one loop.
Note that in the family of matrices described by a simple digraph, the diagonal entries of the matrix are free, whereas in the family of matrices described by a loop digraph, the the absence or presence of loops in the digraph describes the zero-nonzero pattern of the diagonal entries of the matrix. The asymmetric minimum rank problem for a digraph asks us to determine the minimum rank among all real matrices whose zero-nonzero pattern of entries is described by a given digraph. The maximum nullity of G~ is M(G~ ) = max{corank(X): X ∈ Q(G~ )}, and the minimum rank of G is mr(G~ ) = min{rank(X): X ∈ Q(G~ )}; clearly M(G~ ) + mr(G~ ) = |V (G~ )|.
The concept of zero forcing models the process to force zeros in a null vector of a matrix
X ∈ Q(G~ ), implying M(G~ ) ≤ Z(G~ )[58].
Just as it is possible for the maximum nullity of a digraph to be zero, it is possible for the empty set to be a zero forcing set for a digraph (note that both of these are impossible for an undirected graph). For example, in Figure 2.4, the digraph G~ = (V,A) with V = {1, 2} and A = {(1, 2), (2, 1)} can be a simple digraph as well as a loop digraph, and its zero forcing number depends on this: As a loop digraph its zero forcing number is 0, and as a simple digraph its zero forcing number is 1. We have 0 m12 ~ { : m12, m21 6= 0} if we interpret G as a loop digraph, m21 0 Q(G~ ) = m11 m12 ~ { : m12, m21 6= 0} if we interpret G as a simple digraph. m21 m22
Clearly, the maximum nullity is 0 for the loop digraph and 1 for the simple digraph and the bound M(G~ ) ≤ Z(G~ ) remains valid in all cases. 2.2. ZERO FORCING IN GRAPHS 23
1 2
Figure 2.4. A directed graph with empty set as the zero forcing set
Not much has been studied on the zero forcing set in the directed case. The only known results are listed below. The analogous results of Theorem 2.2.1 and Theorem 2.2.2 holds for directed version of the problem as well. The minimum rank problem or maximum nullity problem has been studied over fields F other than the real numbers.
Theorem 2.2.31 ([9]). For any digraph G~ ,
Z(G~ ) ≥ MF (G~ ).
Theorem 2.2.32 ([9]). For any digraph G~ ,
Z(G~ ) ≥ P(G~ ).
It is also shown in [9] that for any directed tree T , Z(T ) = P(T ).
2.2.3. Variants of zero forcing problem. In this section, we give an overview of the many variants of zero forcing that have been studied but that are beyond the scope of this thesis.
Propagation time [60]. Propagation time of a graph is introduced as the number of steps it takes for a zero forcing set to turn the entire graph blue. Propagation time of a zero forcing set was implicit in [16] and explicit in [82]. Chilakamarri et al. [21] determined the propagation time, which they call the iteration index, for a number of families of graphs including Cartesian products and various grid graphs. Control of an entire network by sequential operations on a subset of particles is valuable [82] and the number of steps needed to obtain this control
(propagation time) is a significant part of the process was the main motivation towards the study of this topic.
Positive semidefinite zero forcing set [7]. The positive semidefinite colour change rule is: Let B be the set consisting of all the blue vertices. Let W1,...,Wk be the sets of vertices of the k components of G \ B (note that it is possible that k = 1). Let w ∈ Wi. If u ∈ B 24 2. BACKGROUND THEORY
and w is the only red neighbour of u in G[Wi ∪ B], then change the colour of w to blue. The positive semidefinite zero forcing number of an undirected graph G, denoted by Z+(G), is the minimum of over all positive semidefinite zero forcing sets in G (using the positive semidefinite colour change rule).
Theorem 2.2.33 ([7]). Since any zero forcing set is a positive definite zero forcing set
Z+(G) ≤ Z(G).
k-forcing set [5]. Given a simple undirected graph G and a positive integer k, the k-forcing number of G, denoted Fk(G), is the minimum number of vertices that need to be initially coloured blue in a red/blue colouring of the vertex set of the graph G so that all vertices eventually become blue during the discrete dynamical process described by the following rule.
Starting from an initial set of blue coloured vertices and stopping when all vertices are coloured blue: if a blue coloured vertex has at most k red neighbours, then each of its red neighbours becomes blue. The minimum cardinality of this set is called k-forcing number of G. When k = 1, the problem reduces to zero forcing problem.
2.3. Power domination in graphs
Power domination was introduced by Baldwin et al. in [6], then described as a graph theoretical problem by Haynes et al. in [55]. The problem is motivated by the requirement for constant monitoring of power systems. Electric power companies need to continually monitor their systems state as defined by a set of state variables (for example, the voltage magnitude at loads and the machine phase angle at generators). One method of monitoring these variables is to place phase measurement units (PMUs) at selected locations in the system. Because of the high cost of a PMU, it is desirable to minimize their number while maintaining the ability to monitor (observe) the entire system. A system is said to be observed if all of the state variables of the system can be determined from a set of measurements (e.g., voltages and currents). 2.3. POWER DOMINATION IN GRAPHS 25
Let G = (V,E) be a graph representing an electric power system, where a vertex represents an electrical node (a substation bus where transmission lines, loads, and generators are con- nected) and an edge represents a transmission line joining two electrical nodes. The problem is to locate the minimum number of PMUs.
