Forcing parameters for graphs David Amos November 5, 2014 1 Basic definitions Here we collect a few necessary graph theory definitions. To begin, a graph is a pair G = (V, E) of sets. The set V is the vertex set of G. We always assume that V is finite, and usually take V to be the set {1, 2, . . . , n}. The set E is the edge set of G, and is a subset of {{u, v} : u, v ∈ V }. In particular, we only consider undirected graphs without loops or multiple edges between vertices. If {u, v} ∈ E, we say that u and v are neighbors, or that u is adjacent to v. The degree of a vertex v ∈ V , denoted deg(v), is the number of neighbors of v. A graph is said the be d-regular if every vertex has degree d. Let G = (V, E) be a graph. If X ⊆ V and Y ⊆ E for which every edge in Y is between vertices of X, then the graph H = (X, Y ) is called a subgraph of G. If Y contains every edge between vertices in X, we say that H is an induced subgraph of G. We also that that H is the subgraph induced by X. A path in a graph G from a vertex u to a vertex w is a sequence of vertices u = v1 , v2 , . . . , vk = w such that vi is adjacent to vi+1 for all 1 ≤ i ≤ k − 1. A graph G is connected if there exists a path from each vertex of G to any other vertex of G. 2 Zero forcing Zero forcing was introduced independently by D. Burgarth and V. Giovanneti in 2007 (see [2]) and the AIM Minimum Rank - Special Graphs Work Group (the “AIM Group” for short) in 2008 (see [3]). Burgarth and Giovanneti introduced zero forcing in the study of control of quantum systems; the AIM 1 group introduced zero forcing in order to bound the minimum rank of a graph (the minimum rank problem is discussed in section 3). Definition 1 (AIM Group, [3]). Let G be a simple, finite and undirected graph. Color each vertex of G either black or white. (a) Color change rule: If u is a black vertex of G and u has exactly one white neighbor v, then u “forces” v to become black. (b) Derived coloring: The derived coloring of G is the coloring obtained by applying the color change rule as many times as possible. (c) Zero forcing set: A nonempty set Z of vertices is called a zero forcing set if, when all of the vertices in Z are colored black and all other vertices of G are colored white, the derived coloring of G has no white vertices. (d) The zero forcing number of G, denoted Z(G) is the minimum |Z| over all zero forcing sets Z. Example 1. Consider the following graph G and coloring of G: The set of black vertices is easily seen to be a zero forcing set: 4 forces 3, 3 (or 5) forces 6, and 6 (or 2) forces 1. Moreover, no set of one or two vertices is a zero forcing set, so Z(G) = 3. Example 2. The zero forcing number of a path is 1 (a single degree one vertex is a zero forcing set). Example 3. The zero forcing number of a cycle C is 2 (clearly no single vertex forces, so Z(C) ≥ 2; any set containing a pair of adjacent vertices is a zero forcing set, so Z(C) = 2). 2 Remark. For any graph G with n vertices and at least one edge, Z(G) ≤ n−1 since the set containing every vertex of G except for one of the vertices incident to an edge is a zero forcing set. Proposition 1. For any graph G with minimum degree δ, Z(G) ≥ δ. Proof. Let Z be a smallest zero forcing set for G. If Z contains every vertex of G, we are done, since n ≥ δ + 1. If Z does not contain every vertex of G, then there exists a u in Z that forces one of its neighbors v. This means that all but one of u’s neighbors is in Z so that |Z| ≥ deg(u). Since Z is a smallest zero forcing set, Z(G) ≥ δ. Example 4. Let Kn be the complete graph on n-vertices. Then Z(Kn ) = n − 1. To see this, note that by Proposition 1, Z(Kn ) ≥ n − 1, since the minimum degree of Kn is n − 1. On the other hand, any set of n − 1 vertices is a zero forcing set for Kn , and thus Z(Kn ) = n − 1. Example 5. The zero forcing number of the Peterson graph (shown below) is 5. The five vertices on the outer cycle constitute a zero forcing set, and thus the zero forcing number is at most 5. It follows from Proposition 1 that the zero forcing number is at least 3. That the zero forcing is not 3 or 4 is left as an exercise. 