# Packing functions and graphs with perfect closed neighbourhood matrices

In this work we consider a straightforward linear programming formulation of the recently introduced {k}-packing function problem in graphs, for each fixed value of the positive integer number k. We analyse a special relation between the case k = 1 and k ≥ 2 and give a sufficient condition for optimality ---the perfection--- of the closed neighbourhood matrix N[G] of the input graph G. We begin a structural study of graphs satisfying this condition. In particular, we look for a characterization of graphs that have perfect closed neighbourhood matrices which involves the property of being a clique-node matrix of a perfect graph. We present a necessary and sufficient condition for a graph to have a clique-node closed neighbourhood matrix. Finally, we study the perfection of the graph of maximal cliques associated to N[G].

## Authors

• 1 publication
• 1 publication
• 2 publications
04/06/2021

### Coloring graph classes with no induced fork via perfect divisibility

For a graph G, χ(G) will denote its chromatic number, and ω(G) its cliqu...
01/06/2021

### Perfect domination, Roman domination and perfect Roman domination in lexicographic product graphs

The aim of this paper is to obtain closed formulas for the perfect domin...
07/12/2019

### Linear programming bounds for cliques in Paley graphs

The Lovász theta number is a semidefinite programming bound on the cliqu...
12/13/2018

### A characterization of claw-free CIS graphs and new results on the order of CIS graphs

A graph is CIS if every maximal clique interesects every maximal stable ...
04/30/2021

### Complementation in t-perfect graphs

Inspired by applications of perfect graphs in combinatorial optimization...
09/07/2019

### Perfect Matchings, Rank of Connection Tensors and Graph Homomorphisms

We develop a theory of graph algebras over general fields. This is model...
07/04/2012

### Stable Independence in Perfect Maps

With the aid of the concept of stable independence we can construct, in ...
##### This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

## 1 Preliminaries and notation

### 1.1 Graphs and binary matrices

Throughout this work we consider finite simple connected graphs , where is its node set and is its edge set. When we need to emphasize the relationship between the graph and its node and edge sets, we write and .

The complementary graph of , denoted by , is such that and .

Given and , we denote by the subgraph of induced by the nodes in , i.e., the graph having node set and edge set . Then is a node induced subgraph of , , if for some .

Given the graph , the neighbourhood of is and its closed neighbourhood is . When we need to emphasize the graph , we write (respectively, ) instead. A vertex is universal if .

For any positive integer number , denotes the graphs with distinct nodes corresponding to a complete graph, i.e. a graph where for all distinct nodes , .

A cycle, denoted by , is a graph with distinct nodes, say , and edges for together with the edge . We say that and , and and are consecutive nodes in the cycle. The length of the cycle is the number of its edges. An even (odd

) cycle is a cycle of even (odd) length.

A chord in a cycle is an edge connecting two non consecutive nodes in the cycle.

A hole (resp. anti-hole) is an induced cycle (the complementary graph of a cycle) in the graph (i.e. a hole has no chords).

A graph is chordal if every cycle in of length at least four have a chord. is strongly chordal if it is chordal and also every cycle of even length at least 6 in has an odd chord, i.e., an edge that connects two nodes that are an odd distance apart from each other in the cycle.

A clique in is a subset of pairwise adjacent nodes in .

Although it is not the original definition of perfection in graphs, we say that a graph is perfect if it has neither an odd hole nor an odd anti-hole as an induced subgraph, due to the Strong Perfect Graph Theorem [2].

In this work we make use of strong ties between all the above mentioned graph definitions from the polyhedral point of view, and then we need to present some -matrices associated with graphs.

Given , its closed neighbourhood matrix is a -matrix with rows and columns indexed in and such that when or and otherwise. Throughout this work we can use a different permutation of the node set in rows and columns, but in this case it is mentioned explicitly.

Matrix denotes the square matrix whose entries are all 1’s, and

the identity matrix, both of appropriate sizes.