The next question is then to ask: How does a PMU measure the state variables? A PMU measures the state variable (voltage and phase angle) for the vertex at which it is placed and its incident edges and their end vertices. (These vertices and edges are said to be observed.)
The other observation rules are as follows:
(1) Any vertex that is incident to an observed edge is observed.
(2) Any edge joining two observed vertices is observed.
(3) If a vertex is incident to a total of k > 1 edges and if k −1 of these edges are observed,
then all k of these edges are observed.
It was noticed in [29, 31] that the power domination problem can be studied considering only vertices.
Let G be a connected graph and S a subset of its vertices. Then we denote the set observed by S with X(S) and define it recursively as follows:
(1) (domination)
X(S) ← S ∪ N(S)
(2) (propagation)
As long as there exists v ∈ X(S) such that
N(v) ∩ (V (G) − X(S)) = {w}
set X(S) ← X(S) ∪ {w}
A set S is called a power dominating set of G if X(S) = V (G). The power domination number
γp(G) is the minimum cardinality of a power dominating set of G. A power dominating set of
G with the minimum cardinality is called a γp(G)-set. The first step in the power domination process can be thought of as domination process.
A dominating set of a graph G(V,E) is a set S of vertices of G such that S ∪ N(S) = V (G) and the minimum cardinality of such a set is called domination number denoted γ(G). The second step of the power domination process is called the propagation process and is similar to 26 2. BACKGROUND THEORY the zero forcing process described in Section 2.2 on page 11. Thus, power domination is closely related to domination problem and zero forcing problem.
5 6
4 3
1 2
0 7
Figure 2.5. A graph G with power dominating set S = {5, 6}
In Figure 2.5, the set S = {5, 6} is clearly a power dominating set as exhibited in Figure 2.6.
At first time step (Figure 2.6a), the neighbours of 5 and 6 are observed. In Figure 2.6b, vertex
4 observes vertex 1 and in Figure 2.6c, vertex 3 observes vertex 2. In Figure 2.5, blue vertices denote observed vertices and red vertices are vertices that are not observed.
The problem of deciding if a graph G has a power dominating set of cardinality k has been shown to be NP-complete even for bipartite graphs, chordal graphs [55] or even split graphs [72].
The power domination problem has efficient polynomial time algorithms for the classes of trees [55], graphs with bounded tree-width [51], block graphs [89], block cactus graphs [61], interval graphs [72], and circular-arc graphs [71].
2.3.1. Bounds on the power domination number.
Theorem 2.3.1 ([55]). For any graph G,
1 ≤ γp(G) ≤ γ(G).
Also γp(G) = 1 for G ∈ {Kn,Cn,Pn,K2,k}.
It is evident that any dominating set is trivially a power dominating set and thus we have the trivial upper bound yet the authors proved that there is no forbidden subgraph characterization of the graphs reaching this bound.
The next result allows us to restrict vertices that belong to any power dominating set. 2.3. POWER DOMINATION IN GRAPHS 27
5 6 5 6
4 3 4 3
1 2 1 2
0 7 0 7
(a) First time step (b) Second time step 5 6
4 3
1 2
0 7
(c) Third time step
Figure 2.6. Exhibition of the power dominating steps
Theorem 2.3.2 ([55]). Let G be a graph with ∆(G) ≥ 3. Then there is a minimum power dominating set S in which each vertex in S has degree at least 3.
In [55], the authors studied extensively the power dominating set in trees. For a tree T , define the spider number of T , denoted sp(T ), to be the minimum number of subsets V (T ) can be partioned so that each subset induces a graph homeomorphic to K1,k for some k ∈ N.
Theorem 2.3.3 ([55]). For any tree T , sp(T ) = γp(G).
The following sharp upper bound for the power domination number of a graph was initially shown for trees in [55] and was later generalized to all graphs in [91].
Theorem 2.3.4 ([91]). Let T be the family of graphs obtained from connected graphs H by adding two new vertices v0 and v00 to each vertex v of H and new edge vv0 and vv00, while v0v00 28 2. BACKGROUND THEORY
n may be added or not. If G is a connected graph of order n ≥ 3, then γp(G) ≤ 3 with equality if and only if G ∈ T ∪ {K3,3}.
Zhao and Kang addressed the question whether one can bound the power domination number of a graph with some condition on the diameter, that is the maximum distance between two vertices of the graph. In [90], they gave some general bounds for the power domination number of planar graphs with diameter 2 or 3. They showed that in outerplanar graphs, if the diameter is at most 2, then the graph admits a power dominating set of size one, while if the diameter is 4 or more, the power domination number can be arbitrarily large.
It was shown in [91] by Zhao, Kang and Chang that the bound could be improved for regular graphs.
Theorem 2.3.5 ([91]). If G is a connected claw-free cubic graph on n vertices, then
n γ (G) ≤ . p 4
Zhao et al. in the same paper also characterized the graphs for which the bound is tight which can be described as follows: take an even cycle and replace every second edge by a K4 minus an edge, using the degree 2 vertices of K4 − e as end vertices of the edge. In [27], the result obtained in Theorem 2.3.5 was improved by removing the claw free condition.
