Graphs, friends and acquaintances
Abstract
As is well known, a graph is a mathematical object modeling the existence of a certain relation between pairs of elements of a given set. Therefore, it is not surprising that many of the first results concerning graphs made reference to relationships between people or groups of people. In this article, we comment on four results of this kind, which are related to various general theories on graphs and their applications: the Handshake lemma (related to graph colorings and Boolean algebra), a lemma on known and unknown people at a cocktail party (to Ramsey theory), a theorem on friends in common (to distanceregularity and coding theory), and Hall’s Marriage theorem (to the theory of networks). These four areas of graph theory, often with problems which are easy to state but difficult to solve, are extensively developed and currently give rise to much research work. As examples of representative problems and results of these areas, which are discussed in this paper, we may cite the following: the Four Colors Theorem (4CTC), the Ramsey numbers, problems of the existence of distanceregular graphs and completely regular codes, and finally the study of topological proprieties of interconnection networks.
Keywords: Graph, Edgecoloring, Boolean Algebra, Ramsey Theory, Distanceregularity, Spectral Graph Theory, Completely Regular Code, Hall’s Marriage Theorem, Menger’s Theorem.
AMS classification: 05C50, 05C05, 05C75.
1 Introduction
A graph is a mathematical structure consisting of a vertex set and a set of edges (or nonordered pairs of vertices). Normally, each vertex is represented by a point and each edge by a line joining vertices and . Graph theory belongs to combinatorics, which is the part of mathematics that studies the structure and enumeration of discrete objects, in contrast to the continuous objects studied in mathematical analysis. In particular, graph theory is useful for studying any system with a certain relationship between pairs of elements, which give a binary relation. It is therefore not surprising that many of the problems and results were originally stated in terms of personal relationships. For example, one of the most simple results is the Handshake lemma: At a cocktail party, an even number of people shake an odd number of hands. There is also the socalled Friendship theorem: At a party, if each pair of people has exactly one friend in common, then there is somebody who is friend of everybody. The first and most appealing proof of this theorem is due to Paul Erdős (with Alfred Rényi and Vera Sós), a Hungarian mathematician, probably the most prolific of the 20th century, who like Euler enjoyed coining sentences such as “A mathematician is a device for turning coffee into theorems” or “Another roof, another proof”. The latter phrase shows his great capacity and predisposition for collaborating with other authors from all over the world (he had 509 coauthors). From Erdős we have the Erdős number: the coauthors of Erdős have Erdős number 1, the coauthors of the coauthors of Erdős have Erdős number 2, etc. For more information on Erdős, see Hoffman [30].
It is considered that the first paper on graph theory was published in 1736. Its author was the great Swiss mathematician Leonhard Euler, about who it is said that he wrote papers in the half an hour between the first and the second calls for lunch. This first paper is about the existence of a possible walk across the Königsberg bridges; see Euler [13]. This city was the capital of Oriental Prussia, the birthplace of Immanuel Kant. Nowadays it corresponds to the Russian city of Kaliningrad. The problem of the Königsberg bridges is related to the puzzle of drawing a figure without raising the pencil from the paper and without passing twice through the same place. In the original problem, it was asked if it was possible to walk through the city by crossing all the bridges only once. With an ingenious reasoning, which in fact does not explicitly use any graph, Euler proved the impossibility of this walk.
Another of the most famous problems in graph theory, not solved until 1977 by Appel, Haken and Kock [3, 2], is the Four Colors theorem (4CT), which states that the countries of any map drawn in the plane can be colored with four colors, such that countries with a common border (different from a point) bear different colors. This theorem is regarded as the first important result to be proved using a computer, because in a part of its proof 1,482 configurations were analyzed. For this reason, not all mathematicians accept it. Twenty years later, Robertson, Sanders, Seymour and Thomas [38] gave an independent proof, which is shorter, but also requires the use of a computer, because of the 633 configurations analyzed.
As we have already stated, graph theory is used to study different relations. A first example is an electric circuit, with all its components and its connections. In telecommunications, graph theory contributes to the modeling, design and study of interconnection or communication networks. For instance, interconnection networks are used in multiprocessor systems, where some processors undertake a task of exchanging information, and in local networks consisting of different computers placed at a short distances, which exchange data at very high speed and low cost. As regards communication networks, nowadays the most important example is the Internet, which makes the communication and exchange of data possible between computers all around the world. In fact, we are experiencing a communication revolution, so that we could say that we are ‘weaving’ the communication network.
2 Shaking hands: Colorings and Boolean algebra
In a graph , the degree is the number of adjacent vertices to vertex , namely, the number of incident edges to . We denote by the maximum degree of all the vertices of and by the minimum degree.
We begin with one of the most simple results about graphs, which states that the sum of the degrees of the vertices in equals twice the number of edges in :
(1) 
since in the degree sum, we count each edge twice because each edge is incident to two vertices. From here, we obtain the inequalities:
(2) 
Although these results are apparently trivial, they have some interesting corollaries, such as the following:

Every graph has an even number of vertices with odd degree.
This is the socalled Handshake lemma, because it can be stated as follows: At a cocktail party, the number of people who shake an odd number of other people’s hands is always even.