A -matrix having rows and (non zero) columns is perfect if is an integer polyhedron, i.e. all extreme points of are

-vectors, where

denotes the vector of all 1’s of appropriate size.

The clique-node matrix of a graph , denoted by , is the -matrix whose columns are indexed by the nodes of and whose rows are the incidence vectors of maximal cliques in . Chvátal [3] proved that a -matrix with no zero columns and no dominating rows is a perfect matrix if and only if it is the clique-node matrix of a perfect graph.

We say that a matrix is an extended clique-node matrix of if it is a square -matrix that contains as a row submatrix and all its remaining rows (if any) are incidence vectors of cliques in . Note that the presence of dominating rows do not affect the integrality of the polyhedrom and thus, Chvátal’s result mentioned above can be restated as the following

[3] A square -matrix with no zero columns is a perfect matrix if and only if it is an extended clique-node matrix of a perfect graph.

The following characterization, derived from a similar result for clique-node matrices exposed in [4], is crucial for our analysis, since it yields a polynomial time algorithm to check if a -matrix is an extended clique-node matrix.

[4] Let be a -matrix. The following statements are equivalent:

1. is an extended clique-node matrix,

2. if

 ⎛⎜⎝0111…11011…11101…1⎞⎟⎠ (1)

is a submatrix of , say, with columns , where , then contains a row such that for .

3. If is a submatrix of , where , then contains a row such that for every column of .

Given a -matrix , the clique graph of , , is the graph having one node for each column of and being an edge when there is a row having value 1 in the -th and -th columns of .

Under this definition, the two last conditions in Theorem 1.1 say that the columns in indexed in form a clique in the graph . A necessary and sufficient condition for to be an extended clique-node matrix is that the incidence vector of every of these cliques are also rows of .

According to Chvátal’s result, the matrix is perfect if and only if is the clique-node matrix of and is a perfect graph.

When for some graph , we simply write instead of , and refer to it as the clique graph associated with .

### 1.2 Limited packing functions

For a graph and a positive integer , a function is a -packing function of [9], if for each node ,

 f(N[v])=∑w∈N[v]f(w)≤k.

The maximum possible value of over all -packing functions of is denoted as .

These concepts generalize the notion of -limited packing introduced by Gallant et al. [6]. A -limited packing in is a function such that for each node ,

 f(N[v])≤k.

Again, the maximum possible value of over all -limited packings in is denoted by .

From their definitions, it holds and for every graph . Nevertheless, this last inequality may or may not be strict. In Figure 1 there is a graph , an optimal -packing function and a -limited packing in whose optimal values do not satisfy the equality. In the following two figures we have represented the value of the function next to each of the nodes.

The optimization problem associated with this concept is the following:

-PACKING FUNCTION (PF, for fixed positive integer ): Given a graph , find .

By using simple combinatorial arguments, in [7] it is proved that , for every integer number and every graph . Nevertheless, note that this inequality may be strict as the following example shows.

Let be the graph depicted in Figure 2. It is clear that but , as shown in the figure.

Besides, it is known that PF is linear time solvable in strongly chordal graphs [9] and it is NP-complete for chordal graphs [7].

We present a natural Integer Linear Programming formulation of PF for fixed . We give a sufficient condition for optimality, the perfection of the closed neighbourhood matrix of the input graph.

A different sufficient condition involving dominating sets was already given in [7]. The step forward with this contribution is that, knowing that checking matrix perfection can be performed in polynomial time for any input graph (see Theorem 1.1), we are allowed to find new polynomial instances of problem PF. In the last section, we begin a structural study of graphs satisfying the new sufficient condition and present a characterization of graphs that have closed neighbourhood matrices that are also clique-node matrices and necessary conditions for the associated clique graphs to be perfect.

## 2 A sufficient condition for optimality of the {k}-Packing Function Problem

We now consider an integer programming formulation for PF.