Theorem 2.3.6 ([27]). If G is a connected cubic graph on n vertices, then either G = K3,3
n or γp(G) ≤ 4 .
The above result in Theorem 2.3.6 motivated the authors to pose the following conjecture which is still open to be resolved.
Conjecture 1 ([27]). For r ≥ 3, if G 6= Kr,r is a connected r-regular graph on n vertices, then n γ (G) ≤ . p r + 1
Another area where the power domination of graphs was extensively studied was on graph products. We shall highlight some of the known results in literature. 2.3. POWER DOMINATION IN GRAPHS 29
Theorem 2.3.7 ([31]). The power domination number of s × t grid Ps × Pt for s ≥ t ≥ 1 is s+1 d 4 e if t ≡ 4 (mod 8), γp(Ps × Pt) = s d 4 e otherwise.
This result is quite surprising in some sense because domination problem on grid graphs is still an open problem. Motivated by this result, Ferrero et al. [39] extended the results of grid networks to hexagonal honeycomb grid graph and triangular grid graph.
The power domination numbers for cylinders Ps × Ct for integers s ≥ 2, t ≥ 3, tori Cs × Ct for integers s, t ≥ 3 were studied in [10] and tight bounds were found.
Theorem 2.3.8 ([10]). The power domination number of s × t cylinder Ps × Ct for s ≥ 2, t ≥ 3 is t+1 s+1 min d 4 e, d 2 e if s ≡ 4 (mod 8), γp(Ps × Ct) ≤ t s+1 min d 4 e, d 2 e otherwise.
Theorem 2.3.9 ([10]). The power domination number of s × t torus Cs × Ct for s ≤ t is s d 2 e if s ≡ 0 (mod 4), γp(Cs × Ct) ≤ s+1 d 2 e otherwise.
Recently, Benson et al. [11] showed that the upper bound obtained for torus in [10] is tight.
Theorem 2.3.10 ([11]). For t ≥ s ≥ 3,
l s m γ (C × C ) = . p s t 2
In [29], this study on the products of paths is continued with other graph products. For the direct product (which has two connected components), the bound obtained is as follows:
Theorem 2.3.11 ([29]). The power domination number of the direct product Ps · Pt for t ≥ s ≥ 1 is s 2d 4 e if s is even, γp(Ps · Pt) ≤ t 2d 4 e if s is odd and t even, 30 2. BACKGROUND THEORY
If both s and t are odd,
l t m lt − 2m lt + sm lt + s − 2m γ (P · P ) ≤ max + , + . p s t 4 4 6 6
Theorem 2.3.12 ([88]). The power domination number of the direct product Cs · Kt for t ≥ 3, s ≥ 4 is 2k if s = 4k, γp(Cs · Kt) ≤ 2k + 1 if s = 4k + 1, 2k + 2 if s = 4k + 2 or n = 4k + 3.
Theorem 2.3.13 ([88]). The power domination number of the direct product Ps ·Ct, s even, is s 2d 3 e if t is even, γp(Ps · Ct) = s d 3 e if t is odd. The situation for the strong product of two paths is a little simpler, though not completely solved either.
Theorem 2.3.14 ([29]). The power domination number of the direct product Ps ⊗ Pt for t ≥ s ≥ 2 is l t m ls + 1m γ (P ⊗ P ) = min , p s t 4 2 unless 3s − t − 6 ≡ 4 (mod 8) in which case
l t m lt + s − 2m l t m lt + s − 2m max , ≤ γ (P ⊗ P ) ≤ max , + 1 . 3 4 p s t 3 4
Dean et al. [25] gave a lower bound and upper bound for the power domination number in hypercubes.
Theorem 2.3.15 ([25]). For the r-dimensional hypercube Qr,
2r−1 ≤ γ (Q ) ≤ 2r−dlog re−1. r p r
The exact values of the power domination for hypercube network Qr has been known only for values of r ≤ 7 as shown in [78]. This is an indication of how difficult it is finding the exact values of power domination number of graphs. 2.3. POWER DOMINATION IN GRAPHS 31
Power domination problem has been studied effectively for Sierpi´nskigraphs [28], gener- alized Petersen graphs [10], undirected de Bruijn and Kautz graphs [69] and Mycielskian of graphs [88].
The next couple of results examine what happens to the power domination number under certain graph operations. The results below are obtained independently by Benson et al. [11] and Dorbec et al. [30].
Theorem 2.3.16 ([11, 30]). For every graph G and for every vertex v,
γp(G) − 1 ≤ γp(G − v).
There is no upper bound for γp(G−v) in terms of γp(G). In fact, the authors in [11] proved the following proposition to justify the claim.
Proposition 1 ([11]). For every integer r ≥ −1, there is a graph Gr with a vertex v such that
γp(Gr − v) = γp(Gr) + r.
Theorem 2.3.17 ([11, 30]). If G − e is the graph obtained from G by deleting an edge e = uv, then
γp(G) − 1 ≤ γp(G − e) ≤ γp(G) + 1.