3 Zero forcing and the minimum rank problem Definition 2. Let A = (aij ) be a symmetric matrix over a field F . (a) The graph of A, denoted G(A), is the graph with vertex set {1, 2, . . . , n} and edge set {{i, j} : aij 6= 0, 1 ≤ i < j ≤ n}. Note that we do not 3 consider the diagonal of A when determining G(A). Moreover, for any graph G and field F , there exists a symmetric matrix A whose graph is G (the adjacency matrix, for example). (b) The minimum rank of a graph G over a field F , denoted mrF (G), is the minimum rank over all matrices A for which G(A) ' G. (c) The maximum nullity of a graph G over a field F , denoted M F (G), is the maximum nullity over all matrices A for which G(A) ' G. Remark. For any field F and graph G with n vertices, the rank-nullity relation implies that mrF (G) + M F (G) = n. If F = R, we will omit the F in our notation for minimum rank and maximum nullity. Finally, observe that for any field F , mrF (G) ≤ n − 1. The minimum rank problems asks, given a graph G and field F , what is mrF (G). Example 6. Let F = R and let P mr(P ) ≤ 3. If we label the vertices of of P is 0 1 1 0 0 1 0 0 be the path on 4 vertices. Clearly P sequentially, the adjacency matrix 0 0 1 0 0 1 1 0 which has rank 3. This minimizes the rank among all matrices whose graph is P , so mr(P ) = 3. In general, if Pn is the path on n vertices, mr(Pn ) = n − 1. Example 7. Let F = R and let Kn be the complete graph on n vertices. Clearly mr(Kn ) ≥ 1. Kn is the graph of the n × n matrix of all 1s, which has rank 1, and thus mr(Kn ) = 1. Conversely, if G is a connected graph with n vertices and mr(G) = 1, then G ' Kn . We now explore the relationship between MF (G) and Z(G). Definition 3. Let F be a field. The support of a vector x = (x1 , x2 , . . . , xn ) with components in F , is the set supp(x) = {i : xi 6= 0}. 4 Proposition 2 (AIM Group, [3]). If F is a field, A an n × n matrix with entries in F and null(A) > k, then there is a nonzero vector x ∈ ker(A) vanishing in any k specified positions. The proof is omitted. The interested reader may try to prove the proposition themselves, or find the proof in [3]. Proposition 3 (AIM Group, [3]). Let F be a field, Z a zero forcing set for a graph G = (V, E), and let A be a symmetric matrix with entries in F such that G(A) ' G. If x ∈ ker(A) and supp(x) ∩ Z = ∅, then x = 0. Proof. If Z = V , we are done. If Z 6= V , then since Z is a zero forcing set for G, we must be able to perform a color change. So there is a black vertex u (note that xu is required to be 0) with exactly one neighbor v colored white (so there is no restriction on xv , yet). Consider the u-th component of the vector Ax. By the definition of multiplication of a vector by a matrix, this is given by X auw xw . (Ax)u = w∈V Writing u ∼ w to indicate that u is adjacent to w, we may expand the right hand side of the above equation as X X X auw xw = auu xu + auv xv auw xw + auw xw = auu xu + w∈V w6∼u w∼u since auw = 0 for every w 6∼ u, xw = 0 if w ∈ Z and every neighbor of u is in Z except for v. Then since xu = 0 and x ∈ ker(A), we must have xv = 0. Each color change requires another component of x to be 0, and since Z is a zero forcing set, this implies that x = 0. Remark. The previous proposition explains the name zero forcing. Proposition 4 (AIM Group, [3]). Let G = (V, E) be a graph and let Z ⊆ V be a zero forcing set for G. Then M F (G) ≤ |Z|, and thus M F (G) ≤ Z(G). Proof. Suppose M F (G) > |Z| and let A be a symmetric matrix with entries in F such that G(A) ' G with null(A) > |Z|. By Proposition 2, there is a nonzero vector x ∈ ker(A) that vanishes on all vertices in Z. But x = 0 by Proposition 3, a contradiction. Since mrF (G) + M F (G) = n, Proposition 4 implies that mrF (G) = n − M F (G) ≥ n − Z(G). 