Every regular graph a graph is regular if all its vertices have degree , with odd, has an even number of vertices.

Every planar graph that is, it can be drawn on the plane without edge crossings with girth the girth is the length of the shortest cycle and number of edges satisfies
(3)
To prove , we need the wellknown Euler formula [14] published between 1752 and 1753, and already observed by Descartes in 1640, which can be proved by induction and states that every planar graph with vertices, edges and regions satisfies
(4) 
In this formula, the number of regions includes the exterior one (that is, the ‘sea’, if we have a map or if the graph is imbedded on a sphere). For example, the Euler formula is satisfied by the graphs of the Platonic solids shown in Figure 1. In fact, this formula gives necessary conditions for the existence of these regular polyhedra; see Rademacher and Toeplitz [36]. In proving (4), the key fact is that the removing of a vertex with degree (and its incident edges) leaves a new planar graph whose number of regions, vertices and edges have been reduced, respectively, by , and units.
Returning again to the Euler formula, the number of regions can also be interpreted as the cardinality of the vertex set of the dual graph . Given a planar graph with vertices and edges forming regions, its dual graph has vertices representing the regions of , and there is an edge between two vertices if the corresponding regions are neighbors. Then, and . This interpretation provides a more symmetric Euler formula:
(5) 
which allows us to prove it without using induction, but rather by identifying both parenthesis in Equation (5) as the number of edges of two spanning trees and belonging to and , respectively. A spanning tree of a connected graph (that is, there is a path between any pair of vertices) is composed of the vertex set and edges without forming cycles. An example of this is shown in Figure 2, where each black continuous edge of (the graph of a cube ) belongs to , but where each black dashed edge corresponds to an edge of in (the graph of an octahedron). For more details, see Aigner and Ziegler [1].
In our case, the proof of is as follows: As each edge is the border of two regions and each region has at least edges, we have . Note that this inequality is obtained from (2), considering the dual graph, since , and . Using this inequality and Equation , we obtain .
As a particular case of , we have the following result:

In any planar graph the number of edges satisfies ; if it does not contain triangles , then ; and if it contains neither triangles nor squares , then .
From the first inequality, we can see that the complete graph , is not planar. A graph is complete if there is an edge between every pair of vertices. Similarly, from the second inequality, we also obtain that the complete bipartite graph is not planar. A bipartite graph (that is, the vertex set can be decomposed into two independent subsets such that vertices in every subset are not adjacent) is complete if each pair of vertices in different subsets are adjacent. See both graphs in Figure 3. Notice that, for instance, the third inequality turns out to be an equality in the case of the dodecahedron graph (see again Figure 1, and ).
In this context, we have the famous Kuratowski theorem [33], which characterizes planar graphs (see also the book by West [42, pp. 246–251] and the paper by Thomassen [40], where the relation between the planarity criterion and the Jordan Curve Theorem is explained):

A graph is planar if and only if it contains no homeomorphic subgraph to or .
Recall that a graph is homeomorphic to a graph if the edges of correspond to (independent) paths in .
From Equation (1) and again the inequalities in , we can prove the following:

Every planar graph contains a vertex of degree . Moreover, if does not contain triangles, then it has a vertex of degree .
Indeed, if denotes the number of vertices with degree , then from Equation (1) we have that
whence
so that for some , as claimed. The proof of the case without triangles is analogue.
The existence of a vertex with degree at most five allows us to prove, by induction, the Five Color theorem (5CT), which was first proved by Heawood [29] (see Aigner and Ziegler [1]):

Five colors suffice to get a vertexcoloring of a planar graph.
Recall that in a vertexcoloring, adjacent vertices have different colors.
First note that the result is trivially true for graphs with at most 5 vertices. Then, assume that it is also true for graphs with vertices, and let be a graph with vertices. We know that contains a vertex with degree . Let , , denote the adjacent vertices to . From the induction hypothesis, the graph (obtained from by removing vertex and all its incident edges) has a vertexcoloring with colors. Therefore, if (which is always the case when ), we can restore vertex and give it a color different from the colors of the adjacent vertices . Thus, we obtain a coloring of using at most 5 colors. Otherwise, if we can assume, without lost of generality, that we have a situation as shown in Figure 4 (where vertex has color , ). Now consider the paths with vertices alternatively colored 13 (with final vertices and/or ) and 24 (with final vertices and/or ). As is planar, these possible paths cannot cross each other (that is, they have neither crossed edges nor common vertices). Then if, for example, there exists the path 13 with initialfinal vertices , the path 24 with initial vertex cannot have as final vertex, but another vertex denoted by (see again Figure 4). Therefore, we can interchange the colors 24 in this path, so that gets color 4. We can then restore vertex and assign it color , obtaining a coloring of with 5 colors.
We now consider the case of giving one of three colors to each edge of a graph with maximum degree . This is called a free edgecoloring of . In particular, the (‘notfree’) edgecoloring of a cubic (regular) graph, also called Taitcoloring, corresponds to the case where adjacent edges receive different colors. As we will see later, if is a planar graph, the problem of the existence of Taitcolorings is closely related to the Four Color theorem (4CT). Moreover, we will also see that the construction of cubic graphs which cannot be Taitcolored leads to Boolean algebra, which is commonly used in the study of logic circuits. To this end, we introduce a natural generalization of the concept of ‘color’, which describes in a simple way the coloring (“0” or “1”) of any set of edges or, more abstractly, of any family of colors chosen between three different colors, say , such that color appears times. This situation can be represented by the coloringvector , where . Then, we say that has Boolecoloring 0, denoted by , if
whereas has Boolecoloring 1 (more specifically ), denoted by (or ), if
where . See Fiol and Fiol [20] for more information.
Recalling these definitions, the Boolecoloring of an edge with color is , and the Boolecoloring of a vertex , denoted by , is defined as the Boolecoloring of its incident edges, which can have either different or the same colors. In this context, it is curious to note the following facts:

If , then if and only if the incident edge to vertex has color .

If , then if both incident edges to vertex have the same color, and if not.

If , then if and only if the three incident edges to vertex have three different colors. Thus, in a Taitcoloring of a cubic graph, all its vertices have Boolecoloring 0.
Moreover, a natural sum operation can be defined in the set , of Boolecolorings in the following way: Given the colorings and represented, respectively, by the coloringvectors and , we define the sum as the coloring represented by the coloring vector . Then, is isomorphic to the Klein group, with 0 as identity, , and where ; see Table .
0  

0  0  
0  
0  
0 
Notice that, since every element coincides with its inverse, is 0 if is even and if is odd. From this simple fact, we can imply the following result (see Fiol [18]), which is very useful in the further development of the theory and can be regarded as a generalization of the socalled Parity lemma (see Isaacs [31]):

Let be a graph with vertices, maximum degree , and having a free edgecoloring, such that vertices have Boolecoloring , for , with . Then,
(6)
Indeed, since the Boolecoloring of each vertex is the sum of the Boolecolorings of its incident edges, and recalling again Equation (1), we can write
but this equality is only satisfied if or , for every . Then, from , we get the result.
Note that, as a direct consequence, we also get the following:

There is no edgecoloring of a graph having only one vertex with Boolecoloring 1 and the other vertices with Boolecoloring .
Another consequence is the following result by Tait [39]:

A cubic planar graph is Taitcolorable if and only if its corresponding map is colorable.
Using the Boolecolorings, the proof of this last result is as follows: First, recall that every map has a 3graph associated, because a vertex with degree greater than 3 can be replaced by a polygon, in such a way that the map obtained can be colored with 4 colors, and so can the original map; see an example in Figure 5. Now assume that we have the regions of the map with the colorings . Then, to obtain a Taitcoloring of a cubic planar graph, we only need to assign to each edge the sum of the colorings of both regions separated by this edge. To see that this gives a Taitcoloring, we only have to study one vertex, as shown in Figure 6. Since we have a 4colored map, each two neighboring regions have different colors. Thus, no sum can give 0. Moreover, since the three regions with a common vertex have different colorings and and is a group, the colorings , and must also be different. Figure 7 provides an example of a 4coloring of a map and its Taitcoloring (obtained from Table 1), where the colorings and are denoted by and , respectively.
Conversely, if we want to obtain a 4colored map from a Taitcoloring of the edges of the corresponding graph, we begin by giving the coloring 0 to any region considered as initial. Then, starting from this region, we follow an arbitrary path crossing some edges and visiting all the regions. We give each newly visited region the coloring obtained by adding the coloring of the ‘previous’ region plus the coloring of the last edge crossed. As no edge has the coloring 0, it is obvious that the coloring obtained for each region is different from that of its ‘previous’ region in the path followed; for an example of this process, see Figure 8 (left and center). Now, to finish the proof, we need to show that the coloring of each region is independent of the path followed. With this aim, let and be two paths with the same initial and final regions. We want to prove that the coloring obtained for the final region is the same following both paths; there is an example of this fact in Figure 8 (center and right). The colorings and obtained by following both paths are equal if and only if the sum of the colorings of all edges crossed, respectively, by and is 0. Indeed, let and be the colorings of the edges crossed respectively by and , then and . If , the sums in both parenthesis are equal, so . To prove this equality, we can assume that is a simple curve (see Figure 9) because, otherwise, we could decompose it into some simple curves. If we imagine that we cut the graph with this curve, we obtain two graphs, such that the colorings of the edges crossed by the curve must satisfy , where is the number of edges crossed with coloring . (Just imagine that in every cut we have two vertices of degree 1 and apply (6).) Then, , as claimed.
As previously mentioned, the concept of colorings allows us to use the theory of Boolean algebra for the construction and characterization of snarks, that is, cubic graphs that are not Taitcolorable, also known as class two. The name ‘snark’ was proposed by Gardner [25], who borrowed it from a nonsense poem by the famous English author Lewis Carroll [10]. The most simple example of snark is the Petersen graph [35] (see Figure 10). With the colorings we can obtain infinite families of snarks. An example is the family obtained by joining adequately an odd number of copies of the multipole (cubic graph with edges and semiedges—or ‘dangling edges’— which are edges with only one final vertex), shown in Figure 11 (left). This structure behaves as a NOT gate of logic circuits in the sense that, its edges and semiedges having been Taitcolored, the colorings and are conjugated one to each other, namely (respectively, ) if and only if (respectively, ). This is satisfied for any coloring of semiedge . Two examples of this fact are shown in Figure 11 (center and right). If, as previously stated, we join an odd number of these multipoles in a circular configuration, adding some vertices to connect semiedges , any attempt at Taitcoloring will lead to a conflict, and hence the graph is a snark. An example with five multipoles can be seen in Figure 12. This family of snarks, called flower snarks, was proposed by Loupekhine (see Isaacs [32]). The first infinite families of snarks were given by Isaacs [31], but they can also be obtained by using Boolecolorings. More details on this technique can be found in Fiol [16].
3 Known and unknown: Ramsey theory
Let us consider the following result:

At a cocktail party with six o more people, there are always three people who are known or unknown to each other.
In other words, if the complete graph on vertices can be (free) edgecolored with two colors, say blue and red, then it always contains a monochromatic triangle, namely, a subgraph with its three edges blue or red. Indeed, as each vertex has degree 5, at least 3 of its incident edges , , must have the same color, for example, blue. Then, if any of the 3 edges () is blue, we obtain a blue triangle. Otherwise, we have a red triangle. Although this is an easy proof, it can be extremely difficult to prove similar results having more colors and/or imposing other monochromatic subgraphs. In this context, recall that, given graphs , the Ramsey number is defined as the smallest number , such that, in any edgecoloring of using colors, there always exists a monochromatic subgraph (with color ) isomorphic to for some . If is a complete graph , the Ramsey number is expressed by writing instead of , for sake of simplicity. Some known results of exact values and bounds for Ramsey numbers are the following:
So, the result at the beginning of this section can be expressed as . Moreover, since (it is easy to color with two colors the edges of the complete graph without monochromatic triangles: the ‘outer cycle’ with one color and the ’inner’ cycle with the other) we conclude that . A good updated summary on this subject can be found in Radziszowski [37].
As an example, we now prove the following result:

.
We first see that . We make an edgecoloring of a complete graph using three colors; say blue, red and green. Let us assume that the edgecoloring has no monochromatic triangles. The green neighborhood of a vertex is the set of vertices that have a green edge to . The green neighborhood of cannot contain any green edge in order to avoid monochromatic triangles. Then, the edgecoloring of the green neighborhood of has only two colors: blue and red. Since , the green neighborhood of can contain at most 5 vertices. With the same reasoning, the blue and the red neighborhoods of can have at most 5 vertices each. As every vertex different from is in the green, blue or red neighborhoods of , then the complete graph can have at most vertices. Thus, .
Now, to prove that , we use algebraic graph theory based on the properties of eigenvalues and eigenvectors of the adjacency matrix, that is, a matrix with rows and columns indexed by the vertices of the graph, and whose entries are either 1 or 0, according to whether the corresponding vertices are adjacent or not.
A regular graph with vertices is said to be strongly regular if each pair of adjacent vertices has common neighbors and each pair of nonadjacent vertices has common neighbors.
If , then we can color the edges of the complete graph with three colors, namely, we can make an edgecoloring of without monochromatic triangles. The required edgecoloring is equivalent to a decomposition of into three graphs , and , each one corresponding to one color. It follows that each , , must be a graph on 16 vertices, regular of degree 5 (because each vertex has degree 15 and the neighborhood with one color has at most 5 vertices) and without triangles. Moreover, each vertex has 10 vertices at distance 2, which can be reached by paths of length 2. Then, we can consider a graph in which any two nonadjacent vertices have 2 common neighbors and any two adjacent vertices have no common neighbors. In other words, a strongly regular graph. It is known that there is just one such graph, the Clebsch graph, which is illustrated in two different ways in Figure 13. On the left, there is the Clesbch graph, as the graph whose vertices are labeled with the numbers 0 to 15 in base 2, and where two vertices are adjacent whenever the corresponding labels differ either by one or by all four digits. On the right, there is the Clebsch graph, as the rooted graph with vertices labeled , and the unordered pairs , with , for . In this representation, the adjacencies are , , , and if are all different and . In fact, the Clebsch graph is vertextransitive (informally speaking, we see the same structure from any vertex), so that any vertex can be chosen as vertex 0. Notice that, from this view of the Clebsch graph, it is apparent that the induced subgraph on ten vertices at distance 2 (from the vertex chosen as 0) is the Petersen graph [35]; compare Figure 13 (on the right) and Figure 10.
Therefore, our problem is to find three edgedisjoint copies of the Clebsch graph in . To this end, let us introduce the following terminology: Let be a family of graphs on the same vertex set and such that , for . We define the graph as the graph , where . Notice that the corresponding adjacency matrices satisfy . With denoting a graph isomorphic to the Clebsch graph, our problem now reads: Is it true that ? In terms of their adjacency matrices , we have
(7) 
since the adjacency matrix of is equal to , where denotes the matrix whose entries are all 1 and is the identity matrix.
We now use eigenvalue techniques to address Equation (7). Recall that the spectrum of an adjacency matrix gives the eigenvalues of this matrix (which are real because the matrix is symmetric), and that each eigenvalue has at least one eigenvector associated. To find the spectra of the Clebsch graph and the matrix , we can either compute them or simply find them in some standard reference, such as Godsil and Royle [27]. We then have that and , where the superscripts denote the multiplicity of each eigenvalue. In both cases, the largest eigenvalue has the all1 vector as eigenvector. It follows that the eigenvectors of the other eigenvalues are in the subspace (with vectors the addition of whose components are zero). Denote by the eigenspace of corresponding to the eigenvalue 1, namely, , and consider the subspace . As and , we infer that . From Equation (7), with , and , where , we obtain that and, then, and . This implies that