Theorem 2.3.18 ([11, 30]). If G/e is the graph obtained from G by contracting an edge e = uv, then
γp(G) − 1 ≤ γp(G/e) ≤ γp(G) + 1.
Theorem 2.3.19 ([11, 30]). If Ge is the graph obtained from G by subdividing an edge e = uv, then
γp(G) ≤ γp(Ge) ≤ γp(G) + 1.
2.3.2. Power domination in directed graphs. The directed version of the power dom- ination problem was initiated in [1]. Following the definition in [37], we define a set S ⊆ V is a power dominating set of G~ if and only if N out[S] is a zero forcing set of G~ .A minimum 32 2. BACKGROUND THEORY power dominating set is a power dominating set of minimum cardinality. The power dominating number of G is the cardinality of a minimum power dominating set and is denoted by γp(G~ ).
1− In [1], the authors show hardness of approximation threshold of 2log n for directed acyclic graphs. They also show that the power domination problem for directed graphs can be solved optimally in linear time if the underlying undirected graph has bounded tree-width.
Theorem 2.3.20 ([1]). Given a directed graph G~ and a tree decomposition of width k of its underlying undirected graph, power dominating set problem for directed graphs can be optimally solved in O((ck)2.n) time for a global constant c.
Remark 2.3.21. We note that γp(G) and γp(G~ ) is used to denote the power domination number of an undirected and a directed graph respectively.
2.3.3. Variants of power domination problem. In this section, we give an overview of the variants of power domination problem that have been studied and are beyond the scope of this thesis.
k-power domination. This notion of k-power domination was introduced by Chang et al. [18] which generalizes the concept of power domination. Let G be a connected graph and
i S a subset of its vertices. Then we denote the set monitored by S at step i as PG,k(S) and define it recursively as follows:
0 (1) PG,k(S) = N[S] i+1 S i i (2) PG,k(S) = {N[v]: v ∈ PG,k(S)} such that |N[v] \ PG,k(S)| ≤ k
i i+1 i0 i0+1 Note that PG,k(S) ⊆ PG,k(S) ⊆ V (G) for any i. If PG,k(S) = PG,k (S) for some i0, then
j i0 ∞ i0 PG,k(S) = PG,k(S) for any j ≥ i0. Define PG,k(S) = PG,k(S). ∞ A set S such that PG,k(S) = V (G) is a k-power dominating set of G. The least cardinality of such a set is called the k-power domination number of G, written γp,k(G). A γp,k(G)-set is a minimum k-power dominating set of G.
It is to be noted that when k = 0, the k-power domination problem reduces to the domina- tion problem and when k = 1, the problem reduces to the power domination problem. This is a beautiful generalisation which brings both domination problem and power domination problem under a single roof. Results on k-power domination problem may be found in [27, 28, 30]. 2.4. ZERO FORCING AND POWER DOMINATION 33
Propagation radius. Propagation radius was introduced in [28] as a measure of the efficiency of power dominating sets and is defined for k-power dominating set as follows:
i radp,k(G) = 1 + min{i : PG(S) = V (G),S is a k-power dominating set of G}.
2.4. Zero forcing and power domination
Both zero forcing problem and power domination problem are very closely related problems.
It is not until recently that a connection between these problems was established in undirected graphs.
Theorem 2.4.1 ([25]). The set S is a power dominating set if and only if N[S] is a zero forcing set.
Theorem 2.4.1 implies that every zero forcing set is a power dominating set. That is,
γp(G) ≤ Z(G). It is also known that every dominating set is also a power dominating set (Theorem 2.3.1). But there exists no relationship between zero forcing number and domination number.
Observation 2. Let G be an undirected graph. Then, the zero forcing number Z(G) is neither bounded above nor bounded below by domination number γ(G).
If G = P6, path on 6 vertices, γ(G) = 2 and Z(G) = 1 whereas if G = Kn, complete graph on n vertices, γ(G) = 1 and Z(G) = n − 1.
The next theorem due to Benson et al. [11] was deduced from Theorem 2.4.1.
Theorem 2.4.2 ([11]). Let G be a connected non trivial graph. Then
l Z(G) m ≤ γ (G) ∆(G) p and this bound is is tight.
The graph satisfying Theorem 2.4.2 is the complete graph Kn as Z(Kn) = ∆(Kn) = n − 1 and γp(G) = 1. The next result is immediate from the fact that M(G) ≤ Z(G)[3]. Although weaker than Theorem 2.4.2, it can sometimes be applied using a well known matrix such as the adjacency or Laplacian matrix of the graph, even if M(G) and Z(G) are not known. 34 2. BACKGROUND THEORY
Theorem 2.4.3 ([11]). For a graph G that has an edge and any matrix A ∈ S(R,G),
lcorank Am ≤ γ (G). ∆(G) p
As an application to Theorem 2.4.2, the authors solved the power domination of the torus described in Theorem 2.3.10. They also exhibited that many of the proofs of the values of the power domination number for families can be simplified by application of the relationship between power domination and zero forcing.
In this thesis, we also establish a relation between power domination and zero forcing and exploit it to solve these problems for de Bruijn and Kautz digraphs in Chapter3. CHAPTER 3
Zero forcing and power domination in iterated line
digraphs
In this chapter we make the following contributions.