Given with , let for . For , , and the associated matrix-vector product.

For fixed integer positive number , PF can be formulated as follows:

 L{k}(G)=max{1⋅x:N[G]⋅x≤k1,x∈{0,1,…,k}n}.

Let be the optimal value of the linear relaxation of this problem for , i.e. allowing to take any value in . Then, we have:

For each fixed positive integer and graph , if is perfect then can be found in polynomial time and .

It is clear that, if is a solution of the linear relaxation of the problem above for , then and thus is a solution of the linear relaxation of PF, and conversely. Consequently,

 L{k}(G)≤kmax{1⋅y:N[G]⋅y≤1,y∈[0,1]n}=kLR1(G).

Notice that if is perfect, then and, therefore, . Recall that , for each and every graph [7].

In all, we have the desired equality.

Notice that the cycle is an example that shows that the sufficient condition in Theorem 2 is not necessary. In fact, for every that is not a mutiple of three and but is not perfect since it is not a clique-node matrix.

In the next section we focus on the family of graphs for which their closed neighbourhood matrices are perfect.

## 3 Perfect closed neighbourhood matrices

In this section we begin a structural study of graphs satisfying the sufficient condition in Theorem 2.

Let be the family of graphs whose closed neighbourhood matrices are perfect. There are some trivial graphs in the family , for instance complete graphs. Let us now introduce some conditions for a graph to belong to the family .

Recall that, given , we consider the clique graph associated with it, denoted by . Then, a graph belongs to if and only if is an extended clique-node matrix and is a perfect graph.

Observe that the characterization of extended clique-node matrices shows that if , for every maximal clique in there must be a node in adjacent to all the nodes in this clique.

Therefore,

If a graph has a universal node then it belongs to .

Let , for and a universal node in . Then, for all .

Theorem 1.1 implies that is a clique-node matrix and the graph is a complete graph, thus perfect.

A wheel graph on nodes is a cycle on nodes together with a universal node outside the cycle. From the previous result, wheel graphs are also in . However, a wheel of at least five nodes is not even chordal. Therefore, there are some trivial non chordal graphs in .

A particular subset of chordal graphs is the set of strongly chordal graphs. A well-known characterization of them is given by the total balancedness of their closed neighbourhood matrices [8]. A -matrix is totally balanced if it does not contain as a submatrix the (node-edge) incidence matrix of a cycle of length at least three. Due to Farber [8], a graph is totally balanced if and only if its closed neighbourhood matrix is totally balanced. The fact that strongly chordal graphs are in follows from the fact that totally balanced matrices are perfect [5].

Clearly, the family of chordal graphs and do not coincide but intersect each other in, at least, strongly chordal graphs.

The family of web graphs generalize other families of graphs having strong symmetry, like complete graphs, cycles and the complementary graphs of cycles. Therefore it is important to determine which of them belong to .

Given positive integer numbers and , the web graph has nodes, say , and is an edge if and differ by at most (mod ) and . The complementary graph of a web is an antiweb. Antiwebs have also a circular symmetry of their maximum cliques and stables sets. If the web is the complete graph of nodes. Clearly, for the graph is an odd cycle and is the complementary graph of an odd cycle.

A circulant matrix , for is the -matrix having columns indexed in and rows given by all the incidence vectors of for , addition mod .

There is a strong relationship between circulant matrices and web graphs. In fact, is a clique-node matrix if and only if . Moreover, in this case, . This follows from the fact that, when the maximal cliques in are defined by consecutive nodes in the web and when we have some more maximal cliques, which are defined by appropriate non consecutive nodes, as in the case of .

Also, perfect web graphs are completely identified.

The authors in [1] completely characterized the disjunctive rank of the clique relaxation of the stable set polytope, i.e., the minimum number of nodes that must be deleted from a web in order to become a perfect graph. As a consequence, they proved that a web graph is perfect if and only if it is a complete graph or .