5 4 k-forcing The k-forcing number is a generalization of the zero forcing number, and was introduced by myself, Yair Caro, Randy Davila and Ryan Pepper in [1]. Definition 4 (DA, Y. Caro, R. Davila and R. Pepper, [1]). Let G be a simple, finite and undirected graph and k a positive integer. Color each vertex of G either black or white. (a) k-color change rule: If u is a black vertex of G and u has at most k white neighbors, then u “k-forces” each of these neighbors to become black. (b) Derived coloring: The derived coloring of G is the coloring obtained by applying the k-color change rule as many times as possible. (c) k-forcing set: A nonempty set F of vertices is called a k-forcing set if, when all of the vertices in F are colored black and all other vertices of G are colored white, the derived coloring of G has no white vertices. (d) The k-forcing number of G, denoted Fk (G) is the minimum |F | over all k-forcing sets F . Remark. Any 1-forcing set is a zero forcing set, and any zero forcing set is a 1-forcing set. Thus F1 (G) = Z(G) for any graph G. Example 8. Let k = 2 and consider the following graph G and coloring of G: The set of black vertices is easily seen to be a 2-forcing set, since 4 2-forces 3 and 5, and then 2 can 2-force 6 and 1. Thus F2 (G) ≤ 2. However, F2 (G) = 1, since the set containing only the vertex 4 (or only the vertex 1) is a 2-forcing set. 6 Example 9. The k-forcing number of the path Pn is 1 for all k, since for any k the set containing one of the degree one vertices of Pn is a k-forcing set. Remark. Any k-forcing set is a (k + 1)-forcing set, and thus Fk ≥ Fk+1 for all positive integers k. Also, if G is connected and k is at least the maximum degree of G, then Fk (G) = 1. For this reason, we often restrict k to be at most the maximum degree. Example 10. For any positive integer k, if Kn is the complete graph on n vertices, then Fk (Kn ) = max{n − k, 1}. To see this, note that if k ≥ n − 1, then Fk (G) = 1. Otherwise, if k ≤ n − 2, color any set of n − k vertices black and the remaining vertices white. Each black vertex has exactly k white neighbors, and thus k-forces all of the white vertices. Thus Fk (Kn ) ≤ n − k. On the other hand, if at most n − k − 1 vertices are colored black, then each black vertex has at least k + 1 white neighbors, so no vertex can k-force. The following proposition gives a generalization of the lower bound for zero forcing in Proposition 1. Proposition 5 ([1]). Let G = (V, E) be a graph with minimum degree δ and let k be a positive integer. Then Fk (G) ≥ δ − k + 1. Proof. Let F be a smallest k-forcing set for G and color the vertices of F black (and all other vertices white). If F = V , then |F | = |V |, so the inequality is satisfied trivially. If F 6= V , then let v ∈ F − V be a white vertex that is forced by u ∈ F . Since u forces, u must have at most deg(u) − k ≥ δ − k black neighbors. Hence, counting u, there are at least δ − k + 1 vertices in F. Proposition 6 ([1]). If G is connected with maximum degree ∆ ≥ 1, then F∆−1 (G) ≤ 2 with equality holding if and only if G is ∆-regular. Proof. Let F = {u, v} be any set of two adjacent vertices. Color each vertex in F black and all remaining vertices white. Each vertex in F has at most ∆−1 white neighbors, and thus (∆−1)-forces. Each vertex that was (∆−1)forced to become black by either u or v