• In Section 3.1, we interpret zero forcing and power domination in directed graphs as
set covering problems.
• In Sections 3.2 and 3.3, we solve the zero forcing and power domination problem in
de Bruijn and Kautz digraphs.
• In Sections 3.4 and 3.5, we extend the results of de Bruijn and Kautz digraphs to
obtain bounds for zero forcing number and power domination number in iterated line
digraphs respectively.
3.1. An alternate interpretation
We give an interpretation of the zero forcing problem and the power domination problem as a set cover problem. We call a vertex set W strongly critical if there is no vertex in G~ which has exactly one out neighbour in W . We call a vertex set W weakly critical if there is no vertex outside W which has exactly one out-neighbour in W . If W is strongly (weakly) critical, but no proper subset of W is strongly (weakly) critical, then we call W minimal strongly (weakly) critical.
♦ Remark 3.1.1. Every strongly critical set is a weakly critical set but the converse need not be true. As a consequence, in any digraph, the maximum number of pairwise disjoint strongly critical sets is less than or equal to the maximum number of pairwise disjoint weakly critical sets.
The contents of this chapter reproduce original results from [37] and [45].
35 36 3. ZERO FORCING AND POWER DOMINATION IN ITERATED LINE DIGRAPHS
6
3 4 1 2
5
Figure 3.1. A directed graph G~ .
In Figure 3.1, the set of vertices {1, 2} is both a strongly and weakly critical set. But the set {3, 4} is a weakly critical set but not a strongly critical set.
It is interesting to note that in Figure 3.1, vertex 5 and vertex 6 have to belong in any zero forcing and power dominating set because their in-degree is zero and cannot be forced by any other vertex but itself.
~ ~ in ♦ Remark 3.1.2. Let G be a digraph and let v be a vertex in G such that degG~ (v) = 0, then v must belong to any zero forcing and power dominating set.
Generalising this observation, we now characterise zero forcing sets of digraphs in terms of strongly and weakly critical sets. The singleton {v} is strongly critical if and only if the in-degree of v is 0. Recall that for a loop digraph, a red or a blue vertex can force a colour change.
♦ Lemma 3.1.3. A vertex set S is a zero forcing set in a loop digraph G~ if and only if S ∩ W 6= ∅ for every strongly critical set W ⊆ V , and therefore
Z(G~ ) = min {|S| : S ∩ W 6= ∅ for every strongly critical set W ⊆ V } .
Proof. Let S be a zero forcing set with S ∩ W = ∅ for some strongly critical set W ⊆ V . This implies that even if all vertices of V \ W are in S, no vertices of W can be forced by any vertex in V \ W , a contradiction. By pigeon hole principle, we conclude that at least one vertex from every strongly critical set must belong to any minimum zero forcing set. To prove the other direction, we need to show that if a set S intersects all minimal strongly critical sets, then
S is a zero forcing set. Assume the counter example. Since S is not a zero forcing set, not at 3.1. AN ALTERNATE INTERPRETATION 37 all vertices of G~ are turned blue after the last forcing step possible. This implies that all blue vertices in G~ after the last forcing step are adjacent to at least two red vertices or no red vertex.
Thus, the collection of all red vertices will form a strongly critical set, a contradiction.
For simple digraphs, we recall that only a blue vertex can force a colour change and following the same argument of Lemma 3.1.3, we have
♦ Lemma 3.1.4. A vertex set S is a zero forcing set in a simple digraph G~ if and only if S ∩ W 6= ∅ for every weakly critical set W ⊆ V , and therefore
Z(G~ ) = min {|S| : S ∩ W 6= ∅ for every weakly critical set W ⊆ V } .
♦ Corollary 3.1.5. In any digraph G~ = (V,A), its zero forcing number Z(G~ ) is at least the maximum number of disjoint strongly critical sets in G~ . Moreover, if G~ does not have any loops, then Z(G~ ) is at least the maximum number of disjoint weakly critical sets in G~ .
Proof. Let S be a minimum zero forcing set of G~ and let {W1,...,Wr} be a set of pairwise disjoint strongly critical sets. By Lemma 3.1.3, |S ∩ Wi| ≥ 1 for every i = 1, . . . , r, and since the sets W1,...,Wr are pairwise disjoint, it must be |S| ≥ r. If G~ does not have any loops we apply the same argument with a collection of pairwise disjoint weakly critical sets.
A set S ⊆ V is a power dominating set of G~ if and only if N out[S] is a zero forcing set of
G~ . Thus, one can characterise power dominating sets as follows.
♦ Lemma 3.1.6. A vertex set S is a power dominating set in a simple (loop) digraph G~ if and only if N out[S] ∩ W 6= ∅ for every weakly (strongly) critical set W ⊆ V , and therefore G~
n o γ (G~ ) = min |S| :(S ∪ N out(S)) ∩ W 6= ∅ for every weakly (strongly) critical set W ⊆ V . p G~
In the next section, we apply this new interpretation in solving the zero forcing and power domination of de Bruijn and Kautz digraphs. Due to their attractive connectivity features these digraphs have been widely studied as a topology for interconnection networks [62], and some gen- eralizations of these digraphs were proposed [64]. Recently, Dong et al. (2015) [26] investigated the domination number of generalized de Bruijn and Kautz digraphs. Kuo et al. (2015) [69] gave an upper bound for power domination in undirected de Bruijn and Kautz graphs. This 38 3. ZERO FORCING AND POWER DOMINATION IN ITERATED LINE DIGRAPHS motivated us to study the directed versions, i.e., the zero forcing number and power domination number of de Bruijn and Kautz digraphs.