where , with .
This indicates that the required spectral condition necessary to the existence of the decomposition is satisfied. In this case, this condition is also sufficient, and it is known that there are only two nonisomorphic decompositions. One of these is illustrated in Figure 14, which shows how to color one third of the edges of with one color using the Clebsch graph. By rotating this graph and radians, we obtain the edges to be colored with the two other colors; with this, we get .
In the case of avoiding monochromatic triangles with colors, only bounds of Ramsey numbers are known. By definition, we state that for , that is, is the biggest integer such that can be colored with colors without monochromatic triangles. The following upper bound is known (see Fiol, Garriga and Yebra [23]):
(8) 
Recall that, surprisingly, we find the number . The proof is as follows: Obviously, and we know that and . If we compute from , considering that a vertex can only be adjacent to vertices, we obtain that . For any , we get the recurrence
We solve the corresponding linear equation
first solving its homogeneous equation
where is a constant. Then, we look for a particular solution of the complete equation:
where is a constant. Finally, gives and, hence, , as claimed.
From the examples given at the beginning of this section, we saw that . Using (8), we obtain that
which represents a good upper bound, quite close to the best bound known.
4 Common friends: Distanceregularity and coding theory
As commented by Aigner and Ziegler [1], nobody knows who was the first to state the following result and to give it the human touch:

At a cocktail party with three or more people, if each two people have exactly one friend in common, then there is a person the ‘politician’ who is a friend of everybody.
Nowadays, this result is known as the Friendship theorem. As mentioned in the introduction, the first proof (by contradiction) was given by Erdős, Rényi and Sós [12] in 1966, and is considered to be the most successful. Basically, it has two parts: First, it is proved that if the graph which models such a cocktail party (where people correspond to vertices and friendships are represented by edges) is a counterexample with more than three vertices, then it has to be regular, say with degree . As a consequence, has to be strongly regular with parameters , that is, every two adjacent vertices has exactly one common neighbor, and the same holds for every two nonadjacent vertices. Second, spectral graph theory is used to prove that cannot exist. In fact, the hypothetic graph would be an example of a distanceregular graph, in this case with diameter 2 (the concepts of stronglyregularity and distanceregularity coincide for connected graphs with diameter 2). Generally speaking, we say that a graph is distanceregular if, when it is observed or ‘hung’ from any of its vertices (called root), we obtain a partition of the vertex set into layers, where the layer contains the vertices at distance from the root, and the vertices in a layer are indistinguishable from each other with respect to their adjacencies. A more precise definition of distanceregularity is the following: A graph with diameter is distanceregular if, for every pair of vertices and integers , the number of vertices at distance from and at distance from only depends on the distance between and , . Then, we write , where the constants are called the intersection numbers. Indeed, because of the many relations between these numbers, it is possible to give a much more simple definition, since for each distance we only need the pairs of distances , and . The corresponding intersection numbers are enough to determine all the others; see, for example, Biggs [4]. Therefore, the most common definition of distanceregularity is: A graph is distanceregular if, for every pair of vertices at distance , the numbers , , and of vertices adjacent to , and at distance , , and , respectively, from only depends on , such that , , and . As simple examples of distanceregular graphs, we have the skeleton of regular polyhedrons; see again Figure 1. In Figure 15, we show the layer partition of the cube graph with the socalled intersection diagram of the corresponding intersection numbers. Notice that each layer is represented by a circle containing its number of vertices.
Since their introduction by Biggs in the early 70’s, distanceregular graphs, and their principal generalization called association schemes (see, for example, Brouwer and Haemers [9]), have been key concepts in algebraic combinatorics. These graphs have connections with other areas of mathematics, such as geometry, coding theory, group theory, design theory, and with other parts of graph theory. As pointed out by Brouwer, Cohen and Neumaier in their monumental book on this subject [8], this is because most of the finite objects with ‘enough’ regularity are closely related to distanceregular graphs.
In 1997 Fiol and Garriga [21, 19] gave the following quasispectral characterization of distanceregular graphs:

A regular graph with adjacency matrix and distinct eigenvalues is distanceregular if and only if the number of vertices at distance from each vertex is a constant and only depends on the spectrum of the matrix .
More precisely, consider a regular graph with vertices and spectrum , where are the eigenvalues of and the superscripts denote their multiplicities; is simple because is connected, thus is irreducible (PerronFrobenius theorem for nonnegative matrices, see Godsil [26, p. 31]). Then, is distanceregular if and only if, for each vertex ,
(9) 
where ’s are momentlike parameters, which can be calculated from the distance between eigenvalues with the formula , for . As examples, we give the spectrum, the number of vertices and the value of obtained from Equation (9) of the cube and the Petersen graph (see again Figures 15 and 10, respectively):

Cube: , , .

Petersen: , , .
As previously mentioned, the theory on distanceregular graphs has many applications in coding theory. Recall that a code , with a set of allowed words or codewords, can be simply represented as a vertex subset of a distanceregular graph ; see Godsil [26] and van Lint [41]. The vertex subset represents the ‘universe’ of words, with or without meaning, which can be received. There is an edge between two words if, with a certain probability, one can be transformed into the other in the process of transmission. Then, the shorter the distance between two words in , the more similar the words. If a codeword has not suffered too many changes, the resulting word is not far from the original one and it is possible to retrieve it (decision criterion by proximity). Therefore, a code is better if the words that constitute it are far away from each other. In the study and design of good codes, some algebraic techniques are used to obtain information about the structure of the graph and, in particular, about the vertex subset that represents the code. In the applications of special relevance, there are the socalled completely regular codes, whose graphs are structured in a kind of distanceregularity around the set that constitutes the code. Thus, these codes can be algebraically characterized in a similar way to the characterization of the distanceregular graphs through their spectra; see Fiol and Garriga [22] for more information.
5 Weddings: Hall’s and Menger’s theorems. Multibus networks
Let us imagine two groups of heterosexual people available for marriage, one of women and another of men, the latter at least as large as the former. Also imagine that every woman knows a certain number of men. The Hall Marriage theorem gives necessary and sufficient conditions for every woman to be able to marry a man who she knows:

A complete matching is possible if and only if each group of women, whatever their number, knows altogether at least an equal number of men.
If the sets of women and men are denoted by and , respectively, we can represent the above situation as a bipartite graph , with stable vertex sets and and where edges stand for acquaintances. Then, we can state Hall’s theorem in a more mathematical form:

In a bipartite graph with , a complete matching is possible if and only if, for every ,
(10) where .
(Recall that is the set of vertices adjacent to vertex .)
There are several proofs of Hall’s theorem. The proof we present here is by Rado, although our reasoning is a little different from that in Bollobás [6] or Harary [28]. As necessity is trivial, we are going to prove sufficiency. If graph satisfies Eq. (10), for any with and , it is immediate that contains a complete matching. If , then there exist at least two edges and , with . Now we claim that, after removing one of these edges, the resulting graph still satisfies Eq. (10). Indeed, if this were not the case, there would be two subsets , with and , such that and . Moreover, would be the only vertex of adjacent to (some vertex of) , and would be the only vertex of adjacent to . See this situation in Figure 16. Then, we would have that the common number of adjacent vertices to and would satisfy the inequality:
Moreover, we would also have:
a contradiction since, according to Eq. (10),
Consequently, every vertex with degree can be converted to a vertex with degree , and the resulting graph still satisfies Eq. (10). This completes the proof.
Curiously, Hall’s theorem is closely linked to another classical result in graph theory: Menger’s theorem; see, for example, Bollobás [7]. As in the case of Hall’s theorem, Menger’s theorem states that a certain condition, which is trivially necessary for a result to be true, is also sufficient. In Menger’s case, the result is not on matchings, but on the vertexconnectivity (or edgeconnectivity ) of a graph, which is defined as the minimum cardinality of a vertex (or edge) set whose deletion disconnects the graph or, in particular, two given vertices . This set is called a cutting set or separating set of or, in particular, of . Then, Menger’s theorem states that for every pair of vertices (nonadjacent, in the case of computing ):

The minimum size of a separating set of vertices equals the maximum number of independent paths in vertices from to .