Using the results in these two remarks we can prove the following

Complete graphs are the only web graphs in . We have already mentioned that complete graphs belong to .

Let be a web graph with and . It is clear that . After Remark 3, if and only if . In order to belong to , has to be the clique-node matrix of a perfect graph. Remark 3 implies that , which contradicts the fact that .

Due to their importance on the sequel, we present some small examples of graphs whose closed neighbourhood matrices are not extended clique-node matrices.

The closed neighbourhood matrices of the chordless cycles , for are not clique-node matrices. This holds since for , and they are not clique-node matrices after Remark 3.

Consider again graph depicted in Figure 3. From Theorem 2 and taking into account Example 1.2, it is clear that is not a perfect matrix.

Also, this can be seen regarding at the neighbourhood matrix, since there are three nodes of degree 4 in defining a structure in as in (1) for but has no row of all ones and thus it is not an extended clique-node matrix.

If , we next show that a necessary and sufficient condition for to be an extended clique-node matrix is that all the induced subgraphs of in (if any) have a universal node in .

Let be a graph. is an extended clique-node matrix if and only if for every and , there exists such that .

Suppose is an extended clique-node matrix. From examples 3 and 3, it is clear that does not belong to .

Let with and consider its neighbourhood matrix :

• If then . Theorem 1.1 with implies that there must be a row of corresponding to nodes not in with 1’s in the columns associated with the nodes in , say . This means that there is a node adjacent to every node in and the result holds.

• If then which is not an extended clique-node matrix after Example 3, but that yields the graph . Since is an extended clique-node matrix, there must be such that is a clique in , and moreover, its incidence vector is a row of .

• Similarly, if , then which is not an extended clique-node matrix after Example 3, but that yields the graph . In fact there are two maximal cliques in , say and , whose incidence vectors are not rows of .

Since is an extended clique-node matrix, there must be such that and are cliques in , and their incidence vectors are rows of . If , there is a node adjacent to every node in and we are done. If not, each of the nodes and generates three cycles of length 4 in . By the first case, each of them must have a universal node. This implies, recursively applying the previous case, that there must exist another node in adjacent to all the nodes in .

• Finally, if , the graph is a complete graph, and then there must be a node in adjacent to all the nodes in .

Now suppose is not an extended clique-node matrix. According to Theorem 1.1, contains

 M=⎛⎜⎝0111…11011…11101…1⎞⎟⎠

as a submatrix, say, with columns , for some , but it does not contain a row with for .

It is enough to prove that the submatrix of of order 3 forces to have at least one induced subgraph of the family and, by the assumption that does not contain a row with for , we conclude that there is not a vertex such that .

Clearly, the submatrix of order 3 itself is not the neighbourhood matrix of any induced subgraph of . Then, is a submatrix of an induced subgraph of , say , with at least 4 nodes. Let us prove that has at most 6 nodes. Let us analyse the possible cases:

• If , we have already seen that is the neighbourhood matrix of .

• If , say , w.l.o.g. some rows of may be permuted in such a way that they correspond to the nodes listed on the table of the left-hand-side of Figure 4. We have some freedom to choose the values of . If , then is the neighbourhood matrix of . If then has as a node induced subgraph (for instance when and ).

• If , say , w.l.o.g. may be indexed as in Figure 4. Again, we have some freedom to choose the values of for .

If for all , then is the neighbourhood matrix of . If for and for (or for and for ) then is the neighbourhood matrix of (Figure 3). If for all , then is the neighbourhood matrix of the 3-pyramid also in Figure 3, which has as a node induced subgraph. It is easy to check that, for the remaining possibilities for the ’s and , has or as a node induced subgraph (see Figure 3).

Recall that we need two conditions for a graph to belong to . We have explored the fact that is an extended clique-node matrix. Let us now focus on the perfection of the clique graph associated with .