3.2. Zero forcing and power domination in de Bruijn digraphs
For an integer d ≥ 2, let Zd = {0, 1, . . . , d − 1} denote the cyclic group of order d. The de Bruijn digraph, denoted B(d, `), with parameters d ≥ 2 and ` ≥ 2 is defined to be the graph
G~ = (V,A) with vertex set V and arcs set A where
` V = Zd = {(a1, . . . , a`): ai ∈ Zd for i = 1, . . . , `} ,
A = {((a1, a2, . . . , a`), (a2, . . . , a`, b)) : (a1, a2, . . . , a`) ∈ V, b ∈ Zd} .
Our main result in this section is the following theorem.
♦ Theorem 3.2.1. Let G~ be a de Bruijn digraph with parameters d, ` ≥ 2. Then the zero forcing number and power domination number of G~ are (d−1)d`−1 and (d−1)d`−2, respectively.
Let us define the sets
X(a1, . . . , a`−1) = {(a1, . . . , a`−1, α): α ∈ Zd}
`−1 out which partition the vertex set V into d sets of size d. Furthermore, N (v) = X(a1, . . . , a`−1) for every vertex v of the form (α, a1, a2, . . . , a`−1).
♦ Lemma 3.2.2. Let G~ be a de Bruijn digraph with parameters d and `. Then Z(G~ ) ≥ (d − 1)d`−1.
Proof. Every 2-element subset of each of the sets X(a1, . . . , a`−1) is strongly critical, and therefore, any zero forcing set S needs to intersect X(a1, . . . , a`−1) in at least d − 1 elements, and the result follows.
♦ Lemma 3.2.3. Let G~ be a de Bruijn digraph with parameters d and `. Then Z(G~ ) ≤ (d − 1)d`−1.
Proof. Consider the vertex set S = {(a1, . . . , a`−1, a`) ∈ V : a1 6= a`}. To show that S is a zero forcing set, it is sufficient to verify that each vertex v = (a1, . . . , a`−1, a`) is either in S 3.2. ZERO FORCING AND POWER DOMINATION IN DE BRUIJN DIGRAPHS 39
or is the unique out-neighbour in V \ S for some vertex w. If a1 6= a`, then v ∈ S. If a1 = a`, then for any vertex of the form w = (β, a1, . . . , a`−1), v is the only neighbour of w in V \ S.
Lemmas 3.2.2 and 3.2.3 imply the first statement of Theorem 3.2.1. In order to prove the second part of this theorem we recall that S ⊆ V is a power dominating set if and only if S ∪ N out(S) is a zero forcing set. In particular, it is necessary that |(S ∪ N out(S)) ∩
`−1 X(a1, . . . , a`−1)| > d − 1 for every (a1, . . . , a`−1) ∈ Zd .
♦ Lemma 3.2.4. Let G~ be a de Bruijn graph with parameters d and `. Then every power dominating set has size at least (d − 1)d`−2.
Proof. Let S be a power-dominating set, suppose |S| < (d − 1)d`−2 and set Z = S ∪ N out(S). We have
(Z \ S) ∩ X(a1, . . . , a`−1) 6= ∅ =⇒ X(a1, . . . , a`−1) ⊆ Z.
For k = 0, 1, . . . , d, we set αk = |{(a1, . . . , a`−1): |S ∩ X(a1, . . . , a`−1)| = k}|, and get
|S| = α1 + 2α2 + ··· + (d − 1)αd−1 + dαd.
Now let I0 = {(a1, . . . , a`−1): X(a1, . . . , a`−1) ⊆ Z}. Then
|I0| 6 |S| + αd = α1 + 2α2 + ··· + (d − 1)αd−1 + (d + 1)αd.
For (a1, . . . , a`−1) ∈/ I0 we must have |Z ∩ X(a1, . . . , a`−1)| = d − 1, and this implies that
`−1 |S ∩ X(a1, . . . , a`−1)| = d − 1. We conclude |I0| + αd−1 > d . Therefore
`−1 α1 + 2α2 + ··· + (d − 2)αd−2 + dαd−1 + (d + 1)αd > d , and together with |S| < (d − 1)d`−2 this yields
`−1 `−2 `−2 αd−1 + αd > d − (d − 1)d = d .
`−2 But then |S| > (d − 1)(αd−1 + αd) > (d − 1)d , which is the required contradiction. 40 3. ZERO FORCING AND POWER DOMINATION IN ITERATED LINE DIGRAPHS
We define a set S ⊆ V by {(0, 1), (0, 2),..., (0, d − 1)} if ` = 2, (3.2.1) S = {(a1, a2, a3) ∈ V : a2 = a1, a3 6= a1} if ` = 3, {(a1, . . . , a`) ∈ V : a`−1 = a1 + a`−2, a` 6= a1 + a2 + a`−2} if ` > 4.