The minimum size of a separating set of edges equals the maximum number of independent paths in edges from to .
It has been shown that the vertexconnectivity (or edgeconnectivity ) of a graph or digraph (a digraph is a graph whose edges are associated to one of the two possible directions) reaches its maximum value, which equals the minimum degree of , if in the diameter is small enough with respect to the girth (see Fàbrega and Fiol [15]) or if the number of vertices is large enough with respect to the diameter (see Fiol [17]).
Both the theorems mentioned, Hall’s and Menger’s, have many applications in the study and design of interconnection networks (for example, between processors) and in communication networks. Here we explain an application of Hall’s theorem to the study of multibus interconnection networks: A multiprocessor system with shared memory and multibus interconnection network consists of processors, buses and memory modules with . The processors have access to the memory modules through the buses, so we can establish processorbus and busmemory connections. Let us assume that there are requirements by the processors for accessing to different memory modules. As each processormemory connection requires a bus, if , then memories will be assigned; instead, if , then only memories will be assigned. In the complete scheme (see Figure 17), each bus is connected to all the memories and all the processors. This represents connections, and generally this provides an important saving with respect to the crossbar network with connections, one connection between each pair processormemory, because the number of buses is normally much smaller than the number of processors and memories. For example, if (an usual situation), the saving is obtained if .
Because the cost of the network basically depends on the number of connections, it is useful to consider the redundancy of this scheme. Namely, what is the maximum number of connections (processorbus or busmemory) that can be removed without having system degradation? In other words, how many connections, from all of , can be removed such that any of the processors asking for access to any of the different memory modules do not lose access? The answer is a direct consequence of the following result:

In a multiprocessor system with multibus network without having degradation, each bus can be disconnected from at most altogether processors or memory modules.
The proof is as follows: For each bus , , let and be, respectively, the number of processors and memories connected to it. Analogously, let and be the numbers of processors and memories disconnected from bus . Obviously, and . The result states that, in a nondegrading system, each bus can be disconnected from, at most, processors or memories, namely, for . But we can also state that each bus must have more than connections, such that for . Assume that, on the contrary, for each bus , we have . Let with and with be, respectively, the processors and memories disconnected to the bus . Note that . Now consider other processors and other memories , as in Figure 18. Let be the requirement of processor to access to memory . None of the requirements
can use bus , and this means that the system suffers degradation.
So, as stated before, the conclusion is that the maximum number of redundant connections is . In fact, this value is obtained with the socalled minimum topologies, such as the rhombic and the staircase topologies; see Tables 2 and 3, respectively. More details can be found in Fiol, Valero, Yebra and Land [24] and in Lang, Valero and Fiol [34].
Notice that the result only gives us necessary conditions for suffering degradation. In this context, Hall’s theorem is used to give a characterization for the interconnection topologies to prevent degradation of the system, as in the aforementioned cases of the complete and the minimum topologies:

A multibus system does not suffer degradation if and only if any of the disjoint pairs processormemory are connected to a set of, at least, buses.
As previously stated, this result gives necessary and sufficient conditions for a nondegrading multibus system.
Rhombic scheme  
0  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15 
7  7  7  7  7  7  7  7  7  
6  6  6  6  6  6  6  6  6  
5  5  5  5  5  5  5  5  5  
4  4  4  4  4  4  4  4  4  
3  3  3  3  3  3  3  3  3  
2  2  2  2  2  2  2  2  2  
1  1  1  1  1  1  1  1  1  
0  0  0  0  0  0  0  0  0 
Staircase scheme  
0  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15 
7  7  7  7  7  7  7  7  7  
6  6  6  6  6  6  6  6  6  
5  5  5  5  5  5  5  5  5  
4  4  4  4  4  4  4  4  4  
3  3  3  3  3  3  3  3  3  
2  2  2  2  2  2  2  2  2  
1  1  1  1  1  1  1  1  1  
0  0  0  0  0  0  0  0  0 
Acknowledgments
The authors thank professors J.L.A. Yebra and E. Garriga for their valuable comments. This research was supported by the Ministerio de Econom\́mbox{\boldmath$i$}a y Competitividad and the European Regional Development Fund under project MTM201128800C0201, and by the Catalan Government under project 2014SGR1147.
References
 M. Aigner, G.M. Ziegler, Proofs from THE BOOK, Springer, Berlin, 1998.
 K. Apple, An attempt to understand the four color problem, J. Graph Theory 1 (1977) 193–206.
 K. Apple, W. Haken, J. Koch, Every planar map is four colorable, Illinois J. Math. 21 (1977) 429–567.
 N.L. Biggs, Algebraic Graph Theory, Cambridge University Press, Cambridge, 1993.
 N.L. Biggs, E.K. Lloyd, R.J. Wilson, Graph Theory: 17361936, Claredon Press, Oxford, 1976.
 B. Bollobás, Extremal Graph Theory, Dover Publications, Mineola, N.Y., cop. 2004.
 B. Bollobás, Graph Theory: An Introductory Course, Springer, New York, 1979, 3rd corrected edition, 1990.
 A.E. Brouwer, A.M. Cohen, A. Neumaier, DistanceRegular Graphs, SpringerVerlag, Berlin, 1989.
 A.E. Brouwer, W.H. Haemers, Association schemes, in: R.L. Graham, et al. (eds.), Handbook of Combinatorics, Vol. 1–2, Elsevier, Amsterdam, 1995, 747–771.
 L. Carroll, The Hunting of the Snark, Annotated by M. Gardner, Penguin Books, New York, 1974.
 R. Diestel, Graph Theory. Springer, New York, 1997.
 P. Erdős, A. Rényi, V. Sós, On a problem of graph theory, Studia Sci. Math. 1 (1966) 215–235.
 L. Euler, Solutio problematis ad geometriam situs pertinentis, Commentarii academiae scientiarum petropolitanae 8 (1741) 128–140.
 L. Euler, Elementa doctrine solidorum, Novi comm. acad. scientiarum imperialis petropolitanae 4 (17521753) 109–160.
 J. Fàbrega, M.A. Fiol, Maximally connected digraphs, J. Graph Theory 13 (1989) 657–668.
 M.A. Fiol, A Boolean algebra approach to the construction of snarks, in Graph Theory, Combinatorics and Applications, Y. Alavi, G. Chartrand, O.R. Oellermann and A.J. Schwenk (eds.), Vol. 1, John Wiley & Sons, New York, 1991, 493–524.
 M.A. Fiol, The connectivity of large digraphs and graphs, J. Graph Theory 17 (1993) 31–45.
 M.A. Fiol, Critical graphs with maximum degree three, in: Graph Theory, Combinatorics, and Applications Y. Alavi and A.J. Schwenk, (eds.), Vol. 1, John Wiley & Sons, New York, 1995, 403–411.
 M.A. Fiol, Algebraic characterizations of distanceregular graphs, Discrete Math. 246 (2002), no.13, 111–129.
 M.A. Fiol, M.L. Fiol, Coloracions: un nou concepte dintre la teoria de coloració de grafs, L’Escaire 11 (1984) 33–44 (in Catalan).
 M.A. Fiol, E. Garriga, From local adjacency polynomials to locally pseudodistanceregular graphs, J. Combin. Theory Ser. B 71 (1997) 162–183.
 M.A. Fiol, E. Garriga, On the algebraic theory of pseudodistanceregularity around a set, Linear Algebra Appl. 298 (1999), no. 13, 115–141.
 M.A. Fiol, E. Garriga, J.L.A. Yebra, Avoiding monocoloured triangles when colouring , Research Report, Universitat Politècnica de Catalunya, Barcelona, 1995.
 M.A. Fiol, M. Valero, J.L.A. Yebra, T. Lang, Reduced interconnection networks based on the multiple bus for multimicroprocessor systems, International Journal of Mini and Microcomputers 6 (1984), no. 1, 4–9.
 M. Gardner, Mathematical games: Snarks, Boojums and other conjectures related to the fourcolormap theorem, Sci. Amer. 234 (1976) 126–130.
 C.D. Godsil, Algebraic Combinatorics, Chapman and Hall, New York, 1993.
 C.D. Godsil, G. Royle, Algebraic Graph Theory, SpringerVerlag, New York, 2001.
 F. Harary, Graph Theory, AddisonWesley, Reading, MA, 1969.
 P.J. Heawood, Map Colour Theorems, Quart. J. Math. 24 (1890) 332–338.
 P. Hoffman, The Man Who Loved Only Numbers: The Story of Paul Erdős and the Search for Mathematical Truth. Hyperion, New York, 1998.
 R. Isaacs, Infinite families of nontrivial graphs which are not Tait colorable, Amer. Math. Monthly 82 (1975), no. 3, 221–239.
 R. Isaacs, Loupekhine’s snarks: a bifamily of nonTaitcolorable graphs, Technical Report, 263, Dept. of Math Sci., The Johns Hopkins University, Maryland, 1976.
 C. Kuratowski, Sur le problème des courbes gauches en topologie, Fund. Math. 15 (1930) 217–283.
 T. Lang, M. Valero, M.A. Fiol, Reduction of connections for multibus organization, IEEE Trans. Comput. C32 (1983), no. 8, 707–716.
 J. Petersen, Sur le théorème de Tait, L’Intermédiaire des Mathématiciens 5 (1898) 225–227.
 H. Rademacher, O. Toeplitz, The Enjoyment of Math, Princenton University Press, Princenton, New Jersey, 1957.
 S.P. Radziszowski, Small Ramsey numbers, Electron. J. Combin. 1 (2006) Dynamical Survey 1.
 N. Robertson, D.P. Sanders, P.D. Seymour, R. Thomas, The four colour theorem, J. Combin. Theory Ser. B. 70 (1997) 2–44.
 P.G. Tait, Remarks on the colouring of maps, Proc. Roy. Soc. Edimburgh 10 (1880) 501–503, 729.
 C. Thomassen, A link between the Jordan curve theorem and the Kuratowski planarity criterion, Amer. Math. Monthly 97 (1990) 216–218.
 J.H. van Lint, Introduction to Coding Theory, third edition, Springer, Berlin, 1999.
 D.B. West, Introduction to Graph Theory, PrenticeHall, Englewood Cliffs, NJ, 2nd edition, 2000.