has at moast ∆ − 1 white neighbors, and consequently (∆ − 1)-forces. Again, each newly colored black vertex has 7 at most ∆ − 1 white neighbors, and this will be true at each step of applying the (∆−1)-color change rule. Thus, since G is connected, the derived coloring colors each vertex of G black. Hence F is a (∆ − 1)-forcing set, which means that Fk (G) ≤ 2. To see that equality holds if and only if G is ∆-regular, first observe that if only a single vertex u of G is colored black, and all remaining vertice are colored white, then u does not (∆ − 1)-force since u has at ∆ white neighbors. Hence F∆−1 (G) ≥ 2, so that F∆−1 (G) = 2. The converse is proven by contrapositive. If G is not ∆-regular, then there exists a vertex u of G with deg(u) < ∆. Color u black and all other vertices of G white. Then u has at most ∆ − 1 white neighbors and thus (∆ − 1)-forces. Each vertex of that was (∆ − 1)-forced by u to become black has at most ∆ − 1 white neighbors, and this will be true at each step that the (∆ − 1)-color change rule is applied. Thus the derived coloring colors each vertex of G black, which means that F∆−1 (G) = 1. 5 Upper bounds for the k-forcing number In this section we surevey, mostly without proof, some upper bounds on the k-forcing number of a graph. The missing proofs may be found in [1]. The first result gives a general upper bound in terms of easily computed graph invariants. Theorem 7. Let k be a positive integer and let G = (V, E) be a graph on n ≥ 2 vertices with maximum degree ∆ ≥ k and minimum degree δ ≥ 1. Then (∆ − k + 1)n Fk (G) ≤ . ∆ − k + 1 + min{δ, k} When the minimum degree of G is at most k, Theorem 7 has the following immediate corollary: Corollary 8. Let k be a positive integer and let G = (V, E) be a graph with maximum degree ∆ ≥ k and minimum degree 1 ≤ δ ≤ k. Then Fk (G) ≤ (∆ − k + 1)n . ∆+1 The upper bound in Corollary 8 is sharp for the the complete graph Kn when k ≤ n − 1. First, recall from Example 10 that Fk (Kn ) = n − k whenever 8 k ≤ n − 1. The maximum degree and minimum degree of Kn are both n − 1, so the bound in Corollary 8 gives Fk (Kn ) = (n − k)n ((n − 1) − k + 1)n = = n − k. (n − 1) + 1 n In the special case that k = 1, both Theorem 7 and Corollary 8 have the following immediate corollary: Corollary 9. Let G = (V, E) be a graph with maximum degree ∆ ≥ 1 and minimum degree δ ≥ 1. Then F1 (G) = Z(G) ≤ ∆ n. ∆+1 Remark. The upper bound in Corollary 9 is also sharp for Kn . The appeal of Corollary 9 is that it bounds ratio of the zero forcing number to the number of vertices of a graph G. For example, if G is a graph with n vertices and ≤ 34 . maximum degree ∆ = 3, then Corollary 9 implies that Z(G) n If we impose the mild restriction that a graph G is connected, we can improve the bound in Theorem 7 (the improvement is found in Theorem 12 below). First, we need a few concepts. Let k be a positive integer. We say that a graph G = (V, E) is k-connected if the subgraph induced by the complement of any set of at most k − 1 vertices is connected. Next we introduce the notion of a connected k-dominating set. Let G = (V, E) be a connected graph and let D ⊂ V be any subset of vertices that induces a connected subgraph. We say that D is a connected k-dominating set for G if every vertex in V − D has at least k neighbors in D. The following theorem shows that there is a relationship between k-forcing sets and connected k-dominating sets. Theorem 10. Let k be a positive integer and let G = (V, E) be a k-connected graph with n > k vertices. If F is a smallest k-forcing set such that the subgraph induced by V −F is connected, then V −F is a connected k-dominating set. Let γk,c (G) denote the cardinality of a smallest connected k-dominating set for G. We call γk,c (G) the connected k-domination number of G. Theorem 10 has the following corollary: 9 Corollary 11. Let k be a positive integer and let G = (V, E) be a k-connected graph with n > k vertices. Then Fk (G) ≤ n − γk,c (G). Proof. Any set S of n − 1 vertices of G is a k-forcing set for G, and the subgraph induced by V −S is connected trivially. Thus there exists a smallest k-forcing set F for G for which V − F induces a connected subgraph. By Theorem 10, V −F is a connected k-dominating set for G. Since |F | ≥ Fk (G), it follows that γk,c (G) ≤ |V − F | = |V | − |F | ≤ n − Fk (G). Rearranging gives the desired upper bound. The next theorem gives an upper bound for the k-forcing number of a kconnected graph in terms of the number of vertices, the maximum degree, and k. While the proof is omitted, we remark that the proof uses the result of Theorem 10. Theorem 12. Let k be a positive integer and let G = (V, E) be a k-connected graph with n > k vertices. If G has maximum degree ∆ ≥ 2, then Fk (G) ≤ (∆ − 2)n + 2 . ∆+k−2 Remark. The upper bound in Theorem 12 is an improvement over the upper bound in Theorem 7 whenever k = 1 or 2, and G is a k-connected with n > k vertices and minimum degree δ ≥ k. This can be seen by comparing the upper bounds in Theorems 7 and 12 when k = 1 and k = 2. When k = 1, Theorem 12 has the following immediate corollary that is reminiscent of Corollary 9: Corollary 13. Let G be a connected graph with maximum degree ∆ ≥ 2. Then (∆ − 2)n + 2 . F1 (G) = Z(G) ≤ ∆−1 Remark. Corollary 13 is easily seen to be sharp for cycles, complete graphs, and complete bipartite graphs with equal parts (a bipartite graph is any graph whose vertices can be partition into two parts A and B such that any edge has one end vertex in A and the other in B; a complete bipartite graph contains every possible edge between the parts A and B). No other families of graphs are known to satisfy equality in the upper bound. In fact, it is conjectured that these families characterize the case of equality. 10 6 Open problems Here we list some open problems concering zero forcing and k-forcing. It would be interesting if there is any connection between k-forcing and the minimum rank problem. The first problem considers one way in which such a relationship could be made. Problem 1. Let k be a positive integer, F a field and G = (V, E) a graph. Is there a constant ck (G) such that M F (G) ≤ ck (G)Fk (G) ≤ Z(G)? Another possible application of zero forcing and k-forcing is in the study of Cayley graphs of finite groups. Problem 2. Let G be a finite group. For every positive integer k, we could define Fk (G) to be the k-forcing number of the Cayley graph of G. Is Fk (G) related to any properties of the group? The definition of zero forcing given by the AIM Group was motivated by a problem in linear algebra. The next problem is a natural question in this setting. Problem 3. Let k be a positive integer, G a graph and A the adjacency matrix of G. Is Fk (G) related to the spectrum of A? Are there any bounds for Fk in terms of the spectrum of A? Finally, one would expect the k-forcing number of an edge dense graph to be larger than the k-forcing number of a graph with the same number of vertices but few edges. How well does Fk (G) measure the edge density of G? References [1] David Amos, Yair Caro, Randy Davila, and Ryan Pepper, Upper bounds on the k-forcing number of a graph, Discrete Applied Mathematics (2014), Accepted, in press. [2] Daniel Burgarth and Vittorio Giovannetti, Full control by locally induced relaxation, Phys. Rev. Lett. 99 (2007), 100501. [3] AIM Minimum Rank Special Graphs Work Group, Zero forcing sets and the minimum rank of graphs, Linear Algebra and its Applications 428 (2008), no. 7, 1628 – 1648. 11