Note that |S| = (d − 1)d`−2. The construction of the set S defined in (3.2.1) can be visualized
2 `−2 by arranging the vertices of G~ in a d ×d -array where the rows are indexed by pairs (a`−1, a`) and the columns are indexed by (` − 2)-tuples (a1, . . . , a`−2). Then column (a1, . . . , a`−2) is the the union of the d sets X(a1, . . . , a`−2, a`−1) over a`−1 ∈ Zd, and the set S contains d − 1 elements from each column. More precisely, the intersection of S with column (a1, . . . , a`−2) is
X(a1, . . . , a`−2, a1 + a`−2) \{(a1, . . . , a`−2, a1 + a`−2, a1 + a2 + a`−2)}.
In Figure 3.2 this is illustrated for two columns with d = 5 and ` = 7.
(1, 3, 4, 4, 2) (3, 1, 0, 2, 4)
X(1, 3, 4, 4, 2, 0) X(3, 1, 0, 2, 4, 0)
a6 = 0 = N out(3, 1, 3, 4, 2, 2, 0) = N out(2, 3, 1, 0, 2, 4, 0)
X(1, 3, 4, 4, 2, 1) X(3, 1, 0, 2, 4, 1)
a6 = 1 = N out(3, 1, 3, 4, 2, 2, 1) = N out(2, 3, 1, 0, 2, 4, 1)
X(1, 3, 4, 4, 2, 2)
a6 = 2 = N out(3, 1, 3, 4, 2, 2, 2)
X(3, 1, 0, 2, 4, 3)
a6 = 3 = N out(2, 3, 1, 0, 2, 4, 3)
X(1, 3, 4, 4, 2, 4) X(3, 1, 0, 2, 4, 4)
a6 = 4 = N out(3, 1, 3, 4, 2, 2, 4) = N out(2, 3, 1, 0, 2, 4, 4)
Figure 3.2. Illustration of the construction of the power dominating set S for d = 5 and ` = 7. For the two columns (a1, . . . , a5) = (1, 3, 4, 4, 2) and (a1, . . . , a5) = (3, 1, 0, 2, 4) we show the elements of S (black squares), and we indicate for the sets X(a1, . . . , a6) (enclosed by rectangles) the elements of S having them as their out-neighbourhood. 3.3. ZERO FORCING AND POWER DOMINATION NUMBER IN KAUTZ DIGRAPHS 41
♦ Lemma 3.2.5. The set S defined in (3.2.1) is a power dominating set for G~ = B(d, `).
out Proof. For Z = S ∪ N (S) it is sufficient to show that |Z ∩ X(a1, . . . , a`−1)| > d − 1 for every (a1, . . . , a`−1). We provide the full argument for ` > 4 (the cases ` = 2 and ` = 3 are easy to check).
Case 1.: If a`−1 = a1 + a`−2, then by (3.2.1),
S ∩ X(a1, . . . , a`−1) = {(a1, . . . , a`): a` ∈ Zd \{a1 + a2 + a`−2}},
hence |Z ∩ X(a1, . . . , a`−1)| ≥ |S ∩ X(a1, . . . , a`−1)| = d − 1.
Case 2.: If a`−1 6= a1 + a`−2, then X(a1, . . . , a`−1) ⊆ Z because
out X(a1, . . . , a`−1) = N ((a`−2 − a`−3, a1, a2, . . . , a`−1))
and (a`−2 − a`−3, a1, a2, . . . , a`−1) ∈ S.
The second part of Theorem 3.2.1 follows from Lemmas 3.2.4 and 3.2.5.
3.3. Zero forcing and power domination number in Kautz digraphs
The Kautz digraph, denoted K(d, `), with parameters d ≥ 2 and ` ≥ 2 is defined to be the graph G~ = (V,A) with vertex set V and arcs set A where
V = {(a1, . . . , a`): ai ∈ Zd+1, ai 6= ai+1}
A = {((a1, a2, . . . , a`), (a2, . . . , a`, b)) : (a1, a2, . . . , a`) ∈ V, b ∈ Zd+1 \{a`}} .
Our main result in this section is the following theorem.
♦ Theorem 3.3.1. Let G~ be a Kautz digraph with parameters d ≥ 2 and ` ≥ 3. Then, the zero forcing number and power domination number of G~ are (d − 1)(d + 1)d`−2 and (d −
1)(d + 1)d`−3, respectively.
Let us define the sets
X(a1, . . . , a`−1) = {(a1, . . . , a`−1, a`): a` ∈ Zd+1 \{a`−1}} 42 3. ZERO FORCING AND POWER DOMINATION IN ITERATED LINE DIGRAPHS
`−1 for (a1, . . . , a`−1) ∈ Zd+1 with ai 6= ai+1 for all i. These sets partition the vertex set V into `−2 out (d + 1)d sets of size d. Furthermore, N (v) = X(a1, . . . , a`−1) for every vertex v of the form (a0, a1, a2, . . . , a`−1).