We have already mentioned in Proposition 3 that the cycles for do not belong to . In this case, this holds from the fact that the associated graph is imperfect when due to Remark 3.

If is such that

• there is whose associated clique graph is imperfect (and thus ),

• for all , for some

then the clique graph is imperfect, and then .

Let and . By hypothesis, for some and then the graph obtained from after deleting the nodes in coincides with . This shows that is imperfect.

This result asks for some knowledge of subgraphs of not in with imperfect clique graph. We know that all cycles of length at least 7 satisfy this property. In fact, we can ask if the converse of this result is true, that is, is it enough to look for node induced cycles in the graph in order to have an imperfect clique graph? The answer is negative as shown by the following infinite family of graphs.

Let where and is such that the even nodes in form a clique and for every odd node .

In this case, the clique graph is the union of an odd hole and a complete graph , where also all the nodes in this complete graph are adjacent to every node in the cycle (i.e., is the complete join of the cycle and the complete graph). Clearly, is imperfect. Also, is a chordal graph, i.e., it has no cycle of length at least 4.

## 4 Conclusions

In this paper we have analysed the relationship between the perfection of closed neighbourhood matrices and the -Packing Function Problem in graphs.

We defined a family of graphs, , in which the -Packing Function Problem can be solved in polynomial time. Then, this contribution is devoted to look for a nice characterization of graphs in this family.

The definition of the family and Chvátal’s result ask for two conditions on the closed neighbourhood matrix of a graph in : the property of being a (extended) clique-node matrix and the perfection of its associated clique graph.

Firstly, given a graph , we have studied under which conditions the closed neighbourhood matrix is an extended clique-node matrix and found a set of some small graphs which —in case of being a subgraph of in the family— need a universal node in . We could completely characterized the family of graphs that have extended clique-node matrices, although some of them have an imperfect clique graph.

Secondly, we provided some sufficient conditions for a graph not to be in the family due to the imperfection of its associated clique graph.

Based on the results in Proposition 3 and Example 3, it is clear that even if a graph satisfies the property of having an extended clique-node closed neighbourhood matrix, it is important to study in more depth the imperfection of the associated clique graph and, if possible, identify some minimal forbidden subgraphs for .

In Figure 5 we have represented the relationship between the well-known graph classes of chordal and strongly chordal graphs and the family introduced in this work. Also, we have depicted some other examples of graphs mentioned in this work (like web graphs, cycles, wheels and the graph ) in order to clarify our findings.

## References

• [1] Bianchi, S., M. Escalante and M. S. Montelar, Lift-and-project ranks of the stable set polytope of joined a-perfect graphs, Discrete Applied Mathematics 210 (2016), 176–184.
• [2] Chudnovsky, M., G. Cornuéjols, X.Liu, P.Seymour and K.Vušković, Recognizing Berge Graphs, Combinatorica 25 (2005), 143–187.
• [3] Chvátal, V., On Certain Polytopes Associated with Graphs, Journal of Combinatorial Theory (B) 18 (1975), 138–154. 1991) 166–190.
• [4]

Cornuéjols, G., “Combinatorial Optimization. Packing and Covering” SIAM, Philadelphia, 2001.

• [5] Fulkerson, D. R., A.J. Hoffman and R. Oppenheim, On balanced matrices, Mathematical Programming 1 (1974), 120–132.
• [6] Gallant, R., G. Gunther, B. Hartnell and D. Rall, Limited packing in graphs, Discrete Applied Mathematics 158 12 (2010), 1357–1364.
• [7] Dobson, M. P., E. Hinrichsen and V. Leoni, On the complexity of the -packing function problem, International Transactions in Operations Research 24 (2017), 347–354.
• [8] Farber, M., Characterizations of Strongly Chordal Graph, Discrete Mathematics 43 (1983), 173–189.
• [9] Hinrichsen, E. and V. Leoni, -packing functions of graphs. Lecture Notes in Computer Science (2014), 325–335.