~ ~ ♦ Lemma 3.3.2. Let G be a Kautz digraph with parameters d, ` > 2. Then Z(G) ≥ (d − 1)(d + 1)d`−2.
Proof. Every 2-element subset of each of the sets X(a1, . . . , a`−1) is strongly critical, and therefore, any zero forcing set S needs to intersect X(a1, . . . , a`−1) in at least d − 1 elements, and the result follows.
~ ~ ♦ Lemma 3.3.3. Let G be a Kautz digraph with parameters d, ` > 2. Then Z(G) ≤ (d − 1)(d + 1)d`−2.
Proof. Consider the vertex set {(a1, a2) ∈ V : a2 6= a1 + 1} if ` = 2, S = {(a1, . . . , a`) ∈ V : a` 6= a`−2} if ` > 3.
We have |S| = (d − 1)(d + 1)d`−2, and to show that S is a zero forcing set, it is sufficient to verify that each vertex v = (a1, . . . , a`−1, a`) is either in S or is the unique out-neighbour in V \ S for some vertex w.
Case 1 : ` = 2. If a2 6= a1 + 1 then v ∈ s. If a2 = a1 + 1 then for any vertex of the form
w = (β, a1), v is the only neighbour of w in V \ S.
Case 2 : ` ≥ 3. If a` 6= a`−2, then v ∈ S. If a` = a`−2, then for any vertex of the form
w = (β, a1, . . . , a`−1), v is the only neighbour of w in V \ S.
Lemmas 3.3.2 and 3.3.3 imply the first statement of Theorem 3.3.1.
♦ Lemma 3.3.4. Let G~ be a Kautz digraph with parameters d ≥ 2 and ` ≥ 3. Then, every power dominating set has size at least (d − 1)(d + 1)d`−3.
Proof. Let S be a power-dominating set, suppose |S| < (d − 1)(d + 1)d`−3 and set Z = S ∪ N out(S). We have
(Z \ S) ∩ X(a1, . . . , a`−1) 6= ∅ =⇒ X(a1, . . . , a`−1) ⊆ Z. 3.3. ZERO FORCING AND POWER DOMINATION NUMBER IN KAUTZ DIGRAPHS 43
For k = 0, 1, . . . , d, we set αk = |{(a1, . . . , a`−1): |S ∩ X(a1, . . . , a`−1)| = k}|, and get
|S| = α1 + 2α2 + ··· + (d − 1)αd−1 + dαd.
Now let I0 = {(a1, . . . , a`−1): X(a1, . . . , a`−1) ⊆ Z}. Clearly,
|I0| 6 |S| + αd = α1 + 2α2 + ··· + (d − 1)αd−1 + (d + 1)αd.
For (a1, . . . , a`−1) ∈/ I0 we must have |Z ∩ X(a1, . . . , a`−1)| = d − 1 because Z intersects every weakly critical set. This implies that |S ∩ X(a1, . . . , a`−1)| = d − 1, and we conclude
`−2 |I0| + αd−1 > (d + 1)d . Therefore
`−2 α1 + 2α2 + ··· + (d − 2)αd−2 + dαd−1 + (d + 1)αd > (d + 1)d , and together with |S| < (d − 1)(d + 1)d`−3 this yields
`−2 `−3 `−3 αd−1 + αd > (d + 1)d − (d − 1)(d + 1)d = (d + 1)d .
`−3 But then |S| > (d−1)(αd−1 +αd) > (d−1)(d+ 1)d , which is the required contradiction.
We define a set S ⊆ V by
(3.3.1) {(0, 1), (0, 2),..., (0, d)} if ` = 2, {(a1, a2, a3) ∈ V : a2 = a1 + 1, a3 6= a1 + 2} if ` = 3, S = {(a1, a2, a3, a4) ∈ V : a3 = a1, a4 6= a2} if ` = 4, {(a1, . . . , a`) ∈ V : ((a`−2, a`−1) = (a1, a2) ∧ an 6= a3) ∨ (a`−1 = a1 ∧ a` 6= a2)} if ` > 5.
d if ` = 2, ♦ Lemma 3.3.5. |S| = `−3 (d − 1)(d + 1)d if ` > 3.
Proof. For ` ≤ 4 this is easy to check. For ` ≥ 5 we proceed by the following argument.
We consider the partition S = S1 ∪ S2 where
S1 = {(a1, . . . , a`) ∈ S : a`−3 = a1},S2 = {(a1, . . . , a`) ∈ S : a`−3 6= a1}. 44 3. ZERO FORCING AND POWER DOMINATION IN ITERATED LINE DIGRAPHS
Let sk be the number of words a1 . . . ak over the alphabet Zd+1 which satisfy ak = a1 and k−2 ai 6= ai+1 for all i ∈ {1, . . . , k − 1}. Then s2 = 0 and sk = (d + 1)d − sk−1 for k ≥ 3. It
k−1 k `−3 follows by induction on k that sk = d − (−1) d. Every vector (a1, . . . , a`−3) ∈ Zd+1 with ai 6= ai+1 and a`−3 = a1 can be extended to an element of S1 by choosing a`−2 ∈ Zd+1 \{a1}, a`−1 = a1 and a` ∈ Zd+1 \{a1, a